These counters will track hugetlb reservations rather than hugetlb memory faulted in. This patch only adds the counter, following patches add the charging and uncharging of the counter.
This is patch 1 of an 9 patch series.
Problem: Currently tasks attempting to reserve more hugetlb memory than is available get a failure at mmap/shmget time. This is thanks to Hugetlbfs Reservations [1]. However, if a task attempts to reserve more hugetlb memory than its hugetlb_cgroup limit allows, the kernel will allow the mmap/shmget call, but will SIGBUS the task when it attempts to fault in the excess memory.
We have users hitting their hugetlb_cgroup limits and thus we've been looking at this failure mode. We'd like to improve this behavior such that users violating the hugetlb_cgroup limits get an error on mmap/shmget time, rather than getting SIGBUS'd when they try to fault the excess memory in. This gives the user an opportunity to fallback more gracefully to non-hugetlbfs memory for example.
The underlying problem is that today's hugetlb_cgroup accounting happens at hugetlb memory *fault* time, rather than at *reservation* time. Thus, enforcing the hugetlb_cgroup limit only happens at fault time, and the offending task gets SIGBUS'd.
Proposed Solution: A new page counter named 'hugetlb.xMB.rsvd.[limit|usage|max_usage]_in_bytes'. This counter has slightly different semantics than 'hugetlb.xMB.[limit|usage|max_usage]_in_bytes':
- While usage_in_bytes tracks all *faulted* hugetlb memory, rsvd.usage_in_bytes tracks all *reserved* hugetlb memory and hugetlb memory faulted in without a prior reservation.
- If a task attempts to reserve more memory than limit_in_bytes allows, the kernel will allow it to do so. But if a task attempts to reserve more memory than rsvd.limit_in_bytes, the kernel will fail this reservation.
This proposal is implemented in this patch series, with tests to verify functionality and show the usage.
Alternatives considered: 1. A new cgroup, instead of only a new page_counter attached to the existing hugetlb_cgroup. Adding a new cgroup seemed like a lot of code duplication with hugetlb_cgroup. Keeping hugetlb related page counters under hugetlb_cgroup seemed cleaner as well.
2. Instead of adding a new counter, we considered adding a sysctl that modifies the behavior of hugetlb.xMB.[limit|usage]_in_bytes, to do accounting at reservation time rather than fault time. Adding a new page_counter seems better as userspace could, if it wants, choose to enforce different cgroups differently: one via limit_in_bytes, and another via rsvd.limit_in_bytes. This could be very useful if you're transitioning how hugetlb memory is partitioned on your system one cgroup at a time, for example. Also, someone may find usage for both limit_in_bytes and rsvd.limit_in_bytes concurrently, and this approach gives them the option to do so.
Testing: - Added tests passing. - Used libhugetlbfs for regression testing.
[1]: https://www.kernel.org/doc/html/latest/vm/hugetlbfs_reserv.html
Signed-off-by: Mina Almasry almasrymina@google.com Acked-by: David Rientjes rientjes@google.com
--- Changes in v11: - Renamed resv.* or 'reservation' or 'reserved' to rsvd.* - Renamed hugetlb_cgroup_get_counter() to hugetlb_cgroup_counter_from_cgroup().
Changes in v10: - Renamed reservation_* to resv.*
--- include/linux/hugetlb.h | 4 +- mm/hugetlb_cgroup.c | 116 +++++++++++++++++++++++++++++++++++----- 2 files changed, 105 insertions(+), 15 deletions(-)
diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h index 1e897e4168ac1..dea6143aa0685 100644 --- a/include/linux/hugetlb.h +++ b/include/linux/hugetlb.h @@ -432,8 +432,8 @@ struct hstate { unsigned int surplus_huge_pages_node[MAX_NUMNODES]; #ifdef CONFIG_CGROUP_HUGETLB /* cgroup control files */ - struct cftype cgroup_files_dfl[5]; - struct cftype cgroup_files_legacy[5]; + struct cftype cgroup_files_dfl[7]; + struct cftype cgroup_files_legacy[9]; #endif char name[HSTATE_NAME_LEN]; }; diff --git a/mm/hugetlb_cgroup.c b/mm/hugetlb_cgroup.c index e434b05416c68..cd1f44775259c 100644 --- a/mm/hugetlb_cgroup.c +++ b/mm/hugetlb_cgroup.c @@ -36,6 +36,11 @@ struct hugetlb_cgroup { */ struct page_counter hugepage[HUGE_MAX_HSTATE];
+ /* + * the counter to account for hugepage reservations from hugetlb. + */ + struct page_counter rsvd_hugepage[HUGE_MAX_HSTATE]; + atomic_long_t events[HUGE_MAX_HSTATE][HUGETLB_NR_MEMORY_EVENTS]; atomic_long_t events_local[HUGE_MAX_HSTATE][HUGETLB_NR_MEMORY_EVENTS];
@@ -55,6 +60,15 @@ struct hugetlb_cgroup {
static struct hugetlb_cgroup *root_h_cgroup __read_mostly;
+static inline struct page_counter * +hugetlb_cgroup_counter_from_cgroup(struct hugetlb_cgroup *h_cg, int idx, + bool rsvd) +{ + if (rsvd) + return &h_cg->rsvd_hugepage[idx]; + return &h_cg->hugepage[idx]; +} + static inline struct hugetlb_cgroup *hugetlb_cgroup_from_css(struct cgroup_subsys_state *s) { @@ -295,28 +309,42 @@ void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages,
enum { RES_USAGE, + RES_RSVD_USAGE, RES_LIMIT, + RES_RSVD_LIMIT, RES_MAX_USAGE, + RES_RSVD_MAX_USAGE, RES_FAILCNT, + RES_RSVD_FAILCNT, };
static u64 hugetlb_cgroup_read_u64(struct cgroup_subsys_state *css, struct cftype *cft) { struct page_counter *counter; + struct page_counter *rsvd_counter; struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css);
counter = &h_cg->hugepage[MEMFILE_IDX(cft->private)]; + rsvd_counter = &h_cg->rsvd_hugepage[MEMFILE_IDX(cft->private)];
switch (MEMFILE_ATTR(cft->private)) { case RES_USAGE: return (u64)page_counter_read(counter) * PAGE_SIZE; + case RES_RSVD_USAGE: + return (u64)page_counter_read(rsvd_counter) * PAGE_SIZE; case RES_LIMIT: return (u64)counter->max * PAGE_SIZE; + case RES_RSVD_LIMIT: + return (u64)rsvd_counter->max * PAGE_SIZE; case RES_MAX_USAGE: return (u64)counter->watermark * PAGE_SIZE; + case RES_RSVD_MAX_USAGE: + return (u64)rsvd_counter->watermark * PAGE_SIZE; case RES_FAILCNT: return counter->failcnt; + case RES_RSVD_FAILCNT: + return rsvd_counter->failcnt; default: BUG(); } @@ -338,10 +366,16 @@ static int hugetlb_cgroup_read_u64_max(struct seq_file *seq, void *v) 1 << huge_page_order(&hstates[idx]));
switch (MEMFILE_ATTR(cft->private)) { + case RES_RSVD_USAGE: + counter = &h_cg->rsvd_hugepage[idx]; + /* Fall through. */ case RES_USAGE: val = (u64)page_counter_read(counter); seq_printf(seq, "%llu\n", val * PAGE_SIZE); break; + case RES_RSVD_LIMIT: + counter = &h_cg->rsvd_hugepage[idx]; + /* Fall through. */ case RES_LIMIT: val = (u64)counter->max; if (val == limit) @@ -365,6 +399,7 @@ static ssize_t hugetlb_cgroup_write(struct kernfs_open_file *of, int ret, idx; unsigned long nr_pages; struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(of_css(of)); + bool rsvd = false;
if (hugetlb_cgroup_is_root(h_cg)) /* Can't set limit on root */ return -EINVAL; @@ -378,9 +413,14 @@ static ssize_t hugetlb_cgroup_write(struct kernfs_open_file *of, nr_pages = round_down(nr_pages, 1 << huge_page_order(&hstates[idx]));
switch (MEMFILE_ATTR(of_cft(of)->private)) { + case RES_RSVD_LIMIT: + rsvd = true; + /* Fall through. */ case RES_LIMIT: mutex_lock(&hugetlb_limit_mutex); - ret = page_counter_set_max(&h_cg->hugepage[idx], nr_pages); + ret = page_counter_set_max( + hugetlb_cgroup_counter_from_cgroup(h_cg, idx, rsvd), + nr_pages); mutex_unlock(&hugetlb_limit_mutex); break; default: @@ -406,18 +446,26 @@ static ssize_t hugetlb_cgroup_reset(struct kernfs_open_file *of, char *buf, size_t nbytes, loff_t off) { int ret = 0; - struct page_counter *counter; + struct page_counter *counter, *rsvd_counter; struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(of_css(of));
counter = &h_cg->hugepage[MEMFILE_IDX(of_cft(of)->private)]; + rsvd_counter = + &h_cg->rsvd_hugepage[MEMFILE_IDX(of_cft(of)->private)];
switch (MEMFILE_ATTR(of_cft(of)->private)) { case RES_MAX_USAGE: page_counter_reset_watermark(counter); break; + case RES_RSVD_MAX_USAGE: + page_counter_reset_watermark(rsvd_counter); + break; case RES_FAILCNT: counter->failcnt = 0; break; + case RES_RSVD_FAILCNT: + rsvd_counter->failcnt = 0; + break; default: ret = -EINVAL; break; @@ -472,7 +520,7 @@ static void __init __hugetlb_cgroup_file_dfl_init(int idx) struct hstate *h = &hstates[idx];
/* format the size */ - mem_fmt(buf, 32, huge_page_size(h)); + mem_fmt(buf, sizeof(buf), huge_page_size(h));
/* Add the limit file */ cft = &h->cgroup_files_dfl[0]; @@ -482,15 +530,30 @@ static void __init __hugetlb_cgroup_file_dfl_init(int idx) cft->write = hugetlb_cgroup_write_dfl; cft->flags = CFTYPE_NOT_ON_ROOT;
- /* Add the current usage file */ + /* Add the reservation limit file */ cft = &h->cgroup_files_dfl[1]; + snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.max", buf); + cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_LIMIT); + cft->seq_show = hugetlb_cgroup_read_u64_max; + cft->write = hugetlb_cgroup_write_dfl; + cft->flags = CFTYPE_NOT_ON_ROOT; + + /* Add the current usage file */ + cft = &h->cgroup_files_dfl[2]; snprintf(cft->name, MAX_CFTYPE_NAME, "%s.current", buf); cft->private = MEMFILE_PRIVATE(idx, RES_USAGE); cft->seq_show = hugetlb_cgroup_read_u64_max; cft->flags = CFTYPE_NOT_ON_ROOT;
+ /* Add the current reservation usage file */ + cft = &h->cgroup_files_dfl[3]; + snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.current", buf); + cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_USAGE); + cft->seq_show = hugetlb_cgroup_read_u64_max; + cft->flags = CFTYPE_NOT_ON_ROOT; + /* Add the events file */ - cft = &h->cgroup_files_dfl[2]; + cft = &h->cgroup_files_dfl[4]; snprintf(cft->name, MAX_CFTYPE_NAME, "%s.events", buf); cft->private = MEMFILE_PRIVATE(idx, 0); cft->seq_show = hugetlb_events_show; @@ -498,7 +561,7 @@ static void __init __hugetlb_cgroup_file_dfl_init(int idx) cft->flags = CFTYPE_NOT_ON_ROOT;
/* Add the events.local file */ - cft = &h->cgroup_files_dfl[3]; + cft = &h->cgroup_files_dfl[5]; snprintf(cft->name, MAX_CFTYPE_NAME, "%s.events.local", buf); cft->private = MEMFILE_PRIVATE(idx, 0); cft->seq_show = hugetlb_events_local_show; @@ -507,7 +570,7 @@ static void __init __hugetlb_cgroup_file_dfl_init(int idx) cft->flags = CFTYPE_NOT_ON_ROOT;
/* NULL terminate the last cft */ - cft = &h->cgroup_files_dfl[4]; + cft = &h->cgroup_files_dfl[6]; memset(cft, 0, sizeof(*cft));
WARN_ON(cgroup_add_dfl_cftypes(&hugetlb_cgrp_subsys, @@ -521,7 +584,7 @@ static void __init __hugetlb_cgroup_file_legacy_init(int idx) struct hstate *h = &hstates[idx];
/* format the size */ - mem_fmt(buf, 32, huge_page_size(h)); + mem_fmt(buf, sizeof(buf), huge_page_size(h));
/* Add the limit file */ cft = &h->cgroup_files_legacy[0]; @@ -530,28 +593,55 @@ static void __init __hugetlb_cgroup_file_legacy_init(int idx) cft->read_u64 = hugetlb_cgroup_read_u64; cft->write = hugetlb_cgroup_write_legacy;
- /* Add the usage file */ + /* Add the reservation limit file */ cft = &h->cgroup_files_legacy[1]; + snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.limit_in_bytes", buf); + cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_LIMIT); + cft->read_u64 = hugetlb_cgroup_read_u64; + cft->write = hugetlb_cgroup_write_legacy; + + /* Add the usage file */ + cft = &h->cgroup_files_legacy[2]; snprintf(cft->name, MAX_CFTYPE_NAME, "%s.usage_in_bytes", buf); cft->private = MEMFILE_PRIVATE(idx, RES_USAGE); cft->read_u64 = hugetlb_cgroup_read_u64;
+ /* Add the reservation usage file */ + cft = &h->cgroup_files_legacy[3]; + snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.usage_in_bytes", buf); + cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_USAGE); + cft->read_u64 = hugetlb_cgroup_read_u64; + /* Add the MAX usage file */ - cft = &h->cgroup_files_legacy[2]; + cft = &h->cgroup_files_legacy[4]; snprintf(cft->name, MAX_CFTYPE_NAME, "%s.max_usage_in_bytes", buf); cft->private = MEMFILE_PRIVATE(idx, RES_MAX_USAGE); cft->write = hugetlb_cgroup_reset; cft->read_u64 = hugetlb_cgroup_read_u64;
+ /* Add the MAX reservation usage file */ + cft = &h->cgroup_files_legacy[5]; + snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.max_usage_in_bytes", buf); + cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_MAX_USAGE); + cft->write = hugetlb_cgroup_reset; + cft->read_u64 = hugetlb_cgroup_read_u64; + /* Add the failcntfile */ - cft = &h->cgroup_files_legacy[3]; + cft = &h->cgroup_files_legacy[6]; snprintf(cft->name, MAX_CFTYPE_NAME, "%s.failcnt", buf); - cft->private = MEMFILE_PRIVATE(idx, RES_FAILCNT); + cft->private = MEMFILE_PRIVATE(idx, RES_FAILCNT); + cft->write = hugetlb_cgroup_reset; + cft->read_u64 = hugetlb_cgroup_read_u64; + + /* Add the reservation failcntfile */ + cft = &h->cgroup_files_legacy[7]; + snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.failcnt", buf); + cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_FAILCNT); cft->write = hugetlb_cgroup_reset; cft->read_u64 = hugetlb_cgroup_read_u64;
/* NULL terminate the last cft */ - cft = &h->cgroup_files_legacy[4]; + cft = &h->cgroup_files_legacy[8]; memset(cft, 0, sizeof(*cft));
WARN_ON(cgroup_add_legacy_cftypes(&hugetlb_cgrp_subsys, -- 2.25.0.341.g760bfbb309-goog
Augments hugetlb_cgroup_charge_cgroup to be able to charge hugetlb usage or hugetlb reservation counter.
Adds a new interface to uncharge a hugetlb_cgroup counter via hugetlb_cgroup_uncharge_counter.
Integrates the counter with hugetlb_cgroup, via hugetlb_cgroup_init, hugetlb_cgroup_have_usage, and hugetlb_cgroup_css_offline.
Signed-off-by: Mina Almasry almasrymina@google.com Acked-by: Mike Kravetz mike.kravetz@oracle.com Acked-by: David Rientjes rientjes@google.com
---
Changes in v11: - Changed all 'reserved' or 'reservation' to 'rsvd' to reflect the user interface. - Expanded comment that describes tail pages usage. facing naming. Changes in v10: - Added missing VM_BUG_ON Changes in V9: - Fixed HUGETLB_CGROUP_MIN_ORDER. - Minor variable name update. - Moved some init/cleanup code from later patches in the series to this patch. - Updated reparenting of reservation accounting.
--- include/linux/hugetlb_cgroup.h | 66 +++++++++++++-------- mm/hugetlb.c | 19 +++--- mm/hugetlb_cgroup.c | 103 ++++++++++++++++++++++++--------- 3 files changed, 128 insertions(+), 60 deletions(-)
diff --git a/include/linux/hugetlb_cgroup.h b/include/linux/hugetlb_cgroup.h index 063962f6dfc6a..91bc5e920e9d0 100644 --- a/include/linux/hugetlb_cgroup.h +++ b/include/linux/hugetlb_cgroup.h @@ -20,29 +20,38 @@ struct hugetlb_cgroup; /* * Minimum page order trackable by hugetlb cgroup. - * At least 3 pages are necessary for all the tracking information. + * At least 4 pages are necessary for all the tracking information. + * The second tail page (hpage[2]) is the fault usage cgroup. + * The third tail page (hpage[3]) is the reservation usage cgroup. */ #define HUGETLB_CGROUP_MIN_ORDER 2
#ifdef CONFIG_CGROUP_HUGETLB
-static inline struct hugetlb_cgroup *hugetlb_cgroup_from_page(struct page *page) +static inline struct hugetlb_cgroup *hugetlb_cgroup_from_page(struct page *page, + bool rsvd) { VM_BUG_ON_PAGE(!PageHuge(page), page);
if (compound_order(page) < HUGETLB_CGROUP_MIN_ORDER) return NULL; - return (struct hugetlb_cgroup *)page[2].private; + if (rsvd) + return (struct hugetlb_cgroup *)page[3].private; + else + return (struct hugetlb_cgroup *)page[2].private; }
-static inline -int set_hugetlb_cgroup(struct page *page, struct hugetlb_cgroup *h_cg) +static inline int set_hugetlb_cgroup(struct page *page, + struct hugetlb_cgroup *h_cg, bool rsvd) { VM_BUG_ON_PAGE(!PageHuge(page), page);
if (compound_order(page) < HUGETLB_CGROUP_MIN_ORDER) return -1; - page[2].private = (unsigned long)h_cg; + if (rsvd) + page[3].private = (unsigned long)h_cg; + else + page[2].private = (unsigned long)h_cg; return 0; }
@@ -52,26 +61,33 @@ static inline bool hugetlb_cgroup_disabled(void) }
extern int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages, - struct hugetlb_cgroup **ptr); + struct hugetlb_cgroup **ptr, bool rsvd); extern void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages, struct hugetlb_cgroup *h_cg, - struct page *page); + struct page *page, bool rsvd); extern void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages, - struct page *page); + struct page *page, bool rsvd); + extern void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages, - struct hugetlb_cgroup *h_cg); + struct hugetlb_cgroup *h_cg, + bool rsvd); +extern void hugetlb_cgroup_uncharge_counter(struct page_counter *p, + unsigned long nr_pages, + struct cgroup_subsys_state *css); + extern void hugetlb_cgroup_file_init(void) __init; extern void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage);
#else -static inline struct hugetlb_cgroup *hugetlb_cgroup_from_page(struct page *page) +static inline struct hugetlb_cgroup *hugetlb_cgroup_from_page(struct page *page, + bool rsvd) { return NULL; }
-static inline -int set_hugetlb_cgroup(struct page *page, struct hugetlb_cgroup *h_cg) +static inline int set_hugetlb_cgroup(struct page *page, + struct hugetlb_cgroup *h_cg, bool rsvd) { return 0; } @@ -81,28 +97,28 @@ static inline bool hugetlb_cgroup_disabled(void) return true; }
-static inline int -hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages, - struct hugetlb_cgroup **ptr) +static inline int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages, + struct hugetlb_cgroup **ptr, + bool rsvd) { return 0; }
-static inline void -hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages, - struct hugetlb_cgroup *h_cg, - struct page *page) +static inline void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages, + struct hugetlb_cgroup *h_cg, + struct page *page, bool rsvd) { }
-static inline void -hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages, struct page *page) +static inline void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages, + struct page *page, bool rsvd) { }
-static inline void -hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages, - struct hugetlb_cgroup *h_cg) +static inline void hugetlb_cgroup_uncharge_cgroup(int idx, + unsigned long nr_pages, + struct hugetlb_cgroup *h_cg, + bool rsvd) { }
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index dd8737a94bec4..62a4cf3db4090 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -1068,7 +1068,8 @@ static void update_and_free_page(struct hstate *h, struct page *page) 1 << PG_active | 1 << PG_private | 1 << PG_writeback); } - VM_BUG_ON_PAGE(hugetlb_cgroup_from_page(page), page); + VM_BUG_ON_PAGE(hugetlb_cgroup_from_page(page, false), page); + VM_BUG_ON_PAGE(hugetlb_cgroup_from_page(page, true), page); set_compound_page_dtor(page, NULL_COMPOUND_DTOR); set_page_refcounted(page); if (hstate_is_gigantic(h)) { @@ -1178,8 +1179,8 @@ static void __free_huge_page(struct page *page)
spin_lock(&hugetlb_lock); clear_page_huge_active(page); - hugetlb_cgroup_uncharge_page(hstate_index(h), - pages_per_huge_page(h), page); + hugetlb_cgroup_uncharge_page(hstate_index(h), pages_per_huge_page(h), + page, false); if (restore_reserve) h->resv_huge_pages++;
@@ -1253,7 +1254,8 @@ static void prep_new_huge_page(struct hstate *h, struct page *page, int nid) INIT_LIST_HEAD(&page->lru); set_compound_page_dtor(page, HUGETLB_PAGE_DTOR); spin_lock(&hugetlb_lock); - set_hugetlb_cgroup(page, NULL); + set_hugetlb_cgroup(page, NULL, false); + set_hugetlb_cgroup(page, NULL, true); h->nr_huge_pages++; h->nr_huge_pages_node[nid]++; spin_unlock(&hugetlb_lock); @@ -2039,7 +2041,8 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, gbl_chg = 1; }
- ret = hugetlb_cgroup_charge_cgroup(idx, pages_per_huge_page(h), &h_cg); + ret = hugetlb_cgroup_charge_cgroup(idx, pages_per_huge_page(h), &h_cg, + false); if (ret) goto out_subpool_put;
@@ -2063,7 +2066,8 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, list_move(&page->lru, &h->hugepage_activelist); /* Fall through */ } - hugetlb_cgroup_commit_charge(idx, pages_per_huge_page(h), h_cg, page); + hugetlb_cgroup_commit_charge(idx, pages_per_huge_page(h), h_cg, page, + false); spin_unlock(&hugetlb_lock);
set_page_private(page, (unsigned long)spool); @@ -2087,7 +2091,8 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, return page;
out_uncharge_cgroup: - hugetlb_cgroup_uncharge_cgroup(idx, pages_per_huge_page(h), h_cg); + hugetlb_cgroup_uncharge_cgroup(idx, pages_per_huge_page(h), h_cg, + false); out_subpool_put: if (map_chg || avoid_reserve) hugepage_subpool_put_pages(spool, 1); diff --git a/mm/hugetlb_cgroup.c b/mm/hugetlb_cgroup.c index cd1f44775259c..640555d7f78c6 100644 --- a/mm/hugetlb_cgroup.c +++ b/mm/hugetlb_cgroup.c @@ -97,8 +97,12 @@ static inline bool hugetlb_cgroup_have_usage(struct hugetlb_cgroup *h_cg) int idx;
for (idx = 0; idx < hugetlb_max_hstate; idx++) { - if (page_counter_read(&h_cg->hugepage[idx])) + if (page_counter_read(hugetlb_cgroup_counter_from_cgroup( + h_cg, idx, true)) || + page_counter_read(hugetlb_cgroup_counter_from_cgroup( + h_cg, idx, false))) { return true; + } } return false; } @@ -109,18 +113,34 @@ static void hugetlb_cgroup_init(struct hugetlb_cgroup *h_cgroup, int idx;
for (idx = 0; idx < HUGE_MAX_HSTATE; idx++) { - struct page_counter *counter = &h_cgroup->hugepage[idx]; - struct page_counter *parent = NULL; + struct page_counter *fault_parent = NULL; + struct page_counter *rsvd_parent = NULL; unsigned long limit; int ret;
- if (parent_h_cgroup) - parent = &parent_h_cgroup->hugepage[idx]; - page_counter_init(counter, parent); + if (parent_h_cgroup) { + fault_parent = hugetlb_cgroup_counter_from_cgroup( + parent_h_cgroup, idx, false); + rsvd_parent = hugetlb_cgroup_counter_from_cgroup( + parent_h_cgroup, idx, true); + } + page_counter_init(hugetlb_cgroup_counter_from_cgroup( + h_cgroup, idx, false), + fault_parent); + page_counter_init(hugetlb_cgroup_counter_from_cgroup(h_cgroup, + idx, true), + rsvd_parent);
limit = round_down(PAGE_COUNTER_MAX, 1 << huge_page_order(&hstates[idx])); - ret = page_counter_set_max(counter, limit); + + ret = page_counter_set_max(hugetlb_cgroup_counter_from_cgroup( + h_cgroup, idx, false), + limit); + VM_BUG_ON(ret); + ret = page_counter_set_max( + hugetlb_cgroup_counter_from_cgroup(h_cgroup, idx, true), + limit); VM_BUG_ON(ret); } } @@ -150,7 +170,6 @@ static void hugetlb_cgroup_css_free(struct cgroup_subsys_state *css) kfree(h_cgroup); }
- /* * Should be called with hugetlb_lock held. * Since we are holding hugetlb_lock, pages cannot get moved from @@ -166,7 +185,7 @@ static void hugetlb_cgroup_move_parent(int idx, struct hugetlb_cgroup *h_cg, struct hugetlb_cgroup *page_hcg; struct hugetlb_cgroup *parent = parent_hugetlb_cgroup(h_cg);
- page_hcg = hugetlb_cgroup_from_page(page); + page_hcg = hugetlb_cgroup_from_page(page, false); /* * We can have pages in active list without any cgroup * ie, hugepage with less than 3 pages. We can safely @@ -185,7 +204,7 @@ static void hugetlb_cgroup_move_parent(int idx, struct hugetlb_cgroup *h_cg, /* Take the pages off the local counter */ page_counter_cancel(counter, nr_pages);
- set_hugetlb_cgroup(page, parent); + set_hugetlb_cgroup(page, parent, false); out: return; } @@ -228,7 +247,7 @@ static inline void hugetlb_event(struct hugetlb_cgroup *hugetlb, int idx, }
int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages, - struct hugetlb_cgroup **ptr) + struct hugetlb_cgroup **ptr, bool rsvd) { int ret = 0; struct page_counter *counter; @@ -251,13 +270,20 @@ int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages, } rcu_read_unlock();
- if (!page_counter_try_charge(&h_cg->hugepage[idx], nr_pages, - &counter)) { + if (!page_counter_try_charge( + hugetlb_cgroup_counter_from_cgroup(h_cg, idx, rsvd), + nr_pages, &counter)) { ret = -ENOMEM; hugetlb_event(hugetlb_cgroup_from_counter(counter, idx), idx, HUGETLB_MAX); + css_put(&h_cg->css); + goto done; } - css_put(&h_cg->css); + /* Reservations take a reference to the css because they do not get + * reparented. + */ + if (!rsvd) + css_put(&h_cg->css); done: *ptr = h_cg; return ret; @@ -266,12 +292,12 @@ int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages, /* Should be called with hugetlb_lock held */ void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages, struct hugetlb_cgroup *h_cg, - struct page *page) + struct page *page, bool rsvd) { if (hugetlb_cgroup_disabled() || !h_cg) return;
- set_hugetlb_cgroup(page, h_cg); + set_hugetlb_cgroup(page, h_cg, rsvd); return; }
@@ -279,23 +305,29 @@ void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages, * Should be called with hugetlb_lock held */ void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages, - struct page *page) + struct page *page, bool rsvd) { struct hugetlb_cgroup *h_cg;
if (hugetlb_cgroup_disabled()) return; lockdep_assert_held(&hugetlb_lock); - h_cg = hugetlb_cgroup_from_page(page); + h_cg = hugetlb_cgroup_from_page(page, rsvd); if (unlikely(!h_cg)) return; - set_hugetlb_cgroup(page, NULL); - page_counter_uncharge(&h_cg->hugepage[idx], nr_pages); + set_hugetlb_cgroup(page, NULL, rsvd); + + page_counter_uncharge( + hugetlb_cgroup_counter_from_cgroup(h_cg, idx, rsvd), nr_pages); + + if (rsvd) + css_put(&h_cg->css); + return; }
void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages, - struct hugetlb_cgroup *h_cg) + struct hugetlb_cgroup *h_cg, bool rsvd) { if (hugetlb_cgroup_disabled() || !h_cg) return; @@ -303,8 +335,22 @@ void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages, if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER) return;
- page_counter_uncharge(&h_cg->hugepage[idx], nr_pages); - return; + page_counter_uncharge( + hugetlb_cgroup_counter_from_cgroup(h_cg, idx, rsvd), nr_pages); + + if (rsvd) + css_put(&h_cg->css); +} + +void hugetlb_cgroup_uncharge_counter(struct page_counter *p, + unsigned long nr_pages, + struct cgroup_subsys_state *css) +{ + if (hugetlb_cgroup_disabled() || !p || !css) + return; + + page_counter_uncharge(p, nr_pages); + css_put(css); }
enum { @@ -450,8 +496,7 @@ static ssize_t hugetlb_cgroup_reset(struct kernfs_open_file *of, struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(of_css(of));
counter = &h_cg->hugepage[MEMFILE_IDX(of_cft(of)->private)]; - rsvd_counter = - &h_cg->rsvd_hugepage[MEMFILE_IDX(of_cft(of)->private)]; + rsvd_counter = &h_cg->rsvd_hugepage[MEMFILE_IDX(of_cft(of)->private)];
switch (MEMFILE_ATTR(of_cft(of)->private)) { case RES_MAX_USAGE: @@ -676,6 +721,7 @@ void __init hugetlb_cgroup_file_init(void) void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage) { struct hugetlb_cgroup *h_cg; + struct hugetlb_cgroup *h_cg_rsvd; struct hstate *h = page_hstate(oldhpage);
if (hugetlb_cgroup_disabled()) @@ -683,11 +729,12 @@ void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage)
VM_BUG_ON_PAGE(!PageHuge(oldhpage), oldhpage); spin_lock(&hugetlb_lock); - h_cg = hugetlb_cgroup_from_page(oldhpage); - set_hugetlb_cgroup(oldhpage, NULL); + h_cg = hugetlb_cgroup_from_page(oldhpage, false); + h_cg_rsvd = hugetlb_cgroup_from_page(oldhpage, true); + set_hugetlb_cgroup(oldhpage, NULL, false);
/* move the h_cg details to new cgroup */ - set_hugetlb_cgroup(newhpage, h_cg); + set_hugetlb_cgroup(newhpage, h_cg_rsvd, true); list_move(&newhpage->lru, &h->hugepage_activelist); spin_unlock(&hugetlb_lock); return; -- 2.25.0.341.g760bfbb309-goog
On 2/3/20 3:22 PM, Mina Almasry wrote:
Augments hugetlb_cgroup_charge_cgroup to be able to charge hugetlb usage or hugetlb reservation counter.
Adds a new interface to uncharge a hugetlb_cgroup counter via hugetlb_cgroup_uncharge_counter.
Integrates the counter with hugetlb_cgroup, via hugetlb_cgroup_init, hugetlb_cgroup_have_usage, and hugetlb_cgroup_css_offline.
Signed-off-by: Mina Almasry almasrymina@google.com Acked-by: Mike Kravetz mike.kravetz@oracle.com Acked-by: David Rientjes rientjes@google.com
Changes in v11:
- Changed all 'reserved' or 'reservation' to 'rsvd' to reflect the user
interface.
Thanks.
Small nit,
@@ -450,8 +496,7 @@ static ssize_t hugetlb_cgroup_reset(struct kernfs_open_file *of, struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(of_css(of));
counter = &h_cg->hugepage[MEMFILE_IDX(of_cft(of)->private)];
- rsvd_counter =
&h_cg->rsvd_hugepage[MEMFILE_IDX(of_cft(of)->private)];
- rsvd_counter = &h_cg->rsvd_hugepage[MEMFILE_IDX(of_cft(of)->private)];
That looks like a change just to reformat a line added in the first patch?
switch (MEMFILE_ATTR(of_cft(of)->private)) { case RES_MAX_USAGE:
On Wed, Feb 5, 2020 at 2:08 PM Mike Kravetz mike.kravetz@oracle.com wrote:
On 2/3/20 3:22 PM, Mina Almasry wrote:
Augments hugetlb_cgroup_charge_cgroup to be able to charge hugetlb usage or hugetlb reservation counter.
Adds a new interface to uncharge a hugetlb_cgroup counter via hugetlb_cgroup_uncharge_counter.
Integrates the counter with hugetlb_cgroup, via hugetlb_cgroup_init, hugetlb_cgroup_have_usage, and hugetlb_cgroup_css_offline.
Signed-off-by: Mina Almasry almasrymina@google.com Acked-by: Mike Kravetz mike.kravetz@oracle.com Acked-by: David Rientjes rientjes@google.com
Changes in v11:
- Changed all 'reserved' or 'reservation' to 'rsvd' to reflect the user
interface.
Thanks.
Small nit,
@@ -450,8 +496,7 @@ static ssize_t hugetlb_cgroup_reset(struct kernfs_open_file *of, struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(of_css(of));
counter = &h_cg->hugepage[MEMFILE_IDX(of_cft(of)->private)];
rsvd_counter =&h_cg->rsvd_hugepage[MEMFILE_IDX(of_cft(of)->private)];
rsvd_counter = &h_cg->rsvd_hugepage[MEMFILE_IDX(of_cft(of)->private)];That looks like a change just to reformat a line added in the first patch?
switch (MEMFILE_ATTR(of_cft(of)->private)) { case RES_MAX_USAGE:
Gah, my bad. I'll move this to patch 1.
-- Mike Kravetz
Normally the pointer to the cgroup to uncharge hangs off the struct page, and gets queried when it's time to free the page. With hugetlb_cgroup reservations, this is not possible. Because it's possible for a page to be reserved by one task and actually faulted in by another task.
The best place to put the hugetlb_cgroup pointer to uncharge for reservations is in the resv_map. But, because the resv_map has different semantics for private and shared mappings, the code patch to charge/uncharge shared and private mappings is different. This patch implements charging and uncharging for private mappings.
For private mappings, the counter to uncharge is in resv_map->reservation_counter. On initializing the resv_map this is set to NULL. On reservation of a region in private mapping, the tasks hugetlb_cgroup is charged and the hugetlb_cgroup is placed is resv_map->reservation_counter.
On hugetlb_vm_op_close, we uncharge resv_map->reservation_counter.
Signed-off-by: Mina Almasry almasrymina@google.com Reviewed-by: Mike Kravetz mike.kravetz@oracle.com Acked-by: David Rientjes rientjes@google.com
---
Changes in v11: - Refactored hugetlb_cgroup_uncharge_conuter a bit to eliminate unnecessary #ifdefs. - Added resv_map_set_hugetlb_cgroup_uncharge_info() to eliminate #ifdefs in the middle of hugetlb logic.
Changes in v10: - Fixed cases where the mapping is private but cgroup accounting is disabled.
Changes in V9: - Updated for reparenting of hugetlb reservation accounting.
--- include/linux/hugetlb.h | 10 ++++++++ include/linux/hugetlb_cgroup.h | 39 +++++++++++++++++++++++++--- mm/hugetlb.c | 47 +++++++++++++++++++++++++++++++--- mm/hugetlb_cgroup.c | 41 +++++------------------------ 4 files changed, 97 insertions(+), 40 deletions(-)
diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h index dea6143aa0685..5491932ea5758 100644 --- a/include/linux/hugetlb.h +++ b/include/linux/hugetlb.h @@ -46,6 +46,16 @@ struct resv_map { long adds_in_progress; struct list_head region_cache; long region_cache_count; +#ifdef CONFIG_CGROUP_HUGETLB + /* + * On private mappings, the counter to uncharge reservations is stored + * here. If these fields are 0, then either the mapping is shared, or + * cgroup accounting is disabled for this resv_map. + */ + struct page_counter *reservation_counter; + unsigned long pages_per_hpage; + struct cgroup_subsys_state *css; +#endif }; extern struct resv_map *resv_map_alloc(void); void resv_map_release(struct kref *ref); diff --git a/include/linux/hugetlb_cgroup.h b/include/linux/hugetlb_cgroup.h index 91bc5e920e9d0..6a6c80df95ae3 100644 --- a/include/linux/hugetlb_cgroup.h +++ b/include/linux/hugetlb_cgroup.h @@ -27,6 +27,33 @@ struct hugetlb_cgroup; #define HUGETLB_CGROUP_MIN_ORDER 2
#ifdef CONFIG_CGROUP_HUGETLB +enum hugetlb_memory_event { + HUGETLB_MAX, + HUGETLB_NR_MEMORY_EVENTS, +}; + +struct hugetlb_cgroup { + struct cgroup_subsys_state css; + + /* + * the counter to account for hugepages from hugetlb. + */ + struct page_counter hugepage[HUGE_MAX_HSTATE]; + + /* + * the counter to account for hugepage reservations from hugetlb. + */ + struct page_counter rsvd_hugepage[HUGE_MAX_HSTATE]; + + atomic_long_t events[HUGE_MAX_HSTATE][HUGETLB_NR_MEMORY_EVENTS]; + atomic_long_t events_local[HUGE_MAX_HSTATE][HUGETLB_NR_MEMORY_EVENTS]; + + /* Handle for "hugetlb.events" */ + struct cgroup_file events_file[HUGE_MAX_HSTATE]; + + /* Handle for "hugetlb.events.local" */ + struct cgroup_file events_local_file[HUGE_MAX_HSTATE]; +};
static inline struct hugetlb_cgroup *hugetlb_cgroup_from_page(struct page *page, bool rsvd) @@ -71,9 +98,9 @@ extern void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages, extern void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages, struct hugetlb_cgroup *h_cg, bool rsvd); -extern void hugetlb_cgroup_uncharge_counter(struct page_counter *p, - unsigned long nr_pages, - struct cgroup_subsys_state *css); +extern void hugetlb_cgroup_uncharge_counter(struct resv_map *resv, + unsigned long start, + unsigned long end);
extern void hugetlb_cgroup_file_init(void) __init; extern void hugetlb_cgroup_migrate(struct page *oldhpage, @@ -122,6 +149,12 @@ static inline void hugetlb_cgroup_uncharge_cgroup(int idx, { }
+static inline void hugetlb_cgroup_uncharge_counter(struct resv_map *resv, + unsigned long start, + unsigned long end) +{ +} + static inline void hugetlb_cgroup_file_init(void) { } diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 62a4cf3db4090..fd2f7d1d31f4e 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -650,6 +650,25 @@ static void set_vma_private_data(struct vm_area_struct *vma, vma->vm_private_data = (void *)value; }
+static void +resv_map_set_hugetlb_cgroup_uncharge_info(struct resv_map *resv_map, + struct hugetlb_cgroup *h_cg, + struct hstate *h) +{ +#ifdef CONFIG_CGROUP_HUGETLB + if (!h_cg || !h) { + resv_map->reservation_counter = NULL; + resv_map->pages_per_hpage = 0; + resv_map->css = NULL; + } else { + resv_map->reservation_counter = + &h_cg->rsvd_hugepage[hstate_index(h)]; + resv_map->pages_per_hpage = pages_per_huge_page(h); + resv_map->css = &h_cg->css; + } +#endif +} + struct resv_map *resv_map_alloc(void) { struct resv_map *resv_map = kmalloc(sizeof(*resv_map), GFP_KERNEL); @@ -666,6 +685,13 @@ struct resv_map *resv_map_alloc(void) INIT_LIST_HEAD(&resv_map->regions);
resv_map->adds_in_progress = 0; + /* + * Initialize these to 0. On shared mappings, 0's here indicate these + * fields don't do cgroup accounting. On private mappings, these will be + * re-initialized to the proper values, to indicate that hugetlb cgroup + * reservations are to be un-charged from here. + */ + resv_map_set_hugetlb_cgroup_uncharge_info(resv_map, NULL, NULL);
INIT_LIST_HEAD(&resv_map->region_cache); list_add(&rg->link, &resv_map->region_cache); @@ -3193,9 +3219,7 @@ static void hugetlb_vm_op_close(struct vm_area_struct *vma) end = vma_hugecache_offset(h, vma, vma->vm_end);
reserve = (end - start) - region_count(resv, start, end); - - kref_put(&resv->refs, resv_map_release); - + hugetlb_cgroup_uncharge_counter(resv, start, end); if (reserve) { /* * Decrement reserve counts. The global reserve count may be @@ -3204,6 +3228,8 @@ static void hugetlb_vm_op_close(struct vm_area_struct *vma) gbl_reserve = hugepage_subpool_put_pages(spool, reserve); hugetlb_acct_memory(h, -gbl_reserve); } + + kref_put(&resv->refs, resv_map_release); }
static int hugetlb_vm_op_split(struct vm_area_struct *vma, unsigned long addr) @@ -4550,6 +4576,7 @@ int hugetlb_reserve_pages(struct inode *inode, struct hstate *h = hstate_inode(inode); struct hugepage_subpool *spool = subpool_inode(inode); struct resv_map *resv_map; + struct hugetlb_cgroup *h_cg; long gbl_reserve;
/* This should never happen */ @@ -4583,12 +4610,26 @@ int hugetlb_reserve_pages(struct inode *inode, chg = region_chg(resv_map, from, to);
} else { + /* Private mapping. */ resv_map = resv_map_alloc(); if (!resv_map) return -ENOMEM;
chg = to - from;
+ if (hugetlb_cgroup_charge_cgroup(hstate_index(h), + chg * pages_per_huge_page(h), + &h_cg, true)) { + kref_put(&resv_map->refs, resv_map_release); + return -ENOMEM; + } + + /* + * Since this branch handles private mappings, we attach the + * counter to uncharge for this reservation off resv_map. + */ + resv_map_set_hugetlb_cgroup_uncharge_info(resv_map, h_cg, h); + set_vma_resv_map(vma, resv_map); set_vma_resv_flags(vma, HPAGE_RESV_OWNER); } diff --git a/mm/hugetlb_cgroup.c b/mm/hugetlb_cgroup.c index 640555d7f78c6..e079513c8de0d 100644 --- a/mm/hugetlb_cgroup.c +++ b/mm/hugetlb_cgroup.c @@ -23,34 +23,6 @@ #include <linux/hugetlb.h> #include <linux/hugetlb_cgroup.h>
-enum hugetlb_memory_event { - HUGETLB_MAX, - HUGETLB_NR_MEMORY_EVENTS, -}; - -struct hugetlb_cgroup { - struct cgroup_subsys_state css; - - /* - * the counter to account for hugepages from hugetlb. - */ - struct page_counter hugepage[HUGE_MAX_HSTATE]; - - /* - * the counter to account for hugepage reservations from hugetlb. - */ - struct page_counter rsvd_hugepage[HUGE_MAX_HSTATE]; - - atomic_long_t events[HUGE_MAX_HSTATE][HUGETLB_NR_MEMORY_EVENTS]; - atomic_long_t events_local[HUGE_MAX_HSTATE][HUGETLB_NR_MEMORY_EVENTS]; - - /* Handle for "hugetlb.events" */ - struct cgroup_file events_file[HUGE_MAX_HSTATE]; - - /* Handle for "hugetlb.events.local" */ - struct cgroup_file events_local_file[HUGE_MAX_HSTATE]; -}; - #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) #define MEMFILE_IDX(val) (((val) >> 16) & 0xffff) #define MEMFILE_ATTR(val) ((val) & 0xffff) @@ -342,15 +314,16 @@ void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages, css_put(&h_cg->css); }
-void hugetlb_cgroup_uncharge_counter(struct page_counter *p, - unsigned long nr_pages, - struct cgroup_subsys_state *css) +void hugetlb_cgroup_uncharge_counter(struct resv_map *resv, unsigned long start, + unsigned long end) { - if (hugetlb_cgroup_disabled() || !p || !css) + if (hugetlb_cgroup_disabled() || !resv || !resv->reservation_counter || + !resv->css) return;
- page_counter_uncharge(p, nr_pages); - css_put(css); + page_counter_uncharge(resv->reservation_counter, + (end - start) * resv->pages_per_hpage); + css_put(resv->css); }
enum { -- 2.25.0.341.g760bfbb309-goog
On 2/3/20 3:22 PM, Mina Almasry wrote:
Normally the pointer to the cgroup to uncharge hangs off the struct page, and gets queried when it's time to free the page. With hugetlb_cgroup reservations, this is not possible. Because it's possible for a page to be reserved by one task and actually faulted in by another task.
The best place to put the hugetlb_cgroup pointer to uncharge for reservations is in the resv_map. But, because the resv_map has different semantics for private and shared mappings, the code patch to charge/uncharge shared and private mappings is different. This patch implements charging and uncharging for private mappings.
For private mappings, the counter to uncharge is in resv_map->reservation_counter. On initializing the resv_map this is set to NULL. On reservation of a region in private mapping, the tasks hugetlb_cgroup is charged and the hugetlb_cgroup is placed is resv_map->reservation_counter.
On hugetlb_vm_op_close, we uncharge resv_map->reservation_counter.
Signed-off-by: Mina Almasry almasrymina@google.com Reviewed-by: Mike Kravetz mike.kravetz@oracle.com Acked-by: David Rientjes rientjes@google.com
Changes in v11:
- Refactored hugetlb_cgroup_uncharge_conuter a bit to eliminate
unnecessary #ifdefs.
- Added resv_map_set_hugetlb_cgroup_uncharge_info() to eliminate #ifdefs
in the middle of hugetlb logic.
Thanks. Code looks better without the #ifdefs. You can keep my Reviewed-by:
A follow up patch in this series adds hugetlb cgroup uncharge info the file_region entries in resv->regions. The cgroup uncharge info may differ for different regions, so they can no longer be coalesced at region_add time. So, disable region coalescing in region_add in this patch.
Behavior change:
Say a resv_map exists like this [0->1], [2->3], and [5->6].
Then a region_chg/add call comes in region_chg/add(f=0, t=5).
Old code would generate resv->regions: [0->5], [5->6]. New code would generate resv->regions: [0->1], [1->2], [2->3], [3->5], [5->6].
Special care needs to be taken to handle the resv->adds_in_progress variable correctly. In the past, only 1 region would be added for every region_chg and region_add call. But now, each call may add multiple regions, so we can no longer increment adds_in_progress by 1 in region_chg, or decrement adds_in_progress by 1 after region_add or region_abort. Instead, region_chg calls add_reservation_in_range() to count the number of regions needed and allocates those, and that info is passed to region_add and region_abort to decrement adds_in_progress correctly.
We've also modified the assumption that region_add after region_chg never fails. region_chg now pre-allocates at least 1 region for region_add. If region_add needs more regions than region_chg has allocated for it, then it may fail.
Signed-off-by: Mina Almasry almasrymina@google.com Reviewed-by: Mike Kravetz mike.kravetz@oracle.com
---
Changes in v9: - Added clarifications in the comments and addressed minor issues from code review.
Changes in v7: - region_chg no longer allocates (t-f) / 2 file_region entries.
Changes in v6: - Fix bug in number of region_caches allocated by region_chg
--- mm/hugetlb.c | 327 +++++++++++++++++++++++++++++++++++---------------- 1 file changed, 223 insertions(+), 104 deletions(-)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index fd2f7d1d31f4e..986e9a9cc6fbe 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -245,110 +245,179 @@ struct file_region { long to; };
+/* Helper that removes a struct file_region from the resv_map cache and returns + * it for use. + */ +static struct file_region * +get_file_region_entry_from_cache(struct resv_map *resv, long from, long to) +{ + struct file_region *nrg = NULL; + + VM_BUG_ON(resv->region_cache_count <= 0); + + resv->region_cache_count--; + nrg = list_first_entry(&resv->region_cache, struct file_region, link); + VM_BUG_ON(!nrg); + list_del(&nrg->link); + + nrg->from = from; + nrg->to = to; + + return nrg; +} + /* Must be called with resv->lock held. Calling this with count_only == true * will count the number of pages to be added but will not modify the linked - * list. + * list. If regions_needed != NULL and count_only == true, then regions_needed + * will indicate the number of file_regions needed in the cache to carry out to + * add the regions for this range. */ static long add_reservation_in_range(struct resv_map *resv, long f, long t, - bool count_only) + long *regions_needed, bool count_only) { - long chg = 0; + long add = 0; struct list_head *head = &resv->regions; + long last_accounted_offset = f; struct file_region *rg = NULL, *trg = NULL, *nrg = NULL;
- /* Locate the region we are before or in. */ - list_for_each_entry(rg, head, link) - if (f <= rg->to) - break; - - /* Round our left edge to the current segment if it encloses us. */ - if (f > rg->from) - f = rg->from; + if (regions_needed) + *regions_needed = 0;
- chg = t - f; + /* In this loop, we essentially handle an entry for the range + * [last_accounted_offset, rg->from), at every iteration, with some + * bounds checking. + */ + list_for_each_entry_safe(rg, trg, head, link) { + /* Skip irrelevant regions that start before our range. */ + if (rg->from < f) { + /* If this region ends after the last accounted offset, + * then we need to update last_accounted_offset. + */ + if (rg->to > last_accounted_offset) + last_accounted_offset = rg->to; + continue; + }
- /* Check for and consume any regions we now overlap with. */ - nrg = rg; - list_for_each_entry_safe(rg, trg, rg->link.prev, link) { - if (&rg->link == head) - break; + /* When we find a region that starts beyond our range, we've + * finished. + */ if (rg->from > t) break;
- /* We overlap with this area, if it extends further than - * us then we must extend ourselves. Account for its - * existing reservation. + /* Add an entry for last_accounted_offset -> rg->from, and + * update last_accounted_offset. */ - if (rg->to > t) { - chg += rg->to - t; - t = rg->to; + if (rg->from > last_accounted_offset) { + add += rg->from - last_accounted_offset; + if (!count_only) { + nrg = get_file_region_entry_from_cache( + resv, last_accounted_offset, rg->from); + list_add(&nrg->link, rg->link.prev); + } else if (regions_needed) + *regions_needed += 1; } - chg -= rg->to - rg->from;
- if (!count_only && rg != nrg) { - list_del(&rg->link); - kfree(rg); - } + last_accounted_offset = rg->to; }
- if (!count_only) { - nrg->from = f; - nrg->to = t; + /* Handle the case where our range extends beyond + * last_accounted_offset. + */ + if (last_accounted_offset < t) { + add += t - last_accounted_offset; + if (!count_only) { + nrg = get_file_region_entry_from_cache( + resv, last_accounted_offset, t); + list_add(&nrg->link, rg->link.prev); + } else if (regions_needed) + *regions_needed += 1; }
- return chg; + return add; }
/* * Add the huge page range represented by [f, t) to the reserve - * map. Existing regions will be expanded to accommodate the specified - * range, or a region will be taken from the cache. Sufficient regions - * must exist in the cache due to the previous call to region_chg with - * the same range. + * map. Regions will be taken from the cache to fill in this range. + * Sufficient regions should exist in the cache due to the previous + * call to region_chg with the same range, but in some cases the cache will not + * have sufficient entries due to races with other code doing region_add or + * region_del. The extra needed entries will be allocated. * - * Return the number of new huge pages added to the map. This - * number is greater than or equal to zero. + * regions_needed is the out value provided by a previous call to region_chg. + * + * Return the number of new huge pages added to the map. This number is greater + * than or equal to zero. If file_region entries needed to be allocated for + * this operation and we were not able to allocate, it ruturns -ENOMEM. + * region_add of regions of length 1 never allocate file_regions and cannot + * fail; region_chg will always allocate at least 1 entry and a region_add for + * 1 page will only require at most 1 entry. */ -static long region_add(struct resv_map *resv, long f, long t) +static long region_add(struct resv_map *resv, long f, long t, + long in_regions_needed) { - struct list_head *head = &resv->regions; - struct file_region *rg, *nrg; - long add = 0; + long add = 0, actual_regions_needed = 0, i = 0; + struct file_region *trg = NULL, *rg = NULL; + struct list_head allocated_regions; + + INIT_LIST_HEAD(&allocated_regions);
spin_lock(&resv->lock); - /* Locate the region we are either in or before. */ - list_for_each_entry(rg, head, link) - if (f <= rg->to) - break; +retry: + + /* Count how many regions are actually needed to execute this add. */ + add_reservation_in_range(resv, f, t, &actual_regions_needed, true);
/* - * If no region exists which can be expanded to include the - * specified range, pull a region descriptor from the cache - * and use it for this range. + * Check for sufficient descriptors in the cache to accommodate + * this add operation. Note that actual_regions_needed may be greater + * than in_regions_needed. In this case, we need to make sure that we + * allocate extra entries, such that we have enough for all the + * existing adds_in_progress, plus the excess needed for this + * operation. */ - if (&rg->link == head || t < rg->from) { - VM_BUG_ON(resv->region_cache_count <= 0); + if (resv->region_cache_count < + resv->adds_in_progress + + (actual_regions_needed - in_regions_needed)) { + /* region_add operation of range 1 should never need to + * allocate file_region entries. + */ + VM_BUG_ON(t - f <= 1);
- resv->region_cache_count--; - nrg = list_first_entry(&resv->region_cache, struct file_region, - link); - list_del(&nrg->link); + /* Must drop lock to allocate a new descriptor. */ + spin_unlock(&resv->lock); + for (i = 0; i < (actual_regions_needed - in_regions_needed); + i++) { + trg = kmalloc(sizeof(*trg), GFP_KERNEL); + if (!trg) + goto out_of_memory; + list_add(&trg->link, &allocated_regions); + } + spin_lock(&resv->lock);
- nrg->from = f; - nrg->to = t; - list_add(&nrg->link, rg->link.prev); + list_for_each_entry_safe(rg, trg, &allocated_regions, link) { + list_del(&rg->link); + list_add(&rg->link, &resv->region_cache); + resv->region_cache_count++; + }
- add += t - f; - goto out_locked; + goto retry; }
- add = add_reservation_in_range(resv, f, t, false); + add = add_reservation_in_range(resv, f, t, NULL, false); + + resv->adds_in_progress -= in_regions_needed;
-out_locked: - resv->adds_in_progress--; spin_unlock(&resv->lock); VM_BUG_ON(add < 0); return add; + +out_of_memory: + list_for_each_entry_safe(rg, trg, &allocated_regions, link) { + list_del(&rg->link); + kfree(rg); + } + return -ENOMEM; }
/* @@ -358,49 +427,79 @@ static long region_add(struct resv_map *resv, long f, long t) * call to region_add that will actually modify the reserve * map to add the specified range [f, t). region_chg does * not change the number of huge pages represented by the - * map. A new file_region structure is added to the cache - * as a placeholder, so that the subsequent region_add - * call will have all the regions it needs and will not fail. + * map. A number of new file_region structures is added to the cache as a + * placeholder, for the subsequent region_add call to use. At least 1 + * file_region structure is added. + * + * out_regions_needed is the number of regions added to the + * resv->adds_in_progress. This value needs to be provided to a follow up call + * to region_add or region_abort for proper accounting. * * Returns the number of huge pages that need to be added to the existing * reservation map for the range [f, t). This number is greater or equal to * zero. -ENOMEM is returned if a new file_region structure or cache entry * is needed and can not be allocated. */ -static long region_chg(struct resv_map *resv, long f, long t) +static long region_chg(struct resv_map *resv, long f, long t, + long *out_regions_needed) { - long chg = 0; + struct file_region *trg = NULL, *rg = NULL; + long chg = 0, i = 0, to_allocate = 0; + struct list_head allocated_regions; + + INIT_LIST_HEAD(&allocated_regions);
spin_lock(&resv->lock); -retry_locked: - resv->adds_in_progress++; + + /* Count how many hugepages in this range are NOT respresented. */ + chg = add_reservation_in_range(resv, f, t, out_regions_needed, true); + + if (*out_regions_needed == 0) + *out_regions_needed = 1; + + resv->adds_in_progress += *out_regions_needed;
/* * Check for sufficient descriptors in the cache to accommodate * the number of in progress add operations. */ - if (resv->adds_in_progress > resv->region_cache_count) { - struct file_region *trg; - - VM_BUG_ON(resv->adds_in_progress - resv->region_cache_count > 1); - /* Must drop lock to allocate a new descriptor. */ - resv->adds_in_progress--; + while (resv->region_cache_count < resv->adds_in_progress) { + to_allocate = resv->adds_in_progress - resv->region_cache_count; + + /* Must drop lock to allocate a new descriptor. Note that even + * though we drop the lock here, we do not make another call to + * add_reservation_in_range after re-acquiring the lock. + * Essentially this branch makes sure that we have enough + * descriptors in the cache as suggested by the first call to + * add_reservation_in_range. If more regions turn out to be + * required, region_add will deal with it. + */ spin_unlock(&resv->lock); - - trg = kmalloc(sizeof(*trg), GFP_KERNEL); - if (!trg) - return -ENOMEM; + for (i = 0; i < to_allocate; i++) { + trg = kmalloc(sizeof(*trg), GFP_KERNEL); + if (!trg) + goto out_of_memory; + list_add(&trg->link, &allocated_regions); + }
spin_lock(&resv->lock); - list_add(&trg->link, &resv->region_cache); - resv->region_cache_count++; - goto retry_locked; - }
- chg = add_reservation_in_range(resv, f, t, true); + list_for_each_entry_safe(rg, trg, &allocated_regions, link) { + list_del(&rg->link); + list_add(&rg->link, &resv->region_cache); + resv->region_cache_count++; + } + }
spin_unlock(&resv->lock); return chg; + +out_of_memory: + list_for_each_entry_safe(rg, trg, &allocated_regions, link) { + list_del(&rg->link); + kfree(rg); + } + return -ENOMEM; }
/* @@ -408,17 +507,20 @@ static long region_chg(struct resv_map *resv, long f, long t) * of the resv_map keeps track of the operations in progress between * calls to region_chg and region_add. Operations are sometimes * aborted after the call to region_chg. In such cases, region_abort - * is called to decrement the adds_in_progress counter. + * is called to decrement the adds_in_progress counter. regions_needed + * is the value returned by the region_chg call, it is used to decrement + * the adds_in_progress counter. * * NOTE: The range arguments [f, t) are not needed or used in this * routine. They are kept to make reading the calling code easier as * arguments will match the associated region_chg call. */ -static void region_abort(struct resv_map *resv, long f, long t) +static void region_abort(struct resv_map *resv, long f, long t, + long regions_needed) { spin_lock(&resv->lock); VM_BUG_ON(!resv->region_cache_count); - resv->adds_in_progress--; + resv->adds_in_progress -= regions_needed; spin_unlock(&resv->lock); }
@@ -1898,6 +2000,7 @@ static long __vma_reservation_common(struct hstate *h, struct resv_map *resv; pgoff_t idx; long ret; + long dummy_out_regions_needed;
resv = vma_resv_map(vma); if (!resv) @@ -1906,20 +2009,29 @@ static long __vma_reservation_common(struct hstate *h, idx = vma_hugecache_offset(h, vma, addr); switch (mode) { case VMA_NEEDS_RESV: - ret = region_chg(resv, idx, idx + 1); + ret = region_chg(resv, idx, idx + 1, &dummy_out_regions_needed); + /* We assume that vma_reservation_* routines always operate on + * 1 page, and that adding to resv map a 1 page entry can only + * ever require 1 region. + */ + VM_BUG_ON(dummy_out_regions_needed != 1); break; case VMA_COMMIT_RESV: - ret = region_add(resv, idx, idx + 1); + ret = region_add(resv, idx, idx + 1, 1); + /* region_add calls of range 1 should never fail. */ + VM_BUG_ON(ret < 0); break; case VMA_END_RESV: - region_abort(resv, idx, idx + 1); + region_abort(resv, idx, idx + 1, 1); ret = 0; break; case VMA_ADD_RESV: - if (vma->vm_flags & VM_MAYSHARE) - ret = region_add(resv, idx, idx + 1); - else { - region_abort(resv, idx, idx + 1); + if (vma->vm_flags & VM_MAYSHARE) { + ret = region_add(resv, idx, idx + 1, 1); + /* region_add calls of range 1 should never fail. */ + VM_BUG_ON(ret < 0); + } else { + region_abort(resv, idx, idx + 1, 1); ret = region_del(resv, idx, idx + 1); } break; @@ -4572,12 +4684,12 @@ int hugetlb_reserve_pages(struct inode *inode, struct vm_area_struct *vma, vm_flags_t vm_flags) { - long ret, chg; + long ret, chg, add = -1; struct hstate *h = hstate_inode(inode); struct hugepage_subpool *spool = subpool_inode(inode); struct resv_map *resv_map; struct hugetlb_cgroup *h_cg; - long gbl_reserve; + long gbl_reserve, regions_needed = 0;
/* This should never happen */ if (from > to) { @@ -4607,7 +4719,7 @@ int hugetlb_reserve_pages(struct inode *inode, */ resv_map = inode_resv_map(inode);
- chg = region_chg(resv_map, from, to); + chg = region_chg(resv_map, from, to, ®ions_needed);
} else { /* Private mapping. */ @@ -4673,9 +4785,14 @@ int hugetlb_reserve_pages(struct inode *inode, * else has to be done for private mappings here */ if (!vma || vma->vm_flags & VM_MAYSHARE) { - long add = region_add(resv_map, from, to); - - if (unlikely(chg > add)) { + add = region_add(resv_map, from, to, regions_needed); + + if (unlikely(add < 0)) { + hugetlb_acct_memory(h, -gbl_reserve); + /* put back original number of pages, chg */ + (void)hugepage_subpool_put_pages(spool, chg); + goto out_err; + } else if (unlikely(chg > add)) { /* * pages in this range were added to the reserve * map between region_chg and region_add. This @@ -4693,9 +4810,11 @@ int hugetlb_reserve_pages(struct inode *inode, return 0; out_err: if (!vma || vma->vm_flags & VM_MAYSHARE) - /* Don't call region_abort if region_chg failed */ - if (chg >= 0) - region_abort(resv_map, from, to); + /* Only call region_abort if the region_chg succeeded but the + * region_add failed or didn't run. + */ + if (chg >= 0 && add < 0) + region_abort(resv_map, from, to, regions_needed); if (vma && is_vma_resv_set(vma, HPAGE_RESV_OWNER)) kref_put(&resv_map->refs, resv_map_release); return ret; -- 2.25.0.341.g760bfbb309-goog
On 2/3/20 3:22 PM, Mina Almasry wrote:
A follow up patch in this series adds hugetlb cgroup uncharge info the file_region entries in resv->regions. The cgroup uncharge info may differ for different regions, so they can no longer be coalesced at region_add time. So, disable region coalescing in region_add in this patch.
Behavior change:
Say a resv_map exists like this [0->1], [2->3], and [5->6].
Then a region_chg/add call comes in region_chg/add(f=0, t=5).
Old code would generate resv->regions: [0->5], [5->6]. New code would generate resv->regions: [0->1], [1->2], [2->3], [3->5], [5->6].
Special care needs to be taken to handle the resv->adds_in_progress variable correctly. In the past, only 1 region would be added for every region_chg and region_add call. But now, each call may add multiple regions, so we can no longer increment adds_in_progress by 1 in region_chg, or decrement adds_in_progress by 1 after region_add or region_abort. Instead, region_chg calls add_reservation_in_range() to count the number of regions needed and allocates those, and that info is passed to region_add and region_abort to decrement adds_in_progress correctly.
We've also modified the assumption that region_add after region_chg never fails. region_chg now pre-allocates at least 1 region for region_add. If region_add needs more regions than region_chg has allocated for it, then it may fail.
Signed-off-by: Mina Almasry almasrymina@google.com Reviewed-by: Mike Kravetz mike.kravetz@oracle.com
This is the same as the previous version. My late comment was that we need to rethink the disabling of region coalescing. This is especially important for private mappings where there will be one region for huge page. I know that you are working on this issue. Please remove my Reviewed-by: when sending out the next version.
Thanks,
On Wed, Feb 5, 2020 at 3:57 PM Mike Kravetz mike.kravetz@oracle.com wrote:
On 2/3/20 3:22 PM, Mina Almasry wrote:
A follow up patch in this series adds hugetlb cgroup uncharge info the file_region entries in resv->regions. The cgroup uncharge info may differ for different regions, so they can no longer be coalesced at region_add time. So, disable region coalescing in region_add in this patch.
Behavior change:
Say a resv_map exists like this [0->1], [2->3], and [5->6].
Then a region_chg/add call comes in region_chg/add(f=0, t=5).
Old code would generate resv->regions: [0->5], [5->6]. New code would generate resv->regions: [0->1], [1->2], [2->3], [3->5], [5->6].
Special care needs to be taken to handle the resv->adds_in_progress variable correctly. In the past, only 1 region would be added for every region_chg and region_add call. But now, each call may add multiple regions, so we can no longer increment adds_in_progress by 1 in region_chg, or decrement adds_in_progress by 1 after region_add or region_abort. Instead, region_chg calls add_reservation_in_range() to count the number of regions needed and allocates those, and that info is passed to region_add and region_abort to decrement adds_in_progress correctly.
We've also modified the assumption that region_add after region_chg never fails. region_chg now pre-allocates at least 1 region for region_add. If region_add needs more regions than region_chg has allocated for it, then it may fail.
Signed-off-by: Mina Almasry almasrymina@google.com Reviewed-by: Mike Kravetz mike.kravetz@oracle.com
This is the same as the previous version. My late comment was that we need to rethink the disabling of region coalescing. This is especially important for private mappings where there will be one region for huge page. I know that you are working on this issue. Please remove my Reviewed-by: when sending out the next version.
Yes to address that there is a new patch in the series, which re-enables the coalescing when the hugetlb cgroup uncharge info is the same. I guess it could be squashed with this one but I thought it was better to implement that patch on top of the patch that enabled shared accounting, because that is the patch that sets hugetlb cgroup info on the file region entries.
Let me know what you think.
Thanks,
Mike Kravetz
On 2/5/20 5:43 PM, Mina Almasry wrote:
On Wed, Feb 5, 2020 at 3:57 PM Mike Kravetz mike.kravetz@oracle.com wrote:
On 2/3/20 3:22 PM, Mina Almasry wrote:
A follow up patch in this series adds hugetlb cgroup uncharge info the file_region entries in resv->regions. The cgroup uncharge info may differ for different regions, so they can no longer be coalesced at region_add time. So, disable region coalescing in region_add in this patch.
Behavior change:
Say a resv_map exists like this [0->1], [2->3], and [5->6].
Then a region_chg/add call comes in region_chg/add(f=0, t=5).
Old code would generate resv->regions: [0->5], [5->6]. New code would generate resv->regions: [0->1], [1->2], [2->3], [3->5], [5->6].
Special care needs to be taken to handle the resv->adds_in_progress variable correctly. In the past, only 1 region would be added for every region_chg and region_add call. But now, each call may add multiple regions, so we can no longer increment adds_in_progress by 1 in region_chg, or decrement adds_in_progress by 1 after region_add or region_abort. Instead, region_chg calls add_reservation_in_range() to count the number of regions needed and allocates those, and that info is passed to region_add and region_abort to decrement adds_in_progress correctly.
We've also modified the assumption that region_add after region_chg never fails. region_chg now pre-allocates at least 1 region for region_add. If region_add needs more regions than region_chg has allocated for it, then it may fail.
Signed-off-by: Mina Almasry almasrymina@google.com Reviewed-by: Mike Kravetz mike.kravetz@oracle.com
This is the same as the previous version. My late comment was that we need to rethink the disabling of region coalescing. This is especially important for private mappings where there will be one region for huge page. I know that you are working on this issue. Please remove my Reviewed-by: when sending out the next version.
Yes to address that there is a new patch in the series, which re-enables the coalescing when the hugetlb cgroup uncharge info is the same. I guess it could be squashed with this one but I thought it was better to implement that patch on top of the patch that enabled shared accounting, because that is the patch that sets hugetlb cgroup info on the file region entries.
Let me know what you think.
Thanks, I saw there was an additional patch but I did not get to it yet. I'll take a look and see how involved the changes are.
For shared mappings, the pointer to the hugetlb_cgroup to uncharge lives in the resv_map entries, in file_region->reservation_counter.
After a call to region_chg, we charge the approprate hugetlb_cgroup, and if successful, we pass on the hugetlb_cgroup info to a follow up region_add call. When a file_region entry is added to the resv_map via region_add, we put the pointer to that cgroup in file_region->reservation_counter. If charging doesn't succeed, we report the error to the caller, so that the kernel fails the reservation.
On region_del, which is when the hugetlb memory is unreserved, we also uncharge the file_region->reservation_counter.
Signed-off-by: Mina Almasry almasrymina@google.com
---
Changes in v11: - Created new function, hugetlb_cgroup_uncharge_file_region to clean up some #ifdefs. - Moved file_region definition to hugetlb.h. - Added copy_hugetlb_cgroup_uncharge_info function to clean up more #ifdefs in the middle of hugetlb code.
Changes in v10: - Deleted duplicated code snippet.
Changes in V9: - Updated for hugetlb reservation repareting.
--- include/linux/hugetlb.h | 36 ++++++++ include/linux/hugetlb_cgroup.h | 10 +++ mm/hugetlb.c | 147 +++++++++++++++++++++------------ mm/hugetlb_cgroup.c | 15 ++++ 4 files changed, 155 insertions(+), 53 deletions(-)
diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h index 5491932ea5758..395f5b1fad416 100644 --- a/include/linux/hugetlb.h +++ b/include/linux/hugetlb.h @@ -57,6 +57,42 @@ struct resv_map { struct cgroup_subsys_state *css; #endif }; + +/* + * Region tracking -- allows tracking of reservations and instantiated pages + * across the pages in a mapping. + * + * The region data structures are embedded into a resv_map and protected + * by a resv_map's lock. The set of regions within the resv_map represent + * reservations for huge pages, or huge pages that have already been + * instantiated within the map. The from and to elements are huge page + * indicies into the associated mapping. from indicates the starting index + * of the region. to represents the first index past the end of the region. + * + * For example, a file region structure with from == 0 and to == 4 represents + * four huge pages in a mapping. It is important to note that the to element + * represents the first element past the end of the region. This is used in + * arithmetic as 4(to) - 0(from) = 4 huge pages in the region. + * + * Interval notation of the form [from, to) will be used to indicate that + * the endpoint from is inclusive and to is exclusive. + */ +struct file_region { + struct list_head link; + long from; + long to; +#ifdef CONFIG_CGROUP_HUGETLB + /* + * On shared mappings, each reserved region appears as a struct + * file_region in resv_map. These fields hold the info needed to + * uncharge each reservation. + */ + struct page_counter *reservation_counter; + unsigned long pages_per_hpage; + struct cgroup_subsys_state *css; +#endif +}; + extern struct resv_map *resv_map_alloc(void); void resv_map_release(struct kref *ref);
diff --git a/include/linux/hugetlb_cgroup.h b/include/linux/hugetlb_cgroup.h index 6a6c80df95ae3..c3fd417c268c5 100644 --- a/include/linux/hugetlb_cgroup.h +++ b/include/linux/hugetlb_cgroup.h @@ -102,11 +102,21 @@ extern void hugetlb_cgroup_uncharge_counter(struct resv_map *resv, unsigned long start, unsigned long end);
+extern void hugetlb_cgroup_uncharge_file_region(struct resv_map *resv, + struct file_region *rg, + unsigned long nr_pages); + extern void hugetlb_cgroup_file_init(void) __init; extern void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage);
#else + +static inline void hugetlb_cgroup_uncharge_file_region(struct resv_map *resv, + struct file_region *rg, + unsigned long nr_pages) +{ +} static inline struct hugetlb_cgroup *hugetlb_cgroup_from_page(struct page *page, bool rsvd) { diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 986e9a9cc6fbe..33818ccaf7e89 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -220,31 +220,6 @@ static inline struct hugepage_subpool *subpool_vma(struct vm_area_struct *vma) return subpool_inode(file_inode(vma->vm_file)); }
-/* - * Region tracking -- allows tracking of reservations and instantiated pages - * across the pages in a mapping. - * - * The region data structures are embedded into a resv_map and protected - * by a resv_map's lock. The set of regions within the resv_map represent - * reservations for huge pages, or huge pages that have already been - * instantiated within the map. The from and to elements are huge page - * indicies into the associated mapping. from indicates the starting index - * of the region. to represents the first index past the end of the region. - * - * For example, a file region structure with from == 0 and to == 4 represents - * four huge pages in a mapping. It is important to note that the to element - * represents the first element past the end of the region. This is used in - * arithmetic as 4(to) - 0(from) = 4 huge pages in the region. - * - * Interval notation of the form [from, to) will be used to indicate that - * the endpoint from is inclusive and to is exclusive. - */ -struct file_region { - struct list_head link; - long from; - long to; -}; - /* Helper that removes a struct file_region from the resv_map cache and returns * it for use. */ @@ -266,6 +241,37 @@ get_file_region_entry_from_cache(struct resv_map *resv, long from, long to) return nrg; }
+static void copy_hugetlb_cgroup_uncharge_info(struct file_region *nrg, + struct file_region *rg) +{ +#ifdef CONFIG_CGROUP_HUGETLB + + nrg->reservation_counter = rg->reservation_counter; + nrg->pages_per_hpage = rg->pages_per_hpage; + nrg->css = rg->css; + css_get(rg->css); +#endif +} + +/* Helper that records hugetlb_cgroup uncharge info. */ +static void record_hugetlb_cgroup_uncharge_info(struct hugetlb_cgroup *h_cg, + struct file_region *nrg, + struct hstate *h) +{ +#ifdef CONFIG_CGROUP_HUGETLB + if (h_cg) { + nrg->reservation_counter = + &h_cg->rsvd_hugepage[hstate_index(h)]; + nrg->pages_per_hpage = pages_per_huge_page(h); + nrg->css = &h_cg->css; + } else { + nrg->reservation_counter = NULL; + nrg->pages_per_hpage = 0; + nrg->css = NULL; + } +#endif +} + /* Must be called with resv->lock held. Calling this with count_only == true * will count the number of pages to be added but will not modify the linked * list. If regions_needed != NULL and count_only == true, then regions_needed @@ -273,7 +279,9 @@ get_file_region_entry_from_cache(struct resv_map *resv, long from, long to) * add the regions for this range. */ static long add_reservation_in_range(struct resv_map *resv, long f, long t, - long *regions_needed, bool count_only) + struct hugetlb_cgroup *h_cg, + struct hstate *h, long *regions_needed, + bool count_only) { long add = 0; struct list_head *head = &resv->regions; @@ -312,6 +320,8 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, if (!count_only) { nrg = get_file_region_entry_from_cache( resv, last_accounted_offset, rg->from); + record_hugetlb_cgroup_uncharge_info(h_cg, nrg, + h); list_add(&nrg->link, rg->link.prev); } else if (regions_needed) *regions_needed += 1; @@ -328,11 +338,13 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, if (!count_only) { nrg = get_file_region_entry_from_cache( resv, last_accounted_offset, t); + record_hugetlb_cgroup_uncharge_info(h_cg, nrg, h); list_add(&nrg->link, rg->link.prev); } else if (regions_needed) *regions_needed += 1; }
+ VM_BUG_ON(add < 0); return add; }
@@ -353,7 +365,8 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, * fail; region_chg will always allocate at least 1 entry and a region_add for * 1 page will only require at most 1 entry. */ -static long region_add(struct resv_map *resv, long f, long t, +static long region_add(struct hstate *h, struct hugetlb_cgroup *h_cg, + struct resv_map *resv, long f, long t, long in_regions_needed) { long add = 0, actual_regions_needed = 0, i = 0; @@ -366,7 +379,8 @@ static long region_add(struct resv_map *resv, long f, long t, retry:
/* Count how many regions are actually needed to execute this add. */ - add_reservation_in_range(resv, f, t, &actual_regions_needed, true); + add_reservation_in_range(resv, f, t, NULL, NULL, &actual_regions_needed, + true);
/* * Check for sufficient descriptors in the cache to accommodate @@ -404,7 +418,7 @@ static long region_add(struct resv_map *resv, long f, long t, goto retry; }
- add = add_reservation_in_range(resv, f, t, NULL, false); + add = add_reservation_in_range(resv, f, t, h_cg, h, NULL, false);
resv->adds_in_progress -= in_regions_needed;
@@ -452,7 +466,8 @@ static long region_chg(struct resv_map *resv, long f, long t, spin_lock(&resv->lock);
/* Count how many hugepages in this range are NOT respresented. */ - chg = add_reservation_in_range(resv, f, t, out_regions_needed, true); + chg = add_reservation_in_range(resv, f, t, NULL, NULL, + out_regions_needed, true);
if (*out_regions_needed == 0) *out_regions_needed = 1; @@ -588,11 +603,17 @@ static long region_del(struct resv_map *resv, long f, long t) /* New entry for end of split region */ nrg->from = t; nrg->to = rg->to; + + copy_hugetlb_cgroup_uncharge_info(nrg, rg); + INIT_LIST_HEAD(&nrg->link);
/* Original entry is trimmed */ rg->to = f;
+ hugetlb_cgroup_uncharge_file_region( + resv, rg, nrg->to - nrg->from); + list_add(&nrg->link, &rg->link); nrg = NULL; break; @@ -600,6 +621,8 @@ static long region_del(struct resv_map *resv, long f, long t)
if (f <= rg->from && t >= rg->to) { /* Remove entire region */ del += rg->to - rg->from; + hugetlb_cgroup_uncharge_file_region(resv, rg, + rg->to - rg->from); list_del(&rg->link); kfree(rg); continue; @@ -608,14 +631,21 @@ static long region_del(struct resv_map *resv, long f, long t) if (f <= rg->from) { /* Trim beginning of region */ del += t - rg->from; rg->from = t; + + hugetlb_cgroup_uncharge_file_region(resv, rg, + t - rg->from); } else { /* Trim end of region */ del += rg->to - f; rg->to = f; + + hugetlb_cgroup_uncharge_file_region(resv, rg, + rg->to - f); } }
spin_unlock(&resv->lock); kfree(nrg); + return del; }
@@ -2017,7 +2047,7 @@ static long __vma_reservation_common(struct hstate *h, VM_BUG_ON(dummy_out_regions_needed != 1); break; case VMA_COMMIT_RESV: - ret = region_add(resv, idx, idx + 1, 1); + ret = region_add(NULL, NULL, resv, idx, idx + 1, 1); /* region_add calls of range 1 should never fail. */ VM_BUG_ON(ret < 0); break; @@ -2027,7 +2057,7 @@ static long __vma_reservation_common(struct hstate *h, break; case VMA_ADD_RESV: if (vma->vm_flags & VM_MAYSHARE) { - ret = region_add(resv, idx, idx + 1, 1); + ret = region_add(NULL, NULL, resv, idx, idx + 1, 1); /* region_add calls of range 1 should never fail. */ VM_BUG_ON(ret < 0); } else { @@ -4688,7 +4718,7 @@ int hugetlb_reserve_pages(struct inode *inode, struct hstate *h = hstate_inode(inode); struct hugepage_subpool *spool = subpool_inode(inode); struct resv_map *resv_map; - struct hugetlb_cgroup *h_cg; + struct hugetlb_cgroup *h_cg = NULL; long gbl_reserve, regions_needed = 0;
/* This should never happen */ @@ -4729,19 +4759,6 @@ int hugetlb_reserve_pages(struct inode *inode,
chg = to - from;
- if (hugetlb_cgroup_charge_cgroup(hstate_index(h), - chg * pages_per_huge_page(h), - &h_cg, true)) { - kref_put(&resv_map->refs, resv_map_release); - return -ENOMEM; - } - - /* - * Since this branch handles private mappings, we attach the - * counter to uncharge for this reservation off resv_map. - */ - resv_map_set_hugetlb_cgroup_uncharge_info(resv_map, h_cg, h); - set_vma_resv_map(vma, resv_map); set_vma_resv_flags(vma, HPAGE_RESV_OWNER); } @@ -4751,6 +4768,21 @@ int hugetlb_reserve_pages(struct inode *inode, goto out_err; }
+ ret = hugetlb_cgroup_charge_cgroup( + hstate_index(h), chg * pages_per_huge_page(h), &h_cg, true); + + if (ret < 0) { + ret = -ENOMEM; + goto out_err; + } + + if (vma && !(vma->vm_flags & VM_MAYSHARE) && h_cg) { + /* For private mappings, the hugetlb_cgroup uncharge info hangs + * of the resv_map. + */ + resv_map_set_hugetlb_cgroup_uncharge_info(resv_map, h_cg, h); + } + /* * There must be enough pages in the subpool for the mapping. If * the subpool has a minimum size, there may be some global @@ -4759,7 +4791,7 @@ int hugetlb_reserve_pages(struct inode *inode, gbl_reserve = hugepage_subpool_get_pages(spool, chg); if (gbl_reserve < 0) { ret = -ENOSPC; - goto out_err; + goto out_uncharge_cgroup; }
/* @@ -4768,9 +4800,7 @@ int hugetlb_reserve_pages(struct inode *inode, */ ret = hugetlb_acct_memory(h, gbl_reserve); if (ret < 0) { - /* put back original number of pages, chg */ - (void)hugepage_subpool_put_pages(spool, chg); - goto out_err; + goto out_put_pages; }
/* @@ -4785,7 +4815,7 @@ int hugetlb_reserve_pages(struct inode *inode, * else has to be done for private mappings here */ if (!vma || vma->vm_flags & VM_MAYSHARE) { - add = region_add(resv_map, from, to, regions_needed); + add = region_add(h, h_cg, resv_map, from, to, regions_needed);
if (unlikely(add < 0)) { hugetlb_acct_memory(h, -gbl_reserve); @@ -4802,12 +4832,23 @@ int hugetlb_reserve_pages(struct inode *inode, */ long rsv_adjust;
- rsv_adjust = hugepage_subpool_put_pages(spool, - chg - add); + hugetlb_cgroup_uncharge_cgroup( + hstate_index(h), + (chg - add) * pages_per_huge_page(h), h_cg, + true); + + rsv_adjust = + hugepage_subpool_put_pages(spool, chg - add); hugetlb_acct_memory(h, -rsv_adjust); } } return 0; +out_put_pages: + /* put back original number of pages, chg */ + (void)hugepage_subpool_put_pages(spool, chg); +out_uncharge_cgroup: + hugetlb_cgroup_uncharge_cgroup( + hstate_index(h), chg * pages_per_huge_page(h), h_cg, true); out_err: if (!vma || vma->vm_flags & VM_MAYSHARE) /* Only call region_abort if the region_chg succeeded but the diff --git a/mm/hugetlb_cgroup.c b/mm/hugetlb_cgroup.c index e079513c8de0d..916ee24cc50d3 100644 --- a/mm/hugetlb_cgroup.c +++ b/mm/hugetlb_cgroup.c @@ -326,6 +326,21 @@ void hugetlb_cgroup_uncharge_counter(struct resv_map *resv, unsigned long start, css_put(resv->css); }
+void hugetlb_cgroup_uncharge_file_region(struct resv_map *resv, + struct file_region *rg, + unsigned long nr_pages) +{ + if (hugetlb_cgroup_disabled() || !resv || !rg || !nr_pages) + return; + + if (rg->reservation_counter && rg->pages_per_hpage && nr_pages > 0 && + !resv->reservation_counter) { + page_counter_uncharge(rg->reservation_counter, + nr_pages * rg->pages_per_hpage); + css_put(rg->css); + } +} + enum { RES_USAGE, RES_RSVD_USAGE, -- 2.25.0.341.g760bfbb309-goog
On 2/3/20 3:22 PM, Mina Almasry wrote:
For shared mappings, the pointer to the hugetlb_cgroup to uncharge lives in the resv_map entries, in file_region->reservation_counter.
After a call to region_chg, we charge the approprate hugetlb_cgroup, and if successful, we pass on the hugetlb_cgroup info to a follow up region_add call. When a file_region entry is added to the resv_map via region_add, we put the pointer to that cgroup in file_region->reservation_counter. If charging doesn't succeed, we report the error to the caller, so that the kernel fails the reservation.
On region_del, which is when the hugetlb memory is unreserved, we also uncharge
Lines of commit message wrap.
the file_region->reservation_counter.
Signed-off-by: Mina Almasry almasrymina@google.com
Changes in v11:
- Created new function, hugetlb_cgroup_uncharge_file_region to clean up
some #ifdefs.
- Moved file_region definition to hugetlb.h.
- Added copy_hugetlb_cgroup_uncharge_info function to clean up more
#ifdefs in the middle of hugetlb code.
Changes in v10:
- Deleted duplicated code snippet.
Changes in V9:
- Updated for hugetlb reservation repareting.
include/linux/hugetlb.h | 36 ++++++++ include/linux/hugetlb_cgroup.h | 10 +++ mm/hugetlb.c | 147 +++++++++++++++++++++------------ mm/hugetlb_cgroup.c | 15 ++++ 4 files changed, 155 insertions(+), 53 deletions(-)
diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h index 5491932ea5758..395f5b1fad416 100644 --- a/include/linux/hugetlb.h +++ b/include/linux/hugetlb.h @@ -57,6 +57,42 @@ struct resv_map { struct cgroup_subsys_state *css; #endif };
+/*
- Region tracking -- allows tracking of reservations and instantiated pages
across the pages in a mapping.
- The region data structures are embedded into a resv_map and protected
- by a resv_map's lock. The set of regions within the resv_map represent
- reservations for huge pages, or huge pages that have already been
- instantiated within the map. The from and to elements are huge page
- indicies into the associated mapping. from indicates the starting index
- of the region. to represents the first index past the end of the region.
- For example, a file region structure with from == 0 and to == 4 represents
- four huge pages in a mapping. It is important to note that the to element
- represents the first element past the end of the region. This is used in
- arithmetic as 4(to) - 0(from) = 4 huge pages in the region.
- Interval notation of the form [from, to) will be used to indicate that
- the endpoint from is inclusive and to is exclusive.
- */
+struct file_region {
- struct list_head link;
- long from;
- long to;
+#ifdef CONFIG_CGROUP_HUGETLB
- /*
* On shared mappings, each reserved region appears as a struct* file_region in resv_map. These fields hold the info needed to* uncharge each reservation.*/- struct page_counter *reservation_counter;
- unsigned long pages_per_hpage;
Can we get rid of pages_per_hpage here? It seems redundant as it will be the same for each file region. The same field/information is in the resv_map but only used for private mappings. Perhaps, we should use it for both shared and private?
- struct cgroup_subsys_state *css;
+#endif +};
extern struct resv_map *resv_map_alloc(void); void resv_map_release(struct kref *ref);
diff --git a/include/linux/hugetlb_cgroup.h b/include/linux/hugetlb_cgroup.h index 6a6c80df95ae3..c3fd417c268c5 100644 --- a/include/linux/hugetlb_cgroup.h +++ b/include/linux/hugetlb_cgroup.h @@ -102,11 +102,21 @@ extern void hugetlb_cgroup_uncharge_counter(struct resv_map *resv, unsigned long start, unsigned long end);
+extern void hugetlb_cgroup_uncharge_file_region(struct resv_map *resv,
struct file_region *rg,unsigned long nr_pages);extern void hugetlb_cgroup_file_init(void) __init; extern void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage);
#else
+static inline void hugetlb_cgroup_uncharge_file_region(struct resv_map *resv,
struct file_region *rg,unsigned long nr_pages)+{ +} static inline struct hugetlb_cgroup *hugetlb_cgroup_from_page(struct page *page, bool rsvd) { diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 986e9a9cc6fbe..33818ccaf7e89 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -220,31 +220,6 @@ static inline struct hugepage_subpool *subpool_vma(struct vm_area_struct *vma) return subpool_inode(file_inode(vma->vm_file)); }
-/*
- Region tracking -- allows tracking of reservations and instantiated pages
across the pages in a mapping.
- The region data structures are embedded into a resv_map and protected
- by a resv_map's lock. The set of regions within the resv_map represent
- reservations for huge pages, or huge pages that have already been
- instantiated within the map. The from and to elements are huge page
- indicies into the associated mapping. from indicates the starting index
- of the region. to represents the first index past the end of the region.
- For example, a file region structure with from == 0 and to == 4 represents
- four huge pages in a mapping. It is important to note that the to element
- represents the first element past the end of the region. This is used in
- arithmetic as 4(to) - 0(from) = 4 huge pages in the region.
- Interval notation of the form [from, to) will be used to indicate that
- the endpoint from is inclusive and to is exclusive.
- */
-struct file_region {
- struct list_head link;
- long from;
- long to;
-};
/* Helper that removes a struct file_region from the resv_map cache and returns
- it for use.
*/ @@ -266,6 +241,37 @@ get_file_region_entry_from_cache(struct resv_map *resv, long from, long to) return nrg; }
+static void copy_hugetlb_cgroup_uncharge_info(struct file_region *nrg,
struct file_region *rg)+{ +#ifdef CONFIG_CGROUP_HUGETLB
- nrg->reservation_counter = rg->reservation_counter;
- nrg->pages_per_hpage = rg->pages_per_hpage;
- nrg->css = rg->css;
- css_get(rg->css);
+#endif +}
+/* Helper that records hugetlb_cgroup uncharge info. */ +static void record_hugetlb_cgroup_uncharge_info(struct hugetlb_cgroup *h_cg,
struct file_region *nrg,struct hstate *h)
Not necessary, but I would make nrg first (or last) argument. It seems a bit odd that it is between two items that are most closely related.
+{ +#ifdef CONFIG_CGROUP_HUGETLB
- if (h_cg) {
nrg->reservation_counter =&h_cg->rsvd_hugepage[hstate_index(h)];nrg->pages_per_hpage = pages_per_huge_page(h);nrg->css = &h_cg->css;- } else {
nrg->reservation_counter = NULL;nrg->pages_per_hpage = 0;nrg->css = NULL;- }
+#endif +}
/* Must be called with resv->lock held. Calling this with count_only == true
- will count the number of pages to be added but will not modify the linked
- list. If regions_needed != NULL and count_only == true, then regions_needed
@@ -273,7 +279,9 @@ get_file_region_entry_from_cache(struct resv_map *resv, long from, long to)
- add the regions for this range.
*/ static long add_reservation_in_range(struct resv_map *resv, long f, long t,
long *regions_needed, bool count_only)
struct hugetlb_cgroup *h_cg,struct hstate *h, long *regions_needed,bool count_only)
It seems like count_only could be easily determined by value of other arguments. However, let's leave it as explicit just to make code easier to understand. Not necessary, but I wonder if something like: #define COUNT_ONLY true #define PERFORM_ADD false for arguments to this routine would make the code easier to read/understand.
{ long add = 0; struct list_head *head = &resv->regions; @@ -312,6 +320,8 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, if (!count_only) { nrg = get_file_region_entry_from_cache( resv, last_accounted_offset, rg->from);
record_hugetlb_cgroup_uncharge_info(h_cg, nrg,h); list_add(&nrg->link, rg->link.prev); } else if (regions_needed) *regions_needed += 1;@@ -328,11 +338,13 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, if (!count_only) { nrg = get_file_region_entry_from_cache( resv, last_accounted_offset, t);
record_hugetlb_cgroup_uncharge_info(h_cg, nrg, h); list_add(&nrg->link, rg->link.prev);} else if (regions_needed) *regions_needed += 1; }
VM_BUG_ON(add < 0);
Curious why that was added. The computation of 'add' did not change with these changes.
return add; }
@@ -353,7 +365,8 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t,
- fail; region_chg will always allocate at least 1 entry and a region_add for
- 1 page will only require at most 1 entry.
*/ -static long region_add(struct resv_map *resv, long f, long t, +static long region_add(struct hstate *h, struct hugetlb_cgroup *h_cg,
struct resv_map *resv, long f, long t,
I would prefer keeping "struct resv_map *resv, long f, long t" as the first arguments to this routine. To me that makes most sense as that is the primary purpose of the operation (to add regions to the reserve map to cover the range f -> t).
long in_regions_needed){ long add = 0, actual_regions_needed = 0, i = 0; @@ -366,7 +379,8 @@ static long region_add(struct resv_map *resv, long f, long t, retry:
/* Count how many regions are actually needed to execute this add. */
- add_reservation_in_range(resv, f, t, &actual_regions_needed, true);
add_reservation_in_range(resv, f, t, NULL, NULL, &actual_regions_needed,
true);/*
- Check for sufficient descriptors in the cache to accommodate
@@ -404,7 +418,7 @@ static long region_add(struct resv_map *resv, long f, long t, goto retry; }
- add = add_reservation_in_range(resv, f, t, NULL, false);
add = add_reservation_in_range(resv, f, t, h_cg, h, NULL, false);
resv->adds_in_progress -= in_regions_needed;
@@ -452,7 +466,8 @@ static long region_chg(struct resv_map *resv, long f, long t, spin_lock(&resv->lock);
/* Count how many hugepages in this range are NOT respresented. */
- chg = add_reservation_in_range(resv, f, t, out_regions_needed, true);
chg = add_reservation_in_range(resv, f, t, NULL, NULL,
out_regions_needed, true);if (*out_regions_needed == 0) *out_regions_needed = 1;
@@ -588,11 +603,17 @@ static long region_del(struct resv_map *resv, long f, long t) /* New entry for end of split region */ nrg->from = t; nrg->to = rg->to;
copy_hugetlb_cgroup_uncharge_info(nrg, rg);INIT_LIST_HEAD(&nrg->link); /* Original entry is trimmed */ rg->to = f;hugetlb_cgroup_uncharge_file_region(resv, rg, nrg->to - nrg->from);list_add(&nrg->link, &rg->link); nrg = NULL; break;@@ -600,6 +621,8 @@ static long region_del(struct resv_map *resv, long f, long t)
if (f <= rg->from && t >= rg->to) { /* Remove entire region */ del += rg->to - rg->from;
hugetlb_cgroup_uncharge_file_region(resv, rg,rg->to - rg->from); list_del(&rg->link); kfree(rg); continue;@@ -608,14 +631,21 @@ static long region_del(struct resv_map *resv, long f, long t) if (f <= rg->from) { /* Trim beginning of region */ del += t - rg->from; rg->from = t;
hugetlb_cgroup_uncharge_file_region(resv, rg,t - rg->from);} else { /* Trim end of region */ del += rg->to - f; rg->to = f;
hugetlb_cgroup_uncharge_file_region(resv, rg,rg->to - f);} }
spin_unlock(&resv->lock); kfree(nrg);
return del;
}
@@ -2017,7 +2047,7 @@ static long __vma_reservation_common(struct hstate *h, VM_BUG_ON(dummy_out_regions_needed != 1); break; case VMA_COMMIT_RESV:
ret = region_add(resv, idx, idx + 1, 1);
ret = region_add(NULL, NULL, resv, idx, idx + 1, 1);@@ -2027,7 +2057,7 @@ static long __vma_reservation_common(struct hstate *h, break; case VMA_ADD_RESV: if (vma->vm_flags & VM_MAYSHARE) {
ret = region_add(resv, idx, idx + 1, 1);
ret = region_add(NULL, NULL, resv, idx, idx + 1, 1); /* region_add calls of range 1 should never fail. */ VM_BUG_ON(ret < 0);@@ -4688,7 +4718,7 @@ int hugetlb_reserve_pages(struct inode *inode, struct hstate *h = hstate_inode(inode); struct hugepage_subpool *spool = subpool_inode(inode); struct resv_map *resv_map;
- struct hugetlb_cgroup *h_cg;
struct hugetlb_cgroup *h_cg = NULL; long gbl_reserve, regions_needed = 0;
/* This should never happen */
@@ -4729,19 +4759,6 @@ int hugetlb_reserve_pages(struct inode *inode,
chg = to - from;
if (hugetlb_cgroup_charge_cgroup(hstate_index(h),chg * pages_per_huge_page(h),&h_cg, true)) {kref_put(&resv_map->refs, resv_map_release);return -ENOMEM;}/** Since this branch handles private mappings, we attach the* counter to uncharge for this reservation off resv_map.*/resv_map_set_hugetlb_cgroup_uncharge_info(resv_map, h_cg, h);- set_vma_resv_map(vma, resv_map); set_vma_resv_flags(vma, HPAGE_RESV_OWNER); }
@@ -4751,6 +4768,21 @@ int hugetlb_reserve_pages(struct inode *inode, goto out_err; }
- ret = hugetlb_cgroup_charge_cgroup(
hstate_index(h), chg * pages_per_huge_page(h), &h_cg, true);- if (ret < 0) {
ret = -ENOMEM;goto out_err;- }
- if (vma && !(vma->vm_flags & VM_MAYSHARE) && h_cg) {
/* For private mappings, the hugetlb_cgroup uncharge info hangs* of the resv_map.*/resv_map_set_hugetlb_cgroup_uncharge_info(resv_map, h_cg, h);- }
- /*
- There must be enough pages in the subpool for the mapping. If
- the subpool has a minimum size, there may be some global
@@ -4759,7 +4791,7 @@ int hugetlb_reserve_pages(struct inode *inode, gbl_reserve = hugepage_subpool_get_pages(spool, chg); if (gbl_reserve < 0) { ret = -ENOSPC;
goto out_err;
goto out_uncharge_cgroup;}
/*
@@ -4768,9 +4800,7 @@ int hugetlb_reserve_pages(struct inode *inode, */ ret = hugetlb_acct_memory(h, gbl_reserve); if (ret < 0) {
/* put back original number of pages, chg */(void)hugepage_subpool_put_pages(spool, chg);goto out_err;
goto out_put_pages;}
/*
@@ -4785,7 +4815,7 @@ int hugetlb_reserve_pages(struct inode *inode, * else has to be done for private mappings here */ if (!vma || vma->vm_flags & VM_MAYSHARE) {
add = region_add(resv_map, from, to, regions_needed);
add = region_add(h, h_cg, resv_map, from, to, regions_needed);if (unlikely(add < 0)) { hugetlb_acct_memory(h, -gbl_reserve);
Don't we need to call hugetlb_cgroup_uncharge_cgroup() in the case as well? Also, can you "goto out_put_pages" to avoid hugepage_subpool_put_pages() call?
@@ -4802,12 +4832,23 @@ int hugetlb_reserve_pages(struct inode *inode, */ long rsv_adjust;
rsv_adjust = hugepage_subpool_put_pages(spool,chg - add);
hugetlb_cgroup_uncharge_cgroup(hstate_index(h),(chg - add) * pages_per_huge_page(h), h_cg,true);rsv_adjust =hugepage_subpool_put_pages(spool, chg - add); hugetlb_acct_memory(h, -rsv_adjust);+out_put_pages:
- /* put back original number of pages, chg */
- (void)hugepage_subpool_put_pages(spool, chg);
+out_uncharge_cgroup:
- hugetlb_cgroup_uncharge_cgroup(
hstate_index(h), chg * pages_per_huge_page(h), h_cg, true);out_err: if (!vma || vma->vm_flags & VM_MAYSHARE) /* Only call region_abort if the region_chg succeeded but the diff --git a/mm/hugetlb_cgroup.c b/mm/hugetlb_cgroup.c index e079513c8de0d..916ee24cc50d3 100644 --- a/mm/hugetlb_cgroup.c +++ b/mm/hugetlb_cgroup.c @@ -326,6 +326,21 @@ void hugetlb_cgroup_uncharge_counter(struct resv_map *resv, unsigned long start, css_put(resv->css); }
+void hugetlb_cgroup_uncharge_file_region(struct resv_map *resv,
struct file_region *rg,unsigned long nr_pages)+{
- if (hugetlb_cgroup_disabled() || !resv || !rg || !nr_pages)
return;- if (rg->reservation_counter && rg->pages_per_hpage && nr_pages > 0 &&
!resv->reservation_counter) {page_counter_uncharge(rg->reservation_counter,nr_pages * rg->pages_per_hpage);css_put(rg->css);- }
+}
enum { RES_USAGE, RES_RSVD_USAGE, -- 2.25.0.341.g760bfbb309-goog
On Thu, Feb 6, 2020 at 11:34 AM Mike Kravetz mike.kravetz@oracle.com wrote:
On 2/3/20 3:22 PM, Mina Almasry wrote:
For shared mappings, the pointer to the hugetlb_cgroup to uncharge lives in the resv_map entries, in file_region->reservation_counter.
After a call to region_chg, we charge the approprate hugetlb_cgroup, and if successful, we pass on the hugetlb_cgroup info to a follow up region_add call. When a file_region entry is added to the resv_map via region_add, we put the pointer to that cgroup in file_region->reservation_counter. If charging doesn't succeed, we report the error to the caller, so that the kernel fails the reservation.
On region_del, which is when the hugetlb memory is unreserved, we also uncharge
Lines of commit message wrap.
Will fix.
the file_region->reservation_counter.
Signed-off-by: Mina Almasry almasrymina@google.com
Changes in v11:
- Created new function, hugetlb_cgroup_uncharge_file_region to clean up
some #ifdefs.
- Moved file_region definition to hugetlb.h.
- Added copy_hugetlb_cgroup_uncharge_info function to clean up more
#ifdefs in the middle of hugetlb code.
Changes in v10:
- Deleted duplicated code snippet.
Changes in V9:
- Updated for hugetlb reservation repareting.
include/linux/hugetlb.h | 36 ++++++++ include/linux/hugetlb_cgroup.h | 10 +++ mm/hugetlb.c | 147 +++++++++++++++++++++------------ mm/hugetlb_cgroup.c | 15 ++++ 4 files changed, 155 insertions(+), 53 deletions(-)
diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h index 5491932ea5758..395f5b1fad416 100644 --- a/include/linux/hugetlb.h +++ b/include/linux/hugetlb.h @@ -57,6 +57,42 @@ struct resv_map { struct cgroup_subsys_state *css; #endif };
+/*
- Region tracking -- allows tracking of reservations and instantiated pages
across the pages in a mapping.
- The region data structures are embedded into a resv_map and protected
- by a resv_map's lock. The set of regions within the resv_map represent
- reservations for huge pages, or huge pages that have already been
- instantiated within the map. The from and to elements are huge page
- indicies into the associated mapping. from indicates the starting index
- of the region. to represents the first index past the end of the region.
- For example, a file region structure with from == 0 and to == 4 represents
- four huge pages in a mapping. It is important to note that the to element
- represents the first element past the end of the region. This is used in
- arithmetic as 4(to) - 0(from) = 4 huge pages in the region.
- Interval notation of the form [from, to) will be used to indicate that
- the endpoint from is inclusive and to is exclusive.
- */
+struct file_region {
struct list_head link;long from;long to;+#ifdef CONFIG_CGROUP_HUGETLB
/** On shared mappings, each reserved region appears as a struct* file_region in resv_map. These fields hold the info needed to* uncharge each reservation.*/struct page_counter *reservation_counter;unsigned long pages_per_hpage;Can we get rid of pages_per_hpage here? It seems redundant as it will be the same for each file region. The same field/information is in the resv_map but only used for private mappings. Perhaps, we should use it for both shared and private?
Will do.
struct cgroup_subsys_state *css;+#endif +};
extern struct resv_map *resv_map_alloc(void); void resv_map_release(struct kref *ref);
diff --git a/include/linux/hugetlb_cgroup.h b/include/linux/hugetlb_cgroup.h index 6a6c80df95ae3..c3fd417c268c5 100644 --- a/include/linux/hugetlb_cgroup.h +++ b/include/linux/hugetlb_cgroup.h @@ -102,11 +102,21 @@ extern void hugetlb_cgroup_uncharge_counter(struct resv_map *resv, unsigned long start, unsigned long end);
+extern void hugetlb_cgroup_uncharge_file_region(struct resv_map *resv,
struct file_region *rg,unsigned long nr_pages);extern void hugetlb_cgroup_file_init(void) __init; extern void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage);
#else
+static inline void hugetlb_cgroup_uncharge_file_region(struct resv_map *resv,
struct file_region *rg,unsigned long nr_pages)+{ +} static inline struct hugetlb_cgroup *hugetlb_cgroup_from_page(struct page *page, bool rsvd) { diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 986e9a9cc6fbe..33818ccaf7e89 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -220,31 +220,6 @@ static inline struct hugepage_subpool *subpool_vma(struct vm_area_struct *vma) return subpool_inode(file_inode(vma->vm_file)); }
-/*
- Region tracking -- allows tracking of reservations and instantiated pages
across the pages in a mapping.
- The region data structures are embedded into a resv_map and protected
- by a resv_map's lock. The set of regions within the resv_map represent
- reservations for huge pages, or huge pages that have already been
- instantiated within the map. The from and to elements are huge page
- indicies into the associated mapping. from indicates the starting index
- of the region. to represents the first index past the end of the region.
- For example, a file region structure with from == 0 and to == 4 represents
- four huge pages in a mapping. It is important to note that the to element
- represents the first element past the end of the region. This is used in
- arithmetic as 4(to) - 0(from) = 4 huge pages in the region.
- Interval notation of the form [from, to) will be used to indicate that
- the endpoint from is inclusive and to is exclusive.
- */
-struct file_region {
struct list_head link;long from;long to;-};
/* Helper that removes a struct file_region from the resv_map cache and returns
- it for use.
*/ @@ -266,6 +241,37 @@ get_file_region_entry_from_cache(struct resv_map *resv, long from, long to) return nrg; }
+static void copy_hugetlb_cgroup_uncharge_info(struct file_region *nrg,
struct file_region *rg)+{ +#ifdef CONFIG_CGROUP_HUGETLB
nrg->reservation_counter = rg->reservation_counter;nrg->pages_per_hpage = rg->pages_per_hpage;nrg->css = rg->css;css_get(rg->css);+#endif +}
+/* Helper that records hugetlb_cgroup uncharge info. */ +static void record_hugetlb_cgroup_uncharge_info(struct hugetlb_cgroup *h_cg,
struct file_region *nrg,struct hstate *h)Not necessary, but I would make nrg first (or last) argument. It seems a bit odd that it is between two items that are most closely related.
Makes sense, will do.
+{ +#ifdef CONFIG_CGROUP_HUGETLB
if (h_cg) {nrg->reservation_counter =&h_cg->rsvd_hugepage[hstate_index(h)];nrg->pages_per_hpage = pages_per_huge_page(h);nrg->css = &h_cg->css;} else {nrg->reservation_counter = NULL;nrg->pages_per_hpage = 0;nrg->css = NULL;}+#endif +}
/* Must be called with resv->lock held. Calling this with count_only == true
- will count the number of pages to be added but will not modify the linked
- list. If regions_needed != NULL and count_only == true, then regions_needed
@@ -273,7 +279,9 @@ get_file_region_entry_from_cache(struct resv_map *resv, long from, long to)
- add the regions for this range.
*/ static long add_reservation_in_range(struct resv_map *resv, long f, long t,
long *regions_needed, bool count_only)
struct hugetlb_cgroup *h_cg,struct hstate *h, long *regions_needed,bool count_only)It seems like count_only could be easily determined by value of other arguments. However, let's leave it as explicit just to make code easier to understand. Not necessary, but I wonder if something like: #define COUNT_ONLY true #define PERFORM_ADD false for arguments to this routine would make the code easier to read/understand.
I agree it's better for readability. If you're leaving it to my preference I'd rather not have the macros.
{ long add = 0; struct list_head *head = &resv->regions; @@ -312,6 +320,8 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, if (!count_only) { nrg = get_file_region_entry_from_cache( resv, last_accounted_offset, rg->from);
record_hugetlb_cgroup_uncharge_info(h_cg, nrg,h); list_add(&nrg->link, rg->link.prev); } else if (regions_needed) *regions_needed += 1;@@ -328,11 +338,13 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, if (!count_only) { nrg = get_file_region_entry_from_cache( resv, last_accounted_offset, t);
record_hugetlb_cgroup_uncharge_info(h_cg, nrg, h); list_add(&nrg->link, rg->link.prev); } else if (regions_needed) *regions_needed += 1; }VM_BUG_ON(add < 0);Curious why that was added. The computation of 'add' did not change with these changes.
This belongs better in patch 4/9 'hugetlb: disable region_add file_region coalescing', where the add is actually modified. Having it here is a mistake spitting up the changes into patches. Would you rather I just remove it or move it to patch 4? More importantly, does moving it to patch 4 without any other changes trigger you to re-review of patch 4? I'm currently carrying over your Reviewed-By tag since no changes have been made to that patch. That was one of the harder patches to review and re-reviewing it is not worth the effort for this line I would say. This VM_BUG_ON is really only a sanity check for something going very wrong in the math.
return add;}
@@ -353,7 +365,8 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t,
- fail; region_chg will always allocate at least 1 entry and a region_add for
- 1 page will only require at most 1 entry.
*/ -static long region_add(struct resv_map *resv, long f, long t, +static long region_add(struct hstate *h, struct hugetlb_cgroup *h_cg,
struct resv_map *resv, long f, long t,I would prefer keeping "struct resv_map *resv, long f, long t" as the first arguments to this routine. To me that makes most sense as that is the primary purpose of the operation (to add regions to the reserve map to cover the range f -> t).
Will do.
long in_regions_needed){ long add = 0, actual_regions_needed = 0, i = 0; @@ -366,7 +379,8 @@ static long region_add(struct resv_map *resv, long f, long t, retry:
/* Count how many regions are actually needed to execute this add. */
add_reservation_in_range(resv, f, t, &actual_regions_needed, true);
add_reservation_in_range(resv, f, t, NULL, NULL, &actual_regions_needed,true); /* * Check for sufficient descriptors in the cache to accommodate@@ -404,7 +418,7 @@ static long region_add(struct resv_map *resv, long f, long t, goto retry; }
add = add_reservation_in_range(resv, f, t, NULL, false);
add = add_reservation_in_range(resv, f, t, h_cg, h, NULL, false); resv->adds_in_progress -= in_regions_needed;@@ -452,7 +466,8 @@ static long region_chg(struct resv_map *resv, long f, long t, spin_lock(&resv->lock);
/* Count how many hugepages in this range are NOT respresented. */
chg = add_reservation_in_range(resv, f, t, out_regions_needed, true);
chg = add_reservation_in_range(resv, f, t, NULL, NULL,out_regions_needed, true); if (*out_regions_needed == 0) *out_regions_needed = 1;@@ -588,11 +603,17 @@ static long region_del(struct resv_map *resv, long f, long t) /* New entry for end of split region */ nrg->from = t; nrg->to = rg->to;
copy_hugetlb_cgroup_uncharge_info(nrg, rg);INIT_LIST_HEAD(&nrg->link); /* Original entry is trimmed */ rg->to = f;hugetlb_cgroup_uncharge_file_region(resv, rg, nrg->to - nrg->from);list_add(&nrg->link, &rg->link); nrg = NULL; break;@@ -600,6 +621,8 @@ static long region_del(struct resv_map *resv, long f, long t)
if (f <= rg->from && t >= rg->to) { /* Remove entire region */ del += rg->to - rg->from;
hugetlb_cgroup_uncharge_file_region(resv, rg,rg->to - rg->from); list_del(&rg->link); kfree(rg); continue;@@ -608,14 +631,21 @@ static long region_del(struct resv_map *resv, long f, long t) if (f <= rg->from) { /* Trim beginning of region */ del += t - rg->from; rg->from = t;
hugetlb_cgroup_uncharge_file_region(resv, rg,t - rg->from); } else { /* Trim end of region */ del += rg->to - f; rg->to = f;hugetlb_cgroup_uncharge_file_region(resv, rg,rg->to - f); } } spin_unlock(&resv->lock); kfree(nrg);return del;}
@@ -2017,7 +2047,7 @@ static long __vma_reservation_common(struct hstate *h, VM_BUG_ON(dummy_out_regions_needed != 1); break; case VMA_COMMIT_RESV:
ret = region_add(resv, idx, idx + 1, 1);
ret = region_add(NULL, NULL, resv, idx, idx + 1, 1); /* region_add calls of range 1 should never fail. */ VM_BUG_ON(ret < 0); break;@@ -2027,7 +2057,7 @@ static long __vma_reservation_common(struct hstate *h, break; case VMA_ADD_RESV: if (vma->vm_flags & VM_MAYSHARE) {
ret = region_add(resv, idx, idx + 1, 1);
ret = region_add(NULL, NULL, resv, idx, idx + 1, 1); /* region_add calls of range 1 should never fail. */ VM_BUG_ON(ret < 0); } else {@@ -4688,7 +4718,7 @@ int hugetlb_reserve_pages(struct inode *inode, struct hstate *h = hstate_inode(inode); struct hugepage_subpool *spool = subpool_inode(inode); struct resv_map *resv_map;
struct hugetlb_cgroup *h_cg;
struct hugetlb_cgroup *h_cg = NULL; long gbl_reserve, regions_needed = 0; /* This should never happen */@@ -4729,19 +4759,6 @@ int hugetlb_reserve_pages(struct inode *inode,
chg = to - from;
if (hugetlb_cgroup_charge_cgroup(hstate_index(h),chg * pages_per_huge_page(h),&h_cg, true)) {kref_put(&resv_map->refs, resv_map_release);return -ENOMEM;}/** Since this branch handles private mappings, we attach the* counter to uncharge for this reservation off resv_map.*/resv_map_set_hugetlb_cgroup_uncharge_info(resv_map, h_cg, h);set_vma_resv_map(vma, resv_map); set_vma_resv_flags(vma, HPAGE_RESV_OWNER); }@@ -4751,6 +4768,21 @@ int hugetlb_reserve_pages(struct inode *inode, goto out_err; }
ret = hugetlb_cgroup_charge_cgroup(hstate_index(h), chg * pages_per_huge_page(h), &h_cg, true);if (ret < 0) {ret = -ENOMEM;goto out_err;}if (vma && !(vma->vm_flags & VM_MAYSHARE) && h_cg) {/* For private mappings, the hugetlb_cgroup uncharge info hangs* of the resv_map.*/resv_map_set_hugetlb_cgroup_uncharge_info(resv_map, h_cg, h);}/* * There must be enough pages in the subpool for the mapping. If * the subpool has a minimum size, there may be some global@@ -4759,7 +4791,7 @@ int hugetlb_reserve_pages(struct inode *inode, gbl_reserve = hugepage_subpool_get_pages(spool, chg); if (gbl_reserve < 0) { ret = -ENOSPC;
goto out_err;
goto out_uncharge_cgroup; } /*@@ -4768,9 +4800,7 @@ int hugetlb_reserve_pages(struct inode *inode, */ ret = hugetlb_acct_memory(h, gbl_reserve); if (ret < 0) {
/* put back original number of pages, chg */(void)hugepage_subpool_put_pages(spool, chg);goto out_err;
goto out_put_pages; } /*@@ -4785,7 +4815,7 @@ int hugetlb_reserve_pages(struct inode *inode, * else has to be done for private mappings here */ if (!vma || vma->vm_flags & VM_MAYSHARE) {
add = region_add(resv_map, from, to, regions_needed);
add = region_add(h, h_cg, resv_map, from, to, regions_needed); if (unlikely(add < 0)) { hugetlb_acct_memory(h, -gbl_reserve);Don't we need to call hugetlb_cgroup_uncharge_cgroup() in the case as well? Also, can you "goto out_put_pages" to avoid hugepage_subpool_put_pages() call?
Yes seems that's the case. Just a 'goto out_put_pages;' right here seems right.
@@ -4802,12 +4832,23 @@ int hugetlb_reserve_pages(struct inode *inode, */ long rsv_adjust;
rsv_adjust = hugepage_subpool_put_pages(spool,chg - add);
hugetlb_cgroup_uncharge_cgroup(hstate_index(h),(chg - add) * pages_per_huge_page(h), h_cg,true);rsv_adjust =hugepage_subpool_put_pages(spool, chg - add); hugetlb_acct_memory(h, -rsv_adjust); } } return 0;+out_put_pages:
/* put back original number of pages, chg */(void)hugepage_subpool_put_pages(spool, chg);+out_uncharge_cgroup:
hugetlb_cgroup_uncharge_cgroup(hstate_index(h), chg * pages_per_huge_page(h), h_cg, true);out_err: if (!vma || vma->vm_flags & VM_MAYSHARE) /* Only call region_abort if the region_chg succeeded but the diff --git a/mm/hugetlb_cgroup.c b/mm/hugetlb_cgroup.c index e079513c8de0d..916ee24cc50d3 100644 --- a/mm/hugetlb_cgroup.c +++ b/mm/hugetlb_cgroup.c @@ -326,6 +326,21 @@ void hugetlb_cgroup_uncharge_counter(struct resv_map *resv, unsigned long start, css_put(resv->css); }
+void hugetlb_cgroup_uncharge_file_region(struct resv_map *resv,
struct file_region *rg,unsigned long nr_pages)+{
if (hugetlb_cgroup_disabled() || !resv || !rg || !nr_pages)return;if (rg->reservation_counter && rg->pages_per_hpage && nr_pages > 0 &&!resv->reservation_counter) {page_counter_uncharge(rg->reservation_counter,nr_pages * rg->pages_per_hpage);css_put(rg->css);}+}
enum { RES_USAGE, RES_RSVD_USAGE, -- 2.25.0.341.g760bfbb309-goog
-- Mike Kravetz
Support MAP_NORESERVE accounting as part of the new counter.
For each hugepage allocation, at allocation time we check if there is a reservation for this allocation or not. If there is a reservation for this allocation, then this allocation was charged at reservation time, and we don't re-account it. If there is no reserevation for this allocation, we charge the appropriate hugetlb_cgroup.
The hugetlb_cgroup to uncharge for this allocation is stored in page[3].private. We use new APIs added in an earlier patch to set this pointer.
Signed-off-by: Mina Almasry almasrymina@google.com
---
Changes in v10: - Refactored deferred_reserve check.
--- mm/hugetlb.c | 28 +++++++++++++++++++++++++++- 1 file changed, 27 insertions(+), 1 deletion(-)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 33818ccaf7e89..ec0b55ea1506e 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -1339,6 +1339,9 @@ static void __free_huge_page(struct page *page) clear_page_huge_active(page); hugetlb_cgroup_uncharge_page(hstate_index(h), pages_per_huge_page(h), page, false); + hugetlb_cgroup_uncharge_page(hstate_index(h), pages_per_huge_page(h), + page, true); + if (restore_reserve) h->resv_huge_pages++;
@@ -2172,6 +2175,7 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, long gbl_chg; int ret, idx; struct hugetlb_cgroup *h_cg; + bool deferred_reserve;
idx = hstate_index(h); /* @@ -2209,10 +2213,20 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, gbl_chg = 1; }
+ /* If this allocation is not consuming a reservation, charge it now. + */ + deferred_reserve = map_chg || avoid_reserve || !vma_resv_map(vma); + if (deferred_reserve) { + ret = hugetlb_cgroup_charge_cgroup(idx, pages_per_huge_page(h), + &h_cg, true); + if (ret) + goto out_subpool_put; + } + ret = hugetlb_cgroup_charge_cgroup(idx, pages_per_huge_page(h), &h_cg, false); if (ret) - goto out_subpool_put; + goto out_uncharge_cgroup_reservation;
spin_lock(&hugetlb_lock); /* @@ -2236,6 +2250,14 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, } hugetlb_cgroup_commit_charge(idx, pages_per_huge_page(h), h_cg, page, false); + /* If allocation is not consuming a reservation, also store the + * hugetlb_cgroup pointer on the page. + */ + if (deferred_reserve) { + hugetlb_cgroup_commit_charge(idx, pages_per_huge_page(h), h_cg, + page, true); + } + spin_unlock(&hugetlb_lock);
set_page_private(page, (unsigned long)spool); @@ -2261,6 +2283,10 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, out_uncharge_cgroup: hugetlb_cgroup_uncharge_cgroup(idx, pages_per_huge_page(h), h_cg, false); +out_uncharge_cgroup_reservation: + if (deferred_reserve) + hugetlb_cgroup_uncharge_cgroup(idx, pages_per_huge_page(h), + h_cg, true); out_subpool_put: if (map_chg || avoid_reserve) hugepage_subpool_put_pages(spool, 1); -- 2.25.0.341.g760bfbb309-goog
On 2/3/20 3:22 PM, Mina Almasry wrote:
Support MAP_NORESERVE accounting as part of the new counter.
For each hugepage allocation, at allocation time we check if there is a reservation for this allocation or not. If there is a reservation for this allocation, then this allocation was charged at reservation time, and we don't re-account it. If there is no reserevation for this allocation, we charge the appropriate hugetlb_cgroup.
The hugetlb_cgroup to uncharge for this allocation is stored in page[3].private. We use new APIs added in an earlier patch to set this pointer.
Ah! That reminded me to look at the migration code. Turns out that none of the existing cgroup information (page[2]) is being migrated today. That is a bug. :( I'll confirm and fix in a patch separate from this series. We will need to make sure that new information added by this series in page[3] is also migrated. That would be in an earlier patch where the use of the field is introduced.
Signed-off-by: Mina Almasry almasrymina@google.com
Changes in v10:
- Refactored deferred_reserve check.
mm/hugetlb.c | 28 +++++++++++++++++++++++++++- 1 file changed, 27 insertions(+), 1 deletion(-)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 33818ccaf7e89..ec0b55ea1506e 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -1339,6 +1339,9 @@ static void __free_huge_page(struct page *page) clear_page_huge_active(page); hugetlb_cgroup_uncharge_page(hstate_index(h), pages_per_huge_page(h), page, false);
- hugetlb_cgroup_uncharge_page(hstate_index(h), pages_per_huge_page(h),
page, true);
When looking at the code without change markings, the two above lines look so similar my first thought is there must be a mistake.
A suggestion for better code readability: - hugetlb_cgroup_uncharge_page could just take "struct hstate *h" and get both hstate_index(h) and pages_per_huge_page(h). - Perhaps make hugetlb_cgroup_uncharge_page and hugetlb_cgroup_uncharge_page_rsvd be wrappers around a common routine. Then the above would look like:
hugetlb_cgroup_uncharge_page(h, page); hugetlb_cgroup_uncharge_page_rsvd(h, page);
if (restore_reserve) h->resv_huge_pages++;
@@ -2172,6 +2175,7 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, long gbl_chg; int ret, idx; struct hugetlb_cgroup *h_cg;
bool deferred_reserve;
idx = hstate_index(h); /*
@@ -2209,10 +2213,20 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, gbl_chg = 1; }
- /* If this allocation is not consuming a reservation, charge it now.
*/- deferred_reserve = map_chg || avoid_reserve || !vma_resv_map(vma);
- if (deferred_reserve) {
ret = hugetlb_cgroup_charge_cgroup(idx, pages_per_huge_page(h),&h_cg, true);if (ret)goto out_subpool_put;- }
- ret = hugetlb_cgroup_charge_cgroup(idx, pages_per_huge_page(h), &h_cg, false);
Hmmm? I'm starting to like the wrapper idea more as a way to help with readability of the bool rsvd argument.
hugetlb_cgroup_charge_cgroup_rsvd() hugetlb_cgroup_charge_cgroup()
At least to me it makes it easier to read.
On 2/6/20 2:31 PM, Mike Kravetz wrote:
On 2/3/20 3:22 PM, Mina Almasry wrote:
Support MAP_NORESERVE accounting as part of the new counter.
For each hugepage allocation, at allocation time we check if there is a reservation for this allocation or not. If there is a reservation for this allocation, then this allocation was charged at reservation time, and we don't re-account it. If there is no reserevation for this allocation, we charge the appropriate hugetlb_cgroup.
The hugetlb_cgroup to uncharge for this allocation is stored in page[3].private. We use new APIs added in an earlier patch to set this pointer.
Ah! That reminded me to look at the migration code. Turns out that none of the existing cgroup information (page[2]) is being migrated today. That is a bug. :( I'll confirm and fix in a patch separate from this series. We will need to make sure that new information added by this series in page[3] is also migrated. That would be in an earlier patch where the use of the field is introduced.
My appologies!
cgroup information is migrated and you took care of it for new reservation information in patch 2. Please disregard that statement.
On Thu, Feb 6, 2020 at 2:31 PM Mike Kravetz mike.kravetz@oracle.com wrote:
On 2/3/20 3:22 PM, Mina Almasry wrote:
Support MAP_NORESERVE accounting as part of the new counter.
For each hugepage allocation, at allocation time we check if there is a reservation for this allocation or not. If there is a reservation for this allocation, then this allocation was charged at reservation time, and we don't re-account it. If there is no reserevation for this allocation, we charge the appropriate hugetlb_cgroup.
The hugetlb_cgroup to uncharge for this allocation is stored in page[3].private. We use new APIs added in an earlier patch to set this pointer.
Ah! That reminded me to look at the migration code. Turns out that none of the existing cgroup information (page[2]) is being migrated today. That is a bug. :( I'll confirm and fix in a patch separate from this series. We will need to make sure that new information added by this series in page[3] is also migrated. That would be in an earlier patch where the use of the field is introduced.
Signed-off-by: Mina Almasry almasrymina@google.com
Changes in v10:
- Refactored deferred_reserve check.
mm/hugetlb.c | 28 +++++++++++++++++++++++++++- 1 file changed, 27 insertions(+), 1 deletion(-)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 33818ccaf7e89..ec0b55ea1506e 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -1339,6 +1339,9 @@ static void __free_huge_page(struct page *page) clear_page_huge_active(page); hugetlb_cgroup_uncharge_page(hstate_index(h), pages_per_huge_page(h), page, false);
hugetlb_cgroup_uncharge_page(hstate_index(h), pages_per_huge_page(h),page, true);When looking at the code without change markings, the two above lines look so similar my first thought is there must be a mistake.
A suggestion for better code readability:
hugetlb_cgroup_uncharge_page could just take "struct hstate *h" and get both hstate_index(h) and pages_per_huge_page(h).
Perhaps make hugetlb_cgroup_uncharge_page and hugetlb_cgroup_uncharge_page_rsvd be wrappers around a common routine. Then the above would look like:
hugetlb_cgroup_uncharge_page(h, page); hugetlb_cgroup_uncharge_page_rsvd(h, page);
I did modify the interfaces to this, as it's much better for readability indeed. Unfortunately the patch the adds interfaces probably needs a re-review now as it's changed quite a bit, I did not carry your or David's Reviewed-by.
if (restore_reserve) h->resv_huge_pages++;@@ -2172,6 +2175,7 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, long gbl_chg; int ret, idx; struct hugetlb_cgroup *h_cg;
bool deferred_reserve; idx = hstate_index(h); /*@@ -2209,10 +2213,20 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, gbl_chg = 1; }
/* If this allocation is not consuming a reservation, charge it now.*/deferred_reserve = map_chg || avoid_reserve || !vma_resv_map(vma);if (deferred_reserve) {ret = hugetlb_cgroup_charge_cgroup(idx, pages_per_huge_page(h),&h_cg, true);if (ret)goto out_subpool_put;}ret = hugetlb_cgroup_charge_cgroup(idx, pages_per_huge_page(h), &h_cg, false);Hmmm? I'm starting to like the wrapper idea more as a way to help with readability of the bool rsvd argument.
hugetlb_cgroup_charge_cgroup_rsvd() hugetlb_cgroup_charge_cgroup()
At least to me it makes it easier to read.
Mike Kravetz
if (ret)
goto out_subpool_put;
goto out_uncharge_cgroup_reservation; spin_lock(&hugetlb_lock); /*@@ -2236,6 +2250,14 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, } hugetlb_cgroup_commit_charge(idx, pages_per_huge_page(h), h_cg, page, false);
/* If allocation is not consuming a reservation, also store the* hugetlb_cgroup pointer on the page.*/if (deferred_reserve) {hugetlb_cgroup_commit_charge(idx, pages_per_huge_page(h), h_cg,page, true);}spin_unlock(&hugetlb_lock); set_page_private(page, (unsigned long)spool);@@ -2261,6 +2283,10 @@ struct page *alloc_huge_page(struct vm_area_struct *vma, out_uncharge_cgroup: hugetlb_cgroup_uncharge_cgroup(idx, pages_per_huge_page(h), h_cg, false); +out_uncharge_cgroup_reservation:
if (deferred_reserve)hugetlb_cgroup_uncharge_cgroup(idx, pages_per_huge_page(h),h_cg, true);out_subpool_put: if (map_chg || avoid_reserve) hugepage_subpool_put_pages(spool, 1); -- 2.25.0.341.g760bfbb309-goog
On 2/11/20 1:35 PM, Mina Almasry wrote:
On Thu, Feb 6, 2020 at 2:31 PM Mike Kravetz mike.kravetz@oracle.com wrote:
On 2/3/20 3:22 PM, Mina Almasry wrote:
+++ b/mm/hugetlb.c @@ -1339,6 +1339,9 @@ static void __free_huge_page(struct page *page) clear_page_huge_active(page); hugetlb_cgroup_uncharge_page(hstate_index(h), pages_per_huge_page(h), page, false);
hugetlb_cgroup_uncharge_page(hstate_index(h), pages_per_huge_page(h),page, true);When looking at the code without change markings, the two above lines look so similar my first thought is there must be a mistake.
A suggestion for better code readability:
hugetlb_cgroup_uncharge_page could just take "struct hstate *h" and get both hstate_index(h) and pages_per_huge_page(h).
Perhaps make hugetlb_cgroup_uncharge_page and hugetlb_cgroup_uncharge_page_rsvd be wrappers around a common routine. Then the above would look like:
hugetlb_cgroup_uncharge_page(h, page); hugetlb_cgroup_uncharge_page_rsvd(h, page);
I did modify the interfaces to this, as it's much better for readability indeed. Unfortunately the patch the adds interfaces probably needs a re-review now as it's changed quite a bit, I did not carry your or David's Reviewed-by.
No worries. Happy to review again.
An earlier patch in this series disabled file_region coalescing in order to hang the hugetlb_cgroup uncharge info on the file_region entries.
This patch re-adds support for coalescing of file_region entries. Essentially everytime we add an entry, we check to see if the hugetlb_cgroup uncharge info is the same as any adjacent entries. If it is, instead of adding an entry we simply extend the appropriate entry.
This is an important performance optimization as private mappings add their entries page by page, and we could incur big performance costs for large mappings with lots of file_region entries in their resv_map.
Signed-off-by: Mina Almasry almasrymina@google.com
--- mm/hugetlb.c | 62 +++++++++++++++++++++++++++++++++++++++++++--------- 1 file changed, 52 insertions(+), 10 deletions(-)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index ec0b55ea1506e..058dd9c8269cf 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -272,6 +272,22 @@ static void record_hugetlb_cgroup_uncharge_info(struct hugetlb_cgroup *h_cg, #endif }
+static bool has_same_uncharge_info(struct file_region *rg, + struct hugetlb_cgroup *h_cg, + struct hstate *h) +{ +#ifdef CONFIG_CGROUP_HUGETLB + return rg && + rg->reservation_counter == + &h_cg->rsvd_hugepage[hstate_index(h)] && + rg->pages_per_hpage == pages_per_huge_page(h) && + rg->css == &h_cg->css; + +#else + return true; +#endif +} + /* Must be called with resv->lock held. Calling this with count_only == true * will count the number of pages to be added but will not modify the linked * list. If regions_needed != NULL and count_only == true, then regions_needed @@ -286,7 +302,7 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, long add = 0; struct list_head *head = &resv->regions; long last_accounted_offset = f; - struct file_region *rg = NULL, *trg = NULL, *nrg = NULL; + struct file_region *rg = NULL, *trg = NULL, *nrg = NULL, *prg = NULL;
if (regions_needed) *regions_needed = 0; @@ -318,16 +334,34 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, if (rg->from > last_accounted_offset) { add += rg->from - last_accounted_offset; if (!count_only) { - nrg = get_file_region_entry_from_cache( - resv, last_accounted_offset, rg->from); - record_hugetlb_cgroup_uncharge_info(h_cg, nrg, - h); - list_add(&nrg->link, rg->link.prev); + /* Check if the last region can be extended. */ + if (prg && prg->to == last_accounted_offset && + has_same_uncharge_info(prg, h_cg, h)) { + prg->to = rg->from; + /* Check if the next region can be extended. */ + } else if (has_same_uncharge_info(rg, h_cg, + h)) { + rg->from = last_accounted_offset; + /* If neither of the regions can be extended, + * add a region. + */ + } else { + nrg = get_file_region_entry_from_cache( + resv, last_accounted_offset, + rg->from); + record_hugetlb_cgroup_uncharge_info( + h_cg, nrg, h); + list_add(&nrg->link, rg->link.prev); + } } else if (regions_needed) *regions_needed += 1; }
last_accounted_offset = rg->to; + /* Record rg as the 'previous file region' incase we need it + * for the next iteration. + */ + prg = rg; }
/* Handle the case where our range extends beyond @@ -336,10 +370,18 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, if (last_accounted_offset < t) { add += t - last_accounted_offset; if (!count_only) { - nrg = get_file_region_entry_from_cache( - resv, last_accounted_offset, t); - record_hugetlb_cgroup_uncharge_info(h_cg, nrg, h); - list_add(&nrg->link, rg->link.prev); + /* Check if the last region can be extended. */ + if (prg && prg->to == last_accounted_offset && + has_same_uncharge_info(prg, h_cg, h)) { + prg->to = last_accounted_offset; + } else { + /* If not, just create a new region. */ + nrg = get_file_region_entry_from_cache( + resv, last_accounted_offset, t); + record_hugetlb_cgroup_uncharge_info(h_cg, nrg, + h); + list_add(&nrg->link, rg->link.prev); + } } else if (regions_needed) *regions_needed += 1; } -- 2.25.0.341.g760bfbb309-goog
On 2/3/20 3:22 PM, Mina Almasry wrote:
An earlier patch in this series disabled file_region coalescing in order to hang the hugetlb_cgroup uncharge info on the file_region entries.
This patch re-adds support for coalescing of file_region entries. Essentially everytime we add an entry, we check to see if the hugetlb_cgroup uncharge info is the same as any adjacent entries. If it is, instead of adding an entry we simply extend the appropriate entry.
This is an important performance optimization as private mappings add their entries page by page, and we could incur big performance costs for large mappings with lots of file_region entries in their resv_map.
Signed-off-by: Mina Almasry almasrymina@google.com
mm/hugetlb.c | 62 +++++++++++++++++++++++++++++++++++++++++++--------- 1 file changed, 52 insertions(+), 10 deletions(-)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index ec0b55ea1506e..058dd9c8269cf 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -272,6 +272,22 @@ static void record_hugetlb_cgroup_uncharge_info(struct hugetlb_cgroup *h_cg, #endif }
+static bool has_same_uncharge_info(struct file_region *rg,
struct hugetlb_cgroup *h_cg,struct hstate *h)+{ +#ifdef CONFIG_CGROUP_HUGETLB
- return rg &&
rg->reservation_counter ==&h_cg->rsvd_hugepage[hstate_index(h)] &&rg->pages_per_hpage == pages_per_huge_page(h) &&rg->css == &h_cg->css;+#else
- return true;
+#endif +}
/* Must be called with resv->lock held. Calling this with count_only == true
- will count the number of pages to be added but will not modify the linked
- list. If regions_needed != NULL and count_only == true, then regions_needed
@@ -286,7 +302,7 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, long add = 0; struct list_head *head = &resv->regions; long last_accounted_offset = f;
- struct file_region *rg = NULL, *trg = NULL, *nrg = NULL;
struct file_region *rg = NULL, *trg = NULL, *nrg = NULL, *prg = NULL;
if (regions_needed) *regions_needed = 0;
@@ -318,16 +334,34 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t,
I seem to be missing something. For context, here is the beginning of that loop:
/* In this loop, we essentially handle an entry for the range * [last_accounted_offset, rg->from), at every iteration, with some * bounds checking. */ list_for_each_entry_safe(rg, trg, head, link) { /* Skip irrelevant regions that start before our range. */ if (rg->from < f) { /* If this region ends after the last accounted offset, * then we need to update last_accounted_offset. */ if (rg->to > last_accounted_offset) last_accounted_offset = rg->to; continue; }
/* When we find a region that starts beyond our range, we've * finished. */ if (rg->from > t) break;
Suppose the resv_map contains one entry [0,2) and we are going to add [2,4). Will we not 'continue' after the first entry and then exit loop without setting prg? So, there is no chance for coalescing?
On Thu, Feb 6, 2020 at 4:17 PM Mike Kravetz mike.kravetz@oracle.com wrote:
On 2/3/20 3:22 PM, Mina Almasry wrote:
An earlier patch in this series disabled file_region coalescing in order to hang the hugetlb_cgroup uncharge info on the file_region entries.
This patch re-adds support for coalescing of file_region entries. Essentially everytime we add an entry, we check to see if the hugetlb_cgroup uncharge info is the same as any adjacent entries. If it is, instead of adding an entry we simply extend the appropriate entry.
This is an important performance optimization as private mappings add their entries page by page, and we could incur big performance costs for large mappings with lots of file_region entries in their resv_map.
Signed-off-by: Mina Almasry almasrymina@google.com
mm/hugetlb.c | 62 +++++++++++++++++++++++++++++++++++++++++++--------- 1 file changed, 52 insertions(+), 10 deletions(-)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index ec0b55ea1506e..058dd9c8269cf 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -272,6 +272,22 @@ static void record_hugetlb_cgroup_uncharge_info(struct hugetlb_cgroup *h_cg, #endif }
+static bool has_same_uncharge_info(struct file_region *rg,
struct hugetlb_cgroup *h_cg,struct hstate *h)+{ +#ifdef CONFIG_CGROUP_HUGETLB
return rg &&rg->reservation_counter ==&h_cg->rsvd_hugepage[hstate_index(h)] &&rg->pages_per_hpage == pages_per_huge_page(h) &&rg->css == &h_cg->css;+#else
return true;+#endif +}
/* Must be called with resv->lock held. Calling this with count_only == true
- will count the number of pages to be added but will not modify the linked
- list. If regions_needed != NULL and count_only == true, then regions_needed
@@ -286,7 +302,7 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, long add = 0; struct list_head *head = &resv->regions; long last_accounted_offset = f;
struct file_region *rg = NULL, *trg = NULL, *nrg = NULL;
struct file_region *rg = NULL, *trg = NULL, *nrg = NULL, *prg = NULL; if (regions_needed) *regions_needed = 0;@@ -318,16 +334,34 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t,
I seem to be missing something. For context, here is the beginning of that loop:
/* In this loop, we essentially handle an entry for the range * [last_accounted_offset, rg->from), at every iteration, with some * bounds checking. */ list_for_each_entry_safe(rg, trg, head, link) { /* Skip irrelevant regions that start before our range. */ if (rg->from < f) { /* If this region ends after the last accounted offset, * then we need to update last_accounted_offset. */ if (rg->to > last_accounted_offset) last_accounted_offset = rg->to; continue; } /* When we find a region that starts beyond our range, we've * finished. */ if (rg->from > t) break;Suppose the resv_map contains one entry [0,2) and we are going to add [2,4). Will we not 'continue' after the first entry and then exit loop without setting prg? So, there is no chance for coalescing?
I think you're right; prg needs to be set on all loop exits, including the continue and break. I'm thinking with that added, the logic should work, but I need to find a good way to test this. I thought I had good test coverage but apparently not. I'll fix this in the next iteration.
-- Mike Kravetz
if (rg->from > last_accounted_offset) { add += rg->from - last_accounted_offset; if (!count_only) {
nrg = get_file_region_entry_from_cache(resv, last_accounted_offset, rg->from);record_hugetlb_cgroup_uncharge_info(h_cg, nrg,h);list_add(&nrg->link, rg->link.prev);
/* Check if the last region can be extended. */if (prg && prg->to == last_accounted_offset &&has_same_uncharge_info(prg, h_cg, h)) {prg->to = rg->from;/* Check if the next region can be extended. */} else if (has_same_uncharge_info(rg, h_cg,h)) {rg->from = last_accounted_offset;/* If neither of the regions can be extended,* add a region.*/} else {nrg = get_file_region_entry_from_cache(resv, last_accounted_offset,rg->from);record_hugetlb_cgroup_uncharge_info(h_cg, nrg, h);list_add(&nrg->link, rg->link.prev);} } else if (regions_needed) *regions_needed += 1; } last_accounted_offset = rg->to;/* Record rg as the 'previous file region' incase we need it* for the next iteration.*/prg = rg; } /* Handle the case where our range extends beyond@@ -336,10 +370,18 @@ static long add_reservation_in_range(struct resv_map *resv, long f, long t, if (last_accounted_offset < t) { add += t - last_accounted_offset; if (!count_only) {
nrg = get_file_region_entry_from_cache(resv, last_accounted_offset, t);record_hugetlb_cgroup_uncharge_info(h_cg, nrg, h);list_add(&nrg->link, rg->link.prev);
/* Check if the last region can be extended. */if (prg && prg->to == last_accounted_offset &&has_same_uncharge_info(prg, h_cg, h)) {prg->to = last_accounted_offset;} else {/* If not, just create a new region. */nrg = get_file_region_entry_from_cache(resv, last_accounted_offset, t);record_hugetlb_cgroup_uncharge_info(h_cg, nrg,h);list_add(&nrg->link, rg->link.prev);} } else if (regions_needed) *regions_needed += 1; }-- 2.25.0.341.g760bfbb309-goog
The tests use both shared and private mapped hugetlb memory, and monitors the hugetlb usage counter as well as the hugetlb reservation counter. They test different configurations such as hugetlb memory usage via hugetlbfs, or MAP_HUGETLB, or shmget/shmat, and with and without MAP_POPULATE.
Also add test for hugetlb reservation reparenting, since this is a subtle issue.
Signed-off-by: Mina Almasry almasrymina@google.com Cc: sandipan@linux.ibm.com
---
Changes in v11: - Modify test to not assume 2MB hugepage size. - Updated resv.* to rsvd.* Changes in v10: - Updated tests to resv.* name changes. Changes in v9: - Added tests for hugetlb reparenting. - Make tests explicitly support cgroup v1 and v2 via script argument. Changes in v6: - Updates tests for cgroups-v2 and NORESERVE allocations.
--- tools/testing/selftests/vm/.gitignore | 1 + tools/testing/selftests/vm/Makefile | 1 + .../selftests/vm/charge_reserved_hugetlb.sh | 558 ++++++++++++++++++ .../selftests/vm/hugetlb_reparenting_test.sh | 235 ++++++++ .../selftests/vm/write_hugetlb_memory.sh | 23 + .../testing/selftests/vm/write_to_hugetlbfs.c | 261 ++++++++ 6 files changed, 1079 insertions(+) create mode 100755 tools/testing/selftests/vm/charge_reserved_hugetlb.sh create mode 100755 tools/testing/selftests/vm/hugetlb_reparenting_test.sh create mode 100644 tools/testing/selftests/vm/write_hugetlb_memory.sh create mode 100644 tools/testing/selftests/vm/write_to_hugetlbfs.c
diff --git a/tools/testing/selftests/vm/.gitignore b/tools/testing/selftests/vm/.gitignore index 31b3c98b6d34d..d3bed9407773c 100644 --- a/tools/testing/selftests/vm/.gitignore +++ b/tools/testing/selftests/vm/.gitignore @@ -14,3 +14,4 @@ virtual_address_range gup_benchmark va_128TBswitch map_fixed_noreplace +write_to_hugetlbfs diff --git a/tools/testing/selftests/vm/Makefile b/tools/testing/selftests/vm/Makefile index 7f9a8a8c31da9..662bb95e84c5d 100644 --- a/tools/testing/selftests/vm/Makefile +++ b/tools/testing/selftests/vm/Makefile @@ -22,6 +22,7 @@ TEST_GEN_FILES += userfaultfd ifneq (,$(filter $(ARCH),arm64 ia64 mips64 parisc64 ppc64 riscv64 s390x sh64 sparc64 x86_64)) TEST_GEN_FILES += va_128TBswitch TEST_GEN_FILES += virtual_address_range +TEST_GEN_FILES += write_to_hugetlbfs endif
TEST_PROGS := run_vmtests diff --git a/tools/testing/selftests/vm/charge_reserved_hugetlb.sh b/tools/testing/selftests/vm/charge_reserved_hugetlb.sh new file mode 100755 index 0000000000000..fa82a66e497a2 --- /dev/null +++ b/tools/testing/selftests/vm/charge_reserved_hugetlb.sh @@ -0,0 +1,558 @@ +#!/bin/sh +# SPDX-License-Identifier: GPL-2.0 + +set -e + +if [[ $(id -u) -ne 0 ]]; then + echo "This test must be run as root. Skipping..." + exit 0 +fi + +fault_limit_file=limit_in_bytes +reservation_limit_file=rsvd.limit_in_bytes +fault_usage_file=usage_in_bytes +reservation_usage_file=rsvd.usage_in_bytes + +if [[ "$1" == "-cgroup-v2" ]]; then + cgroup2=1 + fault_limit_file=max + reservation_limit_file=rsvd.max + fault_usage_file=current + reservation_usage_file=rsvd.current +fi + +cgroup_path=/dev/cgroup/memory +if [[ ! -e $cgroup_path ]]; then + mkdir -p $cgroup_path + if [[ $cgroup2 ]]; then + mount -t cgroup2 none $cgroup_path + else + mount -t cgroup memory,hugetlb $cgroup_path + fi +fi + +if [[ $cgroup2 ]]; then + echo "+hugetlb" >/dev/cgroup/memory/cgroup.subtree_control +fi + +function cleanup() { + if [[ $cgroup2 ]]; then + echo $$ >$cgroup_path/cgroup.procs + else + echo $$ >$cgroup_path/tasks + fi + + if [[ -e /mnt/huge ]]; then + rm -rf /mnt/huge/* + umount /mnt/huge || echo error + rmdir /mnt/huge + fi + if [[ -e $cgroup_path/hugetlb_cgroup_test ]]; then + rmdir $cgroup_path/hugetlb_cgroup_test + fi + if [[ -e $cgroup_path/hugetlb_cgroup_test1 ]]; then + rmdir $cgroup_path/hugetlb_cgroup_test1 + fi + if [[ -e $cgroup_path/hugetlb_cgroup_test2 ]]; then + rmdir $cgroup_path/hugetlb_cgroup_test2 + fi + echo 0 >/proc/sys/vm/nr_hugepages + echo CLEANUP DONE +} + +function expect_equal() { + local expected="$1" + local actual="$2" + local error="$3" + + if [[ "$expected" != "$actual" ]]; then + echo "expected ($expected) != actual ($actual): $3" + cleanup + exit 1 + fi +} + +function get_machine_hugepage_size() { + hpz=$(grep -i hugepagesize /proc/meminfo) + kb=${hpz:14:-3} + mb=$(($kb / 1024)) + echo $mb +} + +MB=$(get_machine_hugepage_size) + +function setup_cgroup() { + local name="$1" + local cgroup_limit="$2" + local reservation_limit="$3" + + mkdir $cgroup_path/$name + + echo writing cgroup limit: "$cgroup_limit" + echo "$cgroup_limit" >$cgroup_path/$name/hugetlb.${MB}MB.$fault_limit_file + + echo writing reseravation limit: "$reservation_limit" + echo "$reservation_limit" > \ + $cgroup_path/$name/hugetlb.${MB}MB.$reservation_limit_file + + if [ -e "$cgroup_path/$name/cpuset.cpus" ]; then + echo 0 >$cgroup_path/$name/cpuset.cpus + fi + if [ -e "$cgroup_path/$name/cpuset.mems" ]; then + echo 0 >$cgroup_path/$name/cpuset.mems + fi +} + +function wait_for_hugetlb_memory_to_get_depleted() { + local cgroup="$1" + local path="/dev/cgroup/memory/$cgroup/hugetlb.${MB}MB.$reservation_usage_file" + # Wait for hugetlbfs memory to get depleted. + while [ $(cat $path) != 0 ]; do + echo Waiting for hugetlb memory to get depleted. + cat $path + sleep 0.5 + done +} + +function wait_for_hugetlb_memory_to_get_reserved() { + local cgroup="$1" + local size="$2" + + local path="/dev/cgroup/memory/$cgroup/hugetlb.${MB}MB.$reservation_usage_file" + # Wait for hugetlbfs memory to get written. + while [ $(cat $path) != $size ]; do + echo Waiting for hugetlb memory reservation to reach size $size. + cat $path + sleep 0.5 + done +} + +function wait_for_hugetlb_memory_to_get_written() { + local cgroup="$1" + local size="$2" + + local path="/dev/cgroup/memory/$cgroup/hugetlb.${MB}MB.$fault_usage_file" + # Wait for hugetlbfs memory to get written. + while [ $(cat $path) != $size ]; do + echo Waiting for hugetlb memory to reach size $size. + cat $path + sleep 0.5 + done +} + +function write_hugetlbfs_and_get_usage() { + local cgroup="$1" + local size="$2" + local populate="$3" + local write="$4" + local path="$5" + local method="$6" + local private="$7" + local expect_failure="$8" + local reserve="$9" + + # Function return values. + reservation_failed=0 + oom_killed=0 + hugetlb_difference=0 + reserved_difference=0 + + local hugetlb_usage=$cgroup_path/$cgroup/hugetlb.${MB}MB.$fault_usage_file + local reserved_usage=$cgroup_path/$cgroup/hugetlb.${MB}MB.$reservation_usage_file + + local hugetlb_before=$(cat $hugetlb_usage) + local reserved_before=$(cat $reserved_usage) + + echo + echo Starting: + echo hugetlb_usage="$hugetlb_before" + echo reserved_usage="$reserved_before" + echo expect_failure is "$expect_failure" + + set +e + if [[ "$method" == "1" ]] || [[ "$method" == 2 ]] || + [[ "$private" == "-r" ]] && [[ "$expect_failure" != 1 ]]; then + + bash write_hugetlb_memory.sh "$size" "$populate" "$write" \ + "$cgroup" "$path" "$method" "$private" "-l" "$reserve" & + + local write_result=$? + + if [[ "$reserve" != "-n" ]]; then + wait_for_hugetlb_memory_to_get_reserved "$cgroup" "$size" + elif [[ "$populate" == "-o" ]] || [[ "$write" == "-w" ]]; then + wait_for_hugetlb_memory_to_get_written "$cgroup" "$size" + else + # This case doesn't produce visible effects, but we still have + # to wait for the async process to start and execute... + sleep 0.5 + fi + + echo write_result is $write_result + else + bash write_hugetlb_memory.sh "$size" "$populate" "$write" \ + "$cgroup" "$path" "$method" "$private" "$reserve" + local write_result=$? + + if [[ "$reserve" != "-n" ]]; then + wait_for_hugetlb_memory_to_get_reserved "$cgroup" "$size" + fi + fi + set -e + + if [[ "$write_result" == 1 ]]; then + reservation_failed=1 + fi + + # On linus/master, the above process gets SIGBUS'd on oomkill, with + # return code 135. On earlier kernels, it gets actual oomkill, with return + # code 137, so just check for both conditions in case we're testing + # against an earlier kernel. + if [[ "$write_result" == 135 ]] || [[ "$write_result" == 137 ]]; then + oom_killed=1 + fi + + local hugetlb_after=$(cat $hugetlb_usage) + local reserved_after=$(cat $reserved_usage) + + echo After write: + echo hugetlb_usage="$hugetlb_after" + echo reserved_usage="$reserved_after" + + hugetlb_difference=$(($hugetlb_after - $hugetlb_before)) + reserved_difference=$(($reserved_after - $reserved_before)) +} + +function cleanup_hugetlb_memory() { + set +e + local cgroup="$1" + if [[ "$(pgrep write_to_hugetlbfs)" != "" ]]; then + echo kiling write_to_hugetlbfs + killall -2 write_to_hugetlbfs + wait_for_hugetlb_memory_to_get_depleted $cgroup + fi + set -e + + if [[ -e /mnt/huge ]]; then + rm -rf /mnt/huge/* + umount /mnt/huge + rmdir /mnt/huge + fi +} + +function run_test() { + local size=$(($1 * ${MB} * 1024 * 1024)) + local populate="$2" + local write="$3" + local cgroup_limit=$(($4 * ${MB} * 1024 * 1024)) + local reservation_limit=$(($5 * ${MB} * 1024 * 1024)) + local nr_hugepages="$6" + local method="$7" + local private="$8" + local expect_failure="$9" + local reserve="${10}" + + # Function return values. + hugetlb_difference=0 + reserved_difference=0 + reservation_failed=0 + oom_killed=0 + + echo nr hugepages = "$nr_hugepages" + echo "$nr_hugepages" >/proc/sys/vm/nr_hugepages + + setup_cgroup "hugetlb_cgroup_test" "$cgroup_limit" "$reservation_limit" + + mkdir -p /mnt/huge + mount -t hugetlbfs -o pagesize=2M,size=256M none /mnt/huge + + write_hugetlbfs_and_get_usage "hugetlb_cgroup_test" "$size" "$populate" \ + "$write" "/mnt/huge/test" "$method" "$private" "$expect_failure" \ + "$reserve" + + cleanup_hugetlb_memory "hugetlb_cgroup_test" + + local final_hugetlb=$(cat $cgroup_path/hugetlb_cgroup_test/hugetlb.${MB}MB.$fault_usage_file) + local final_reservation=$(cat $cgroup_path/hugetlb_cgroup_test/hugetlb.${MB}MB.$reservation_usage_file) + + echo $hugetlb_difference + echo $reserved_difference + expect_equal "0" "$final_hugetlb" "final hugetlb is not zero" + expect_equal "0" "$final_reservation" "final reservation is not zero" +} + +function run_multiple_cgroup_test() { + local size1="$1" + local populate1="$2" + local write1="$3" + local cgroup_limit1="$4" + local reservation_limit1="$5" + + local size2="$6" + local populate2="$7" + local write2="$8" + local cgroup_limit2="$9" + local reservation_limit2="${10}" + + local nr_hugepages="${11}" + local method="${12}" + local private="${13}" + local expect_failure="${14}" + local reserve="${15}" + + # Function return values. + hugetlb_difference1=0 + reserved_difference1=0 + reservation_failed1=0 + oom_killed1=0 + + hugetlb_difference2=0 + reserved_difference2=0 + reservation_failed2=0 + oom_killed2=0 + + echo nr hugepages = "$nr_hugepages" + echo "$nr_hugepages" >/proc/sys/vm/nr_hugepages + + setup_cgroup "hugetlb_cgroup_test1" "$cgroup_limit1" "$reservation_limit1" + setup_cgroup "hugetlb_cgroup_test2" "$cgroup_limit2" "$reservation_limit2" + + mkdir -p /mnt/huge + mount -t hugetlbfs -o pagesize=2M,size=256M none /mnt/huge + + write_hugetlbfs_and_get_usage "hugetlb_cgroup_test1" "$size1" \ + "$populate1" "$write1" "/mnt/huge/test1" "$method" "$private" \ + "$expect_failure" "$reserve" + + hugetlb_difference1=$hugetlb_difference + reserved_difference1=$reserved_difference + reservation_failed1=$reservation_failed + oom_killed1=$oom_killed + + local cgroup1_hugetlb_usage=$cgroup_path/hugetlb_cgroup_test1/hugetlb.${MB}MB.$fault_usage_file + local cgroup1_reservation_usage=$cgroup_path/hugetlb_cgroup_test1/hugetlb.${MB}MB.$reservation_usage_file + local cgroup2_hugetlb_usage=$cgroup_path/hugetlb_cgroup_test2/hugetlb.${MB}MB.$fault_usage_file + local cgroup2_reservation_usage=$cgroup_path/hugetlb_cgroup_test2/hugetlb.${MB}MB.$reservation_usage_file + + local usage_before_second_write=$(cat $cgroup1_hugetlb_usage) + local reservation_usage_before_second_write=$(cat $cgroup1_reservation_usage) + + write_hugetlbfs_and_get_usage "hugetlb_cgroup_test2" "$size2" \ + "$populate2" "$write2" "/mnt/huge/test2" "$method" "$private" \ + "$expect_failure" "$reserve" + + hugetlb_difference2=$hugetlb_difference + reserved_difference2=$reserved_difference + reservation_failed2=$reservation_failed + oom_killed2=$oom_killed + + expect_equal "$usage_before_second_write" \ + "$(cat $cgroup1_hugetlb_usage)" "Usage changed." + expect_equal "$reservation_usage_before_second_write" \ + "$(cat $cgroup1_reservation_usage)" "Reservation usage changed." + + cleanup_hugetlb_memory + + local final_hugetlb=$(cat $cgroup1_hugetlb_usage) + local final_reservation=$(cat $cgroup1_reservation_usage) + + expect_equal "0" "$final_hugetlb" \ + "hugetlbt_cgroup_test1 final hugetlb is not zero" + expect_equal "0" "$final_reservation" \ + "hugetlbt_cgroup_test1 final reservation is not zero" + + local final_hugetlb=$(cat $cgroup2_hugetlb_usage) + local final_reservation=$(cat $cgroup2_reservation_usage) + + expect_equal "0" "$final_hugetlb" \ + "hugetlb_cgroup_test2 final hugetlb is not zero" + expect_equal "0" "$final_reservation" \ + "hugetlb_cgroup_test2 final reservation is not zero" +} + +cleanup + +for populate in "" "-o"; do + for method in 0 1 2; do + for private in "" "-r"; do + for reserve in "" "-n"; do + + # Skip mmap(MAP_HUGETLB | MAP_SHARED). Doesn't seem to be supported. + if [[ "$method" == 1 ]] && [[ "$private" == "" ]]; then + continue + fi + + # Skip populated shmem tests. Doesn't seem to be supported. + if [[ "$method" == 2"" ]] && [[ "$populate" == "-o" ]]; then + continue + fi + + if [[ "$method" == 2"" ]] && [[ "$reserve" == "-n" ]]; then + continue + fi + + cleanup + echo + echo + echo + echo Test normal case. + echo private=$private, populate=$populate, method=$method, reserve=$reserve + run_test 5 "$populate" "" 10 10 10 "$method" "$private" "0" "$reserve" + + echo Memory charged to hugtlb=$hugetlb_difference + echo Memory charged to reservation=$reserved_difference + + if [[ "$populate" == "-o" ]]; then + expect_equal "$((5 * $MB * 1024 * 1024))" "$hugetlb_difference" \ + "Reserved memory charged to hugetlb cgroup." + else + expect_equal "0" "$hugetlb_difference" \ + "Reserved memory charged to hugetlb cgroup." + fi + + if [[ "$reserve" != "-n" ]] || [[ "$populate" == "-o" ]]; then + expect_equal "$((5 * $MB * 1024 * 1024))" "$reserved_difference" \ + "Reserved memory not charged to reservation usage." + else + expect_equal "0" "$reserved_difference" \ + "Reserved memory not charged to reservation usage." + fi + + echo 'PASS' + + cleanup + echo + echo + echo + echo Test normal case with write. + echo private=$private, populate=$populate, method=$method, reserve=$reserve + run_test 5 "$populate" '-w' 5 5 10 "$method" "$private" "0" "$reserve" + + echo Memory charged to hugtlb=$hugetlb_difference + echo Memory charged to reservation=$reserved_difference + + expect_equal "$((5 * $MB * 1024 * 1024))" "$hugetlb_difference" \ + "Reserved memory charged to hugetlb cgroup." + + expect_equal "$((5 * $MB * 1024 * 1024))" "$reserved_difference" \ + "Reserved memory not charged to reservation usage." + + echo 'PASS' + + cleanup + continue + echo + echo + echo + echo Test more than reservation case. + echo private=$private, populate=$populate, method=$method, reserve=$reserve + + if [ "$reserve" != "-n" ]; then + run_test "5" "$populate" '' "10" "2" "10" "$method" "$private" "1" \ + "$reserve" + + expect_equal "1" "$reservation_failed" "Reservation succeeded." + fi + + echo 'PASS' + + cleanup + + echo + echo + echo + echo Test more than cgroup limit case. + echo private=$private, populate=$populate, method=$method, reserve=$reserve + + # Not sure if shm memory can be cleaned up when the process gets sigbus'd. + if [[ "$method" != 2 ]]; then + run_test 5 "$populate" "-w" 2 10 10 "$method" "$private" "1" "$reserve" + + expect_equal "1" "$oom_killed" "Not oom killed." + fi + echo 'PASS' + + cleanup + + echo + echo + echo + echo Test normal case, multiple cgroups. + echo private=$private, populate=$populate, method=$method, reserve=$reserve + run_multiple_cgroup_test "3" "$populate" "" "10" "10" "5" \ + "$populate" "" "10" "10" "10" \ + "$method" "$private" "0" "$reserve" + + echo Memory charged to hugtlb1=$hugetlb_difference1 + echo Memory charged to reservation1=$reserved_difference1 + echo Memory charged to hugtlb2=$hugetlb_difference2 + echo Memory charged to reservation2=$reserved_difference2 + + if [[ "$reserve" != "-n" ]] || [[ "$populate" == "-o" ]]; then + expect_equal "3" "$reserved_difference1" \ + "Incorrect reservations charged to cgroup 1." + + expect_equal "5" "$reserved_difference2" \ + "Incorrect reservation charged to cgroup 2." + + else + expect_equal "0" "$reserved_difference1" \ + "Incorrect reservations charged to cgroup 1." + + expect_equal "0" "$reserved_difference2" \ + "Incorrect reservation charged to cgroup 2." + fi + + if [[ "$populate" == "-o" ]]; then + expect_equal "3" "$hugetlb_difference1" \ + "Incorrect hugetlb charged to cgroup 1." + + expect_equal "5" "$hugetlb_difference2" \ + "Incorrect hugetlb charged to cgroup 2." + + else + expect_equal "0" "$hugetlb_difference1" \ + "Incorrect hugetlb charged to cgroup 1." + + expect_equal "0" "$hugetlb_difference2" \ + "Incorrect hugetlb charged to cgroup 2." + fi + echo 'PASS' + + cleanup + echo + echo + echo + echo Test normal case with write, multiple cgroups. + echo private=$private, populate=$populate, method=$method, reserve=$reserve + run_multiple_cgroup_test "3" "$populate" "-w" "10" "10" "5" \ + "$populate" "-w" "10" "10" "10" \ + "$method" "$private" "0" "$reserve" + + echo Memory charged to hugtlb1=$hugetlb_difference1 + echo Memory charged to reservation1=$reserved_difference1 + echo Memory charged to hugtlb2=$hugetlb_difference2 + echo Memory charged to reservation2=$reserved_difference2 + + expect_equal "3" "$hugetlb_difference1" \ + "Incorrect hugetlb charged to cgroup 1." + + expect_equal "3" "$reserved_difference1" \ + "Incorrect reservation charged to cgroup 1." + + expect_equal "5" "$hugetlb_difference2" \ + "Incorrect hugetlb charged to cgroup 2." + + expect_equal "5" "$reserved_difference2" \ + "Incorrected reservation charged to cgroup 2." + echo 'PASS' + + cleanup + + done # reserve + done # private + done # populate +done # method + +umount $cgroup_path +rmdir $cgroup_path diff --git a/tools/testing/selftests/vm/hugetlb_reparenting_test.sh b/tools/testing/selftests/vm/hugetlb_reparenting_test.sh new file mode 100755 index 0000000000000..2be672c2b311a --- /dev/null +++ b/tools/testing/selftests/vm/hugetlb_reparenting_test.sh @@ -0,0 +1,235 @@ +#!/bin/bash +# SPDX-License-Identifier: GPL-2.0 + +set -e + +if [[ $(id -u) -ne 0 ]]; then + echo "This test must be run as root. Skipping..." + exit 0 +fi + +usage_file=usage_in_bytes + +if [[ "$1" == "-cgroup-v2" ]]; then + cgroup2=1 + usage_file=current +fi + +CGROUP_ROOT='/dev/cgroup/memory' +MNT='/mnt/huge/' + +if [[ ! -e $CGROUP_ROOT ]]; then + mkdir -p $CGROUP_ROOT + if [[ $cgroup2 ]]; then + mount -t cgroup2 none $CGROUP_ROOT + sleep 1 + echo "+hugetlb +memory" >$CGROUP_ROOT/cgroup.subtree_control + else + mount -t cgroup memory,hugetlb $CGROUP_ROOT + fi +fi + +function cleanup() { + echo cleanup + set +e + rm -rf "$MNT"/* 2>/dev/null + umount "$MNT" 2>/dev/null + rmdir "$MNT" 2>/dev/null + rmdir "$CGROUP_ROOT"/a/b 2>/dev/null + rmdir "$CGROUP_ROOT"/a 2>/dev/null + rmdir "$CGROUP_ROOT"/test1 2>/dev/null + echo 0 >/proc/sys/vm/nr_hugepages + set -e +} + +function assert_state() { + local expected_a="$1" + local expected_a_hugetlb="$2" + local expected_b="" + local expected_b_hugetlb="" + + if [ ! -z ${3:-} ] && [ ! -z ${4:-} ]; then + expected_b="$3" + expected_b_hugetlb="$4" + fi + local tolerance=$((5 * 1024 * 1024)) + + local actual_a + actual_a="$(cat "$CGROUP_ROOT"/a/memory.$usage_file)" + if [[ $actual_a -lt $(($expected_a - $tolerance)) ]] || + [[ $actual_a -gt $(($expected_a + $tolerance)) ]]; then + echo actual a = $((${actual_a%% *} / 1024 / 1024)) MB + echo expected a = $((${expected_a%% *} / 1024 / 1024)) MB + echo fail + + cleanup + exit 1 + fi + + local actual_a_hugetlb + actual_a_hugetlb="$(cat "$CGROUP_ROOT"/a/hugetlb.2MB.$usage_file)" + if [[ $actual_a_hugetlb -lt $(($expected_a_hugetlb - $tolerance)) ]] || + [[ $actual_a_hugetlb -gt $(($expected_a_hugetlb + $tolerance)) ]]; then + echo actual a hugetlb = $((${actual_a_hugetlb%% *} / 1024 / 1024)) MB + echo expected a hugetlb = $((${expected_a_hugetlb%% *} / 1024 / 1024)) MB + echo fail + + cleanup + exit 1 + fi + + if [[ -z "$expected_b" || -z "$expected_b_hugetlb" ]]; then + return + fi + + local actual_b + actual_b="$(cat "$CGROUP_ROOT"/a/b/memory.$usage_file)" + if [[ $actual_b -lt $(($expected_b - $tolerance)) ]] || + [[ $actual_b -gt $(($expected_b + $tolerance)) ]]; then + echo actual b = $((${actual_b%% *} / 1024 / 1024)) MB + echo expected b = $((${expected_b%% *} / 1024 / 1024)) MB + echo fail + + cleanup + exit 1 + fi + + local actual_b_hugetlb + actual_b_hugetlb="$(cat "$CGROUP_ROOT"/a/b/hugetlb.2MB.$usage_file)" + if [[ $actual_b_hugetlb -lt $(($expected_b_hugetlb - $tolerance)) ]] || + [[ $actual_b_hugetlb -gt $(($expected_b_hugetlb + $tolerance)) ]]; then + echo actual b hugetlb = $((${actual_b_hugetlb%% *} / 1024 / 1024)) MB + echo expected b hugetlb = $((${expected_b_hugetlb%% *} / 1024 / 1024)) MB + echo fail + + cleanup + exit 1 + fi +} + +function setup() { + echo 100 >/proc/sys/vm/nr_hugepages + mkdir "$CGROUP_ROOT"/a + sleep 1 + if [[ $cgroup2 ]]; then + echo "+hugetlb +memory" >$CGROUP_ROOT/a/cgroup.subtree_control + else + echo 0 >$CGROUP_ROOT/a/cpuset.mems + echo 0 >$CGROUP_ROOT/a/cpuset.cpus + fi + + mkdir "$CGROUP_ROOT"/a/b + + if [[ ! $cgroup2 ]]; then + echo 0 >$CGROUP_ROOT/a/b/cpuset.mems + echo 0 >$CGROUP_ROOT/a/b/cpuset.cpus + fi + + mkdir -p "$MNT" + mount -t hugetlbfs none "$MNT" +} + +write_hugetlbfs() { + local cgroup="$1" + local path="$2" + local size="$3" + + if [[ $cgroup2 ]]; then + echo $$ >$CGROUP_ROOT/$cgroup/cgroup.procs + else + echo 0 >$CGROUP_ROOT/$cgroup/cpuset.mems + echo 0 >$CGROUP_ROOT/$cgroup/cpuset.cpus + echo $$ >"$CGROUP_ROOT/$cgroup/tasks" + fi + ./write_to_hugetlbfs -p "$path" -s "$size" -m 0 -o + if [[ $cgroup2 ]]; then + echo $$ >$CGROUP_ROOT/cgroup.procs + else + echo $$ >"$CGROUP_ROOT/tasks" + fi + echo +} + +set -e + +size=$((2 * 1024 * 1024 * 25)) # 50MB = 25 * 2MB hugepages. + +cleanup + +echo +echo +echo Test charge, rmdir, uncharge +setup +echo mkdir +mkdir $CGROUP_ROOT/test1 + +echo write +write_hugetlbfs test1 "$MNT"/test $size + +echo rmdir +rmdir $CGROUP_ROOT/test1 +mkdir $CGROUP_ROOT/test1 + +echo uncharge +rm -rf /mnt/huge/* + +cleanup + +echo done +echo +echo +if [[ ! $cgroup2 ]]; then + echo "Test parent and child hugetlb usage" + setup + + echo write + write_hugetlbfs a "$MNT"/test $size + + echo Assert memory charged correctly for parent use. + assert_state 0 $size 0 0 + + write_hugetlbfs a/b "$MNT"/test2 $size + + echo Assert memory charged correctly for child use. + assert_state 0 $(($size * 2)) 0 $size + + rmdir "$CGROUP_ROOT"/a/b + sleep 5 + echo Assert memory reparent correctly. + assert_state 0 $(($size * 2)) + + rm -rf "$MNT"/* + umount "$MNT" + echo Assert memory uncharged correctly. + assert_state 0 0 + + cleanup +fi + +echo +echo +echo "Test child only hugetlb usage" +echo setup +setup + +echo write +write_hugetlbfs a/b "$MNT"/test2 $size + +echo Assert memory charged correctly for child only use. +assert_state 0 $(($size)) 0 $size + +rmdir "$CGROUP_ROOT"/a/b +echo Assert memory reparent correctly. +assert_state 0 $size + +rm -rf "$MNT"/* +umount "$MNT" +echo Assert memory uncharged correctly. +assert_state 0 0 + +cleanup + +echo ALL PASS + +umount $CGROUP_ROOT +rm -rf $CGROUP_ROOT diff --git a/tools/testing/selftests/vm/write_hugetlb_memory.sh b/tools/testing/selftests/vm/write_hugetlb_memory.sh new file mode 100644 index 0000000000000..d3d0d108924d4 --- /dev/null +++ b/tools/testing/selftests/vm/write_hugetlb_memory.sh @@ -0,0 +1,23 @@ +#!/bin/sh +# SPDX-License-Identifier: GPL-2.0 + +set -e + +size=$1 +populate=$2 +write=$3 +cgroup=$4 +path=$5 +method=$6 +private=$7 +want_sleep=$8 +reserve=$9 + +echo "Putting task in cgroup '$cgroup'" +echo $$ > /dev/cgroup/memory/"$cgroup"/cgroup.procs + +echo "Method is $method" + +set +e +./write_to_hugetlbfs -p "$path" -s "$size" "$write" "$populate" -m "$method" \ + "$private" "$want_sleep" "$reserve" diff --git a/tools/testing/selftests/vm/write_to_hugetlbfs.c b/tools/testing/selftests/vm/write_to_hugetlbfs.c new file mode 100644 index 0000000000000..85811c3384a10 --- /dev/null +++ b/tools/testing/selftests/vm/write_to_hugetlbfs.c @@ -0,0 +1,261 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * This program reserves and uses hugetlb memory, supporting a bunch of + * scenarios needed by the charged_reserved_hugetlb.sh test. + */ + +#include <err.h> +#include <errno.h> +#include <signal.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> +#include <fcntl.h> +#include <sys/types.h> +#include <sys/shm.h> +#include <sys/stat.h> +#include <sys/mman.h> + +/* Global definitions. */ +enum method { + HUGETLBFS, + MMAP_MAP_HUGETLB, + SHM, + MAX_METHOD +}; + + +/* Global variables. */ +static const char *self; +static char *shmaddr; +static int shmid; + +/* + * Show usage and exit. + */ +static void exit_usage(void) +{ + printf("Usage: %s -p <path to hugetlbfs file> -s <size to map> " + "[-m <0=hugetlbfs | 1=mmap(MAP_HUGETLB)>] [-l] [-r] " + "[-o] [-w] [-n]\n", + self); + exit(EXIT_FAILURE); +} + +void sig_handler(int signo) +{ + printf("Received %d.\n", signo); + if (signo == SIGINT) { + printf("Deleting the memory\n"); + if (shmdt((const void *)shmaddr) != 0) { + perror("Detach failure"); + shmctl(shmid, IPC_RMID, NULL); + exit(4); + } + + shmctl(shmid, IPC_RMID, NULL); + printf("Done deleting the memory\n"); + } + exit(2); +} + +int main(int argc, char **argv) +{ + int fd = 0; + int key = 0; + int *ptr = NULL; + int c = 0; + int size = 0; + char path[256] = ""; + enum method method = MAX_METHOD; + int want_sleep = 0, private = 0; + int populate = 0; + int write = 0; + int reserve = 1; + + unsigned long i; + + if (signal(SIGINT, sig_handler) == SIG_ERR) + err(1, "\ncan't catch SIGINT\n"); + + /* Parse command-line arguments. */ + setvbuf(stdout, NULL, _IONBF, 0); + self = argv[0]; + + while ((c = getopt(argc, argv, "s:p:m:owlrn")) != -1) { + switch (c) { + case 's': + size = atoi(optarg); + break; + case 'p': + strncpy(path, optarg, sizeof(path)); + break; + case 'm': + if (atoi(optarg) >= MAX_METHOD) { + errno = EINVAL; + perror("Invalid -m."); + exit_usage(); + } + method = atoi(optarg); + break; + case 'o': + populate = 1; + break; + case 'w': + write = 1; + break; + case 'l': + want_sleep = 1; + break; + case 'r': + private + = 1; + break; + case 'n': + reserve = 0; + break; + default: + errno = EINVAL; + perror("Invalid arg"); + exit_usage(); + } + } + + if (strncmp(path, "", sizeof(path)) != 0) { + printf("Writing to this path: %s\n", path); + } else { + errno = EINVAL; + perror("path not found"); + exit_usage(); + } + + if (size != 0) { + printf("Writing this size: %d\n", size); + } else { + errno = EINVAL; + perror("size not found"); + exit_usage(); + } + + if (!populate) + printf("Not populating.\n"); + else + printf("Populating.\n"); + + if (!write) + printf("Not writing to memory.\n"); + + if (method == MAX_METHOD) { + errno = EINVAL; + perror("-m Invalid"); + exit_usage(); + } else + printf("Using method=%d\n", method); + + if (!private) + printf("Shared mapping.\n"); + else + printf("Private mapping.\n"); + + if (!reserve) + printf("NO_RESERVE mapping.\n"); + else + printf("RESERVE mapping.\n"); + + switch (method) { + case HUGETLBFS: + printf("Allocating using HUGETLBFS.\n"); + fd = open(path, O_CREAT | O_RDWR, 0777); + if (fd == -1) + err(1, "Failed to open file."); + + ptr = mmap(NULL, size, PROT_READ | PROT_WRITE, + (private ? MAP_PRIVATE : MAP_SHARED) | + (populate ? MAP_POPULATE : 0) | + (reserve ? 0 : MAP_NORESERVE), + fd, 0); + + if (ptr == MAP_FAILED) { + close(fd); + err(1, "Error mapping the file"); + } + break; + case MMAP_MAP_HUGETLB: + printf("Allocating using MAP_HUGETLB.\n"); + ptr = mmap(NULL, size, PROT_READ | PROT_WRITE, + (private ? (MAP_PRIVATE | MAP_ANONYMOUS) : + MAP_SHARED) | + MAP_HUGETLB | (populate ? MAP_POPULATE : 0) | + (reserve ? 0 : MAP_NORESERVE), + -1, 0); + + if (ptr == MAP_FAILED) + err(1, "mmap"); + + printf("Returned address is %p\n", ptr); + break; + case SHM: + printf("Allocating using SHM.\n"); + shmid = shmget(key, size, + SHM_HUGETLB | IPC_CREAT | SHM_R | SHM_W); + if (shmid < 0) { + shmid = shmget(++key, size, + SHM_HUGETLB | IPC_CREAT | SHM_R | SHM_W); + if (shmid < 0) + err(1, "shmget"); + } + printf("shmid: 0x%x, shmget key:%d\n", shmid, key); + + shmaddr = shmat(shmid, NULL, 0); + if (shmaddr == (char *)-1) { + perror("Shared memory attach failure"); + shmctl(shmid, IPC_RMID, NULL); + exit(2); + } + printf("shmaddr: %p\n", shmaddr); + + break; + default: + errno = EINVAL; + err(1, "Invalid method."); + } + + if (write) { + printf("Writing to memory.\n"); + if (method != SHM) { + memset(ptr, 1, size); + } else { + printf("Starting the writes:\n"); + for (i = 0; i < size; i++) { + shmaddr[i] = (char)(i); + if (!(i % (1024 * 1024))) + printf("."); + } + printf("\n"); + + printf("Starting the Check..."); + for (i = 0; i < size; i++) + if (shmaddr[i] != (char)i) { + printf("\nIndex %lu mismatched\n", i); + exit(3); + } + printf("Done.\n"); + } + } + + if (want_sleep) { + /* Signal to caller that we're done. */ + printf("DONE\n"); + + /* Hold memory until external kill signal is delivered. */ + while (1) + sleep(100); + } + + switch (method == HUGETLBFS) { + close(fd); + } + + return 0; +} -- 2.25.0.341.g760bfbb309-goog
Hi Mina,
On 04/02/20 4:52 am, Mina Almasry wrote:
The tests use both shared and private mapped hugetlb memory, and monitors the hugetlb usage counter as well as the hugetlb reservation counter. They test different configurations such as hugetlb memory usage via hugetlbfs, or MAP_HUGETLB, or shmget/shmat, and with and without MAP_POPULATE.
Also add test for hugetlb reservation reparenting, since this is a subtle issue.
Signed-off-by: Mina Almasry almasrymina@google.com Cc: sandipan@linux.ibm.com
Changes in v11:
- Modify test to not assume 2MB hugepage size.
- Updated resv.* to rsvd.*
Changes in v10:
- Updated tests to resv.* name changes.
Changes in v9:
- Added tests for hugetlb reparenting.
- Make tests explicitly support cgroup v1 and v2 via script argument.
Changes in v6:
- Updates tests for cgroups-v2 and NORESERVE allocations.
There are still a couple of places where 2MB page size is being used. These are my workarounds to get the tests running on ppc64.
diff --git a/tools/testing/selftests/vm/hugetlb_reparenting_test.sh b/tools/testing/selftests/vm/hugetlb_reparenting_test.sh index 2be672c2b311..d11d1febccc3 100755 --- a/tools/testing/selftests/vm/hugetlb_reparenting_test.sh +++ b/tools/testing/selftests/vm/hugetlb_reparenting_test.sh @@ -29,6 +29,15 @@ if [[ ! -e $CGROUP_ROOT ]]; then fi fi
+function get_machine_hugepage_size() { + hpz=$(grep -i hugepagesize /proc/meminfo) + kb=${hpz:14:-3} + mb=$(($kb / 1024)) + echo $mb +} + +MB=$(get_machine_hugepage_size) + function cleanup() { echo cleanup set +e @@ -67,7 +76,7 @@ function assert_state() { fi
local actual_a_hugetlb - actual_a_hugetlb="$(cat "$CGROUP_ROOT"/a/hugetlb.2MB.$usage_file)" + actual_a_hugetlb="$(cat "$CGROUP_ROOT"/a/hugetlb.${MB}MB.$usage_file)" if [[ $actual_a_hugetlb -lt $(($expected_a_hugetlb - $tolerance)) ]] || [[ $actual_a_hugetlb -gt $(($expected_a_hugetlb + $tolerance)) ]]; then echo actual a hugetlb = $((${actual_a_hugetlb%% *} / 1024 / 1024)) MB @@ -95,7 +104,7 @@ function assert_state() { fi
local actual_b_hugetlb - actual_b_hugetlb="$(cat "$CGROUP_ROOT"/a/b/hugetlb.2MB.$usage_file)" + actual_b_hugetlb="$(cat "$CGROUP_ROOT"/a/b/hugetlb.${MB}MB.$usage_file)" if [[ $actual_b_hugetlb -lt $(($expected_b_hugetlb - $tolerance)) ]] || [[ $actual_b_hugetlb -gt $(($expected_b_hugetlb + $tolerance)) ]]; then echo actual b hugetlb = $((${actual_b_hugetlb%% *} / 1024 / 1024)) MB @@ -152,7 +161,7 @@ write_hugetlbfs() {
set -e
-size=$((2 * 1024 * 1024 * 25)) # 50MB = 25 * 2MB hugepages. +size=$((${MB} * 1024 * 1024 * 25)) # 50MB = 25 * 2MB hugepages.
cleanup
diff --git a/tools/testing/selftests/vm/charge_reserved_hugetlb.sh b/tools/testing/selftests/vm/charge_reserved_hugetlb.sh index fa82a66e497a..ca98ad229b75 100755 --- a/tools/testing/selftests/vm/charge_reserved_hugetlb.sh +++ b/tools/testing/selftests/vm/charge_reserved_hugetlb.sh @@ -226,7 +226,7 @@ function write_hugetlbfs_and_get_usage() { function cleanup_hugetlb_memory() { set +e local cgroup="$1" - if [[ "$(pgrep write_to_hugetlbfs)" != "" ]]; then + if [[ "$(pgrep -f write_to_hugetlbfs)" != "" ]]; then echo kiling write_to_hugetlbfs killall -2 write_to_hugetlbfs wait_for_hugetlb_memory_to_get_depleted $cgroup @@ -264,7 +264,7 @@ function run_test() { setup_cgroup "hugetlb_cgroup_test" "$cgroup_limit" "$reservation_limit"
mkdir -p /mnt/huge - mount -t hugetlbfs -o pagesize=2M,size=256M none /mnt/huge + mount -t hugetlbfs -o pagesize=${MB}M,size=256M none /mnt/huge
write_hugetlbfs_and_get_usage "hugetlb_cgroup_test" "$size" "$populate" \ "$write" "/mnt/huge/test" "$method" "$private" "$expect_failure" \ @@ -318,7 +318,7 @@ function run_multiple_cgroup_test() { setup_cgroup "hugetlb_cgroup_test2" "$cgroup_limit2" "$reservation_limit2"
mkdir -p /mnt/huge - mount -t hugetlbfs -o pagesize=2M,size=256M none /mnt/huge + mount -t hugetlbfs -o pagesize=${MB}M,size=256M none /mnt/huge
write_hugetlbfs_and_get_usage "hugetlb_cgroup_test1" "$size1" \ "$populate1" "$write1" "/mnt/huge/test1" "$method" "$private" \
---
Also I had missed running charge_reserved_hugetlb.sh the last time. Right now, it stops at the following scenario.
Test normal case with write. private=, populate=, method=2, reserve= nr hugepages = 10 writing cgroup limit: 83886080 writing reseravation limit: 83886080
Starting: hugetlb_usage=0 reserved_usage=0 expect_failure is 0 Putting task in cgroup 'hugetlb_cgroup_test' Method is 2 Executing ./write_to_hugetlbfs -p /mnt/huge/test -s 83886080 -w -m 2 -l Writing to this path: /mnt/huge/test Writing this size: 83886080 Not populating. Using method=2 Shared mapping. RESERVE mapping. Allocating using SHM. shmid: 0x5, shmget key:0 shmaddr: 0x7dfffb000000 Writing to memory. Starting the writes: .write_result is 0 .After write: hugetlb_usage=16777216 reserved_usage=83886080 ....kiling write_to_hugetlbfs ...Received 2. Deleting the memory Done deleting the memory 16777216 83886080 Memory charged to hugtlb=16777216 Memory charged to reservation=83886080 expected (83886080) != actual (16777216): Reserved memory charged to hugetlb cgroup. CLEANUP DONE
The other test script (hugetlb_reparenting_test.sh) passes. Did not observe anything unusual with hugepage accounting either.
- Sandipan
On Tue, Feb 4, 2020 at 8:26 AM Sandipan Das sandipan@linux.ibm.com wrote:
There are still a couple of places where 2MB page size is being used. These are my workarounds to get the tests running on ppc64.
Thanks for the changes!
Also I had missed running charge_reserved_hugetlb.sh the last time. Right now, it stops at the following scenario.
Test normal case with write. private=, populate=, method=2, reserve= nr hugepages = 10 writing cgroup limit: 83886080 writing reseravation limit: 83886080
Starting: hugetlb_usage=0 reserved_usage=0 expect_failure is 0 Putting task in cgroup 'hugetlb_cgroup_test' Method is 2 Executing ./write_to_hugetlbfs -p /mnt/huge/test -s 83886080 -w -m 2 -l Writing to this path: /mnt/huge/test Writing this size: 83886080 Not populating. Using method=2 Shared mapping. RESERVE mapping. Allocating using SHM. shmid: 0x5, shmget key:0 shmaddr: 0x7dfffb000000 Writing to memory. Starting the writes: .write_result is 0 .After write: hugetlb_usage=16777216 reserved_usage=83886080 ....kiling write_to_hugetlbfs ...Received 2. Deleting the memory Done deleting the memory 16777216 83886080 Memory charged to hugtlb=16777216 Memory charged to reservation=83886080 expected (83886080) != actual (16777216): Reserved memory charged to hugetlb cgroup. CLEANUP DONE
So the problem in this log seems to be that this log line is missing: echo Waiting for hugetlb memory to reach size $size.
The way the test works is that it starts a process that writes the hugetlb memory, then it *should* wait until the memory is written, then it should record the cgroup accounting and kill the process. It seems from your log that the wait doesn't happen, so the test continues before the background process has had time to write the memory properly. Essentially wait_for_hugetlb_memory_to_get_written() never gets called in your log.
Can you try this additional attached diff on top of your changes? I attached the diff and pasted the same here, hopefully one works for you:
diff --git a/tools/testing/selftests/vm/charge_reserved_hugetlb.sh b/tools/testing/selftests/vm/charge_reserved_hugetlb.sh index efd68093ce3e9..18d33684faade 100755 --- a/tools/testing/selftests/vm/charge_reserved_hugetlb.sh +++ b/tools/testing/selftests/vm/charge_reserved_hugetlb.sh @@ -169,19 +169,36 @@ function write_hugetlbfs_and_get_usage() { echo reserved_usage="$reserved_before" echo expect_failure is "$expect_failure"
+ output=$(mktemp) set +e if [[ "$method" == "1" ]] || [[ "$method" == 2 ]] || [[ "$private" == "-r" ]] && [[ "$expect_failure" != 1 ]]; then
bash write_hugetlb_memory.sh "$size" "$populate" "$write" \ - "$cgroup" "$path" "$method" "$private" "-l" "$reserve" & + "$cgroup" "$path" "$method" "$private" "-l" "$reserve" 2>&1 | tee $output &
local write_result=$? + local write_pid=$!
- if [[ "$reserve" != "-n" ]]; then - wait_for_hugetlb_memory_to_get_reserved "$cgroup" "$size" - elif [[ "$populate" == "-o" ]] || [[ "$write" == "-w" ]]; then + until grep -q -i "DONE" $output; do + echo waiting for DONE signal. + if ! ps $write_pid > /dev/null + then + echo "FAIL: The write died" + cleanup + exit 1 + fi + sleep 0.5 + done + + echo ================= write_hugetlb_memory.sh output is: + cat $output + echo ================= end output. + + if [[ "$populate" == "-o" ]] || [[ "$write" == "-w" ]]; then wait_for_hugetlb_memory_to_get_written "$cgroup" "$size" + elif [[ "$reserve" != "-n" ]]; then + wait_for_hugetlb_memory_to_get_reserved "$cgroup" "$size" else # This case doesn't produce visible effects, but we still have # to wait for the async process to start and execute... @@ -227,7 +244,7 @@ function cleanup_hugetlb_memory() { set +e local cgroup="$1" if [[ "$(pgrep -f write_to_hugetlbfs)" != "" ]]; then - echo kiling write_to_hugetlbfs + echo killing write_to_hugetlbfs killall -2 write_to_hugetlbfs wait_for_hugetlb_memory_to_get_depleted $cgroup fi diff --git a/tools/testing/selftests/vm/write_to_hugetlbfs.c b/tools/testing/selftests/vm/write_to_hugetlbfs.c index 85811c3384a10..7f75ad5f7b580 100644 --- a/tools/testing/selftests/vm/write_to_hugetlbfs.c +++ b/tools/testing/selftests/vm/write_to_hugetlbfs.c @@ -207,13 +207,13 @@ int main(int argc, char **argv) } printf("shmid: 0x%x, shmget key:%d\n", shmid, key);
- shmaddr = shmat(shmid, NULL, 0); - if (shmaddr == (char *)-1) { + ptr = shmat(shmid, NULL, 0); + if (ptr == (int *)-1) { perror("Shared memory attach failure"); shmctl(shmid, IPC_RMID, NULL); exit(2); } - printf("shmaddr: %p\n", shmaddr); + printf("shmaddr: %p\n", ptr);
break; default: @@ -223,25 +223,7 @@ int main(int argc, char **argv)
if (write) { printf("Writing to memory.\n"); - if (method != SHM) { - memset(ptr, 1, size); - } else { - printf("Starting the writes:\n"); - for (i = 0; i < size; i++) { - shmaddr[i] = (char)(i); - if (!(i % (1024 * 1024))) - printf("."); - } - printf("\n"); - - printf("Starting the Check..."); - for (i = 0; i < size; i++) - if (shmaddr[i] != (char)i) { - printf("\nIndex %lu mismatched\n", i); - exit(3); - } - printf("Done.\n"); - } + memset(ptr, 1, size); }
if (want_sleep) { @@ -253,7 +235,7 @@ int main(int argc, char **argv) sleep(100); }
- switch (method == HUGETLBFS) { + if (method == HUGETLBFS) { close(fd); }
On Tue, Feb 4, 2020 at 12:36 PM Mina Almasry almasrymina@google.com wrote:
On Tue, Feb 4, 2020 at 8:26 AM Sandipan Das sandipan@linux.ibm.com wrote:
There are still a couple of places where 2MB page size is being used. These are my workarounds to get the tests running on ppc64.
Thanks for the changes!
Also I had missed running charge_reserved_hugetlb.sh the last time. Right now, it stops at the following scenario.
Test normal case with write. private=, populate=, method=2, reserve= nr hugepages = 10 writing cgroup limit: 83886080 writing reseravation limit: 83886080
Starting: hugetlb_usage=0 reserved_usage=0 expect_failure is 0 Putting task in cgroup 'hugetlb_cgroup_test' Method is 2 Executing ./write_to_hugetlbfs -p /mnt/huge/test -s 83886080 -w -m 2 -l Writing to this path: /mnt/huge/test Writing this size: 83886080 Not populating. Using method=2 Shared mapping. RESERVE mapping. Allocating using SHM. shmid: 0x5, shmget key:0 shmaddr: 0x7dfffb000000 Writing to memory. Starting the writes: .write_result is 0 .After write: hugetlb_usage=16777216 reserved_usage=83886080 ....kiling write_to_hugetlbfs ...Received 2. Deleting the memory Done deleting the memory 16777216 83886080 Memory charged to hugtlb=16777216 Memory charged to reservation=83886080 expected (83886080) != actual (16777216): Reserved memory charged to hugetlb cgroup. CLEANUP DONE
So the problem in this log seems to be that this log line is missing: echo Waiting for hugetlb memory to reach size $size.
The way the test works is that it starts a process that writes the hugetlb memory, then it *should* wait until the memory is written, then it should record the cgroup accounting and kill the process. It seems from your log that the wait doesn't happen, so the test continues before the background process has had time to write the memory properly. Essentially wait_for_hugetlb_memory_to_get_written() never gets called in your log.
Can you try this additional attached diff on top of your changes? I attached the diff and pasted the same here, hopefully one works for you:
I got my hands on a machine with 16MB default hugepage size and charge_reserved_hugetlb.sh passes now after my changes. Please let me know if you still run into issues.
Hi,
On 05/02/20 4:03 am, Mina Almasry wrote:
On Tue, Feb 4, 2020 at 12:36 PM Mina Almasry almasrymina@google.com wrote:
So the problem in this log seems to be that this log line is missing: echo Waiting for hugetlb memory to reach size $size.
The way the test works is that it starts a process that writes the hugetlb memory, then it *should* wait until the memory is written, then it should record the cgroup accounting and kill the process. It seems from your log that the wait doesn't happen, so the test continues before the background process has had time to write the memory properly. Essentially wait_for_hugetlb_memory_to_get_written() never gets called in your log.
Can you try this additional attached diff on top of your changes? I attached the diff and pasted the same here, hopefully one works for you: ...
I got my hands on a machine with 16MB default hugepage size and charge_reserved_hugetlb.sh passes now after my changes. Please let me know if you still run into issues.
With your updates, the tests are passing. Ran the tests on a ppc64 system that uses radix MMU (2MB hugepages) and everything passed there as well.
- Sandipan
On Wed, Feb 5, 2020 at 4:42 AM Sandipan Das sandipan@linux.ibm.com wrote:
Hi,
On 05/02/20 4:03 am, Mina Almasry wrote:
On Tue, Feb 4, 2020 at 12:36 PM Mina Almasry almasrymina@google.com wrote:
So the problem in this log seems to be that this log line is missing: echo Waiting for hugetlb memory to reach size $size.
The way the test works is that it starts a process that writes the hugetlb memory, then it *should* wait until the memory is written, then it should record the cgroup accounting and kill the process. It seems from your log that the wait doesn't happen, so the test continues before the background process has had time to write the memory properly. Essentially wait_for_hugetlb_memory_to_get_written() never gets called in your log.
Can you try this additional attached diff on top of your changes? I attached the diff and pasted the same here, hopefully one works for you: ...
I got my hands on a machine with 16MB default hugepage size and charge_reserved_hugetlb.sh passes now after my changes. Please let me know if you still run into issues.
With your updates, the tests are passing. Ran the tests on a ppc64 system that uses radix MMU (2MB hugepages) and everything passed there as well.
Thanks, please consider reviewing the next iteration of the patch then.
- Sandipan
Add docs for how to use hugetlb_cgroup reservations, and their behavior.
Signed-off-by: Mina Almasry almasrymina@google.com
---
Changes in v11: - Changed resv.* to rsvd.* Changes in v10: - Clarify reparenting behavior. - Reword benefits of reservation limits. Changes in v6: - Updated docs to reflect the new design based on a new counter that tracks both reservations and faults.
--- .../admin-guide/cgroup-v1/hugetlb.rst | 103 ++++++++++++++++-- 1 file changed, 92 insertions(+), 11 deletions(-)
diff --git a/Documentation/admin-guide/cgroup-v1/hugetlb.rst b/Documentation/admin-guide/cgroup-v1/hugetlb.rst index a3902aa253a96..338f2c7d7a1cd 100644 --- a/Documentation/admin-guide/cgroup-v1/hugetlb.rst +++ b/Documentation/admin-guide/cgroup-v1/hugetlb.rst @@ -2,13 +2,6 @@ HugeTLB Controller ==================
-The HugeTLB controller allows to limit the HugeTLB usage per control group and -enforces the controller limit during page fault. Since HugeTLB doesn't -support page reclaim, enforcing the limit at page fault time implies that, -the application will get SIGBUS signal if it tries to access HugeTLB pages -beyond its limit. This requires the application to know beforehand how much -HugeTLB pages it would require for its use. - HugeTLB controller can be created by first mounting the cgroup filesystem.
# mount -t cgroup -o hugetlb none /sys/fs/cgroup @@ -28,10 +21,14 @@ process (bash) into it.
Brief summary of control files::
- hugetlb.<hugepagesize>.limit_in_bytes # set/show limit of "hugepagesize" hugetlb usage - hugetlb.<hugepagesize>.max_usage_in_bytes # show max "hugepagesize" hugetlb usage recorded - hugetlb.<hugepagesize>.usage_in_bytes # show current usage for "hugepagesize" hugetlb - hugetlb.<hugepagesize>.failcnt # show the number of allocation failure due to HugeTLB limit + hugetlb.<hugepagesize>.rsvd.limit_in_bytes # set/show limit of "hugepagesize" hugetlb reservations + hugetlb.<hugepagesize>.rsvd.max_usage_in_bytes # show max "hugepagesize" hugetlb reservations and no-reserve faults + hugetlb.<hugepagesize>.rsvd.usage_in_bytes # show current reservations and no-reserve faults for "hugepagesize" hugetlb + hugetlb.<hugepagesize>.rsvd.failcnt # show the number of allocation failure due to HugeTLB reservation limit + hugetlb.<hugepagesize>.limit_in_bytes # set/show limit of "hugepagesize" hugetlb faults + hugetlb.<hugepagesize>.max_usage_in_bytes # show max "hugepagesize" hugetlb usage recorded + hugetlb.<hugepagesize>.usage_in_bytes # show current usage for "hugepagesize" hugetlb + hugetlb.<hugepagesize>.failcnt # show the number of allocation failure due to HugeTLB usage limit
For a system supporting three hugepage sizes (64k, 32M and 1G), the control files include:: @@ -40,11 +37,95 @@ files include:: hugetlb.1GB.max_usage_in_bytes hugetlb.1GB.usage_in_bytes hugetlb.1GB.failcnt + hugetlb.1GB.rsvd.limit_in_bytes + hugetlb.1GB.rsvd.max_usage_in_bytes + hugetlb.1GB.rsvd.usage_in_bytes + hugetlb.1GB.rsvd.failcnt hugetlb.64KB.limit_in_bytes hugetlb.64KB.max_usage_in_bytes hugetlb.64KB.usage_in_bytes hugetlb.64KB.failcnt + hugetlb.64KB.rsvd.limit_in_bytes + hugetlb.64KB.rsvd.max_usage_in_bytes + hugetlb.64KB.rsvd.usage_in_bytes + hugetlb.64KB.rsvd.failcnt hugetlb.32MB.limit_in_bytes hugetlb.32MB.max_usage_in_bytes hugetlb.32MB.usage_in_bytes hugetlb.32MB.failcnt + hugetlb.32MB.rsvd.limit_in_bytes + hugetlb.32MB.rsvd.max_usage_in_bytes + hugetlb.32MB.rsvd.usage_in_bytes + hugetlb.32MB.rsvd.failcnt + + +1. Page fault accounting + +hugetlb.<hugepagesize>.limit_in_bytes +hugetlb.<hugepagesize>.max_usage_in_bytes +hugetlb.<hugepagesize>.usage_in_bytes +hugetlb.<hugepagesize>.failcnt + +The HugeTLB controller allows users to limit the HugeTLB usage (page fault) per +control group and enforces the limit during page fault. Since HugeTLB +doesn't support page reclaim, enforcing the limit at page fault time implies +that, the application will get SIGBUS signal if it tries to fault in HugeTLB +pages beyond its limit. Therefore the application needs to know exactly how many +HugeTLB pages it uses before hand, and the sysadmin needs to make sure that +there are enough available on the machine for all the users to avoid processes +getting SIGBUS. + + +2. Reservation accounting + +hugetlb.<hugepagesize>.rsvd.limit_in_bytes +hugetlb.<hugepagesize>.rsvd.max_usage_in_bytes +hugetlb.<hugepagesize>.rsvd.usage_in_bytes +hugetlb.<hugepagesize>.rsvd.failcnt + +The HugeTLB controller allows to limit the HugeTLB reservations per control +group and enforces the controller limit at reservation time and at the fault of +HugeTLB memory for which no reservation exists. Since reservation limits are +enforced at reservation time (on mmap or shget), reservation limits never causes +the application to get SIGBUS signal if the memory was reserved before hand. For +MAP_NORESERVE allocations, the reservation limit behaves the same as the fault +limit, enforcing memory usage at fault time and causing the application to +receive a SIGBUS if it's crossing its limit. + +Reservation limits are superior to page fault limits described above, since +reservation limits are enforced at reservation time (on mmap or shget), and +never causes the application to get SIGBUS signal if the memory was reserved +before hand. This allows for easier fallback to alternatives such as +non-HugeTLB memory for example. In the case of page fault accounting, it's very +hard to avoid processes getting SIGBUS since the sysadmin needs precisely know +the HugeTLB usage of all the tasks in the system and make sure there is enough +pages to satisfy all requests. Avoiding tasks getting SIGBUS on overcommited +systems is practically impossible with page fault accounting. + + +3. Caveats with shared memory + +For shared HugeTLB memory, both HugeTLB reservation and page faults are charged +to the first task that causes the memory to be reserved or faulted, and all +subsequent uses of this reserved or faulted memory is done without charging. + +Shared HugeTLB memory is only uncharged when it is unreserved or deallocated. +This is usually when the HugeTLB file is deleted, and not when the task that +caused the reservation or fault has exited. + + +4. Caveats with HugeTLB cgroup offline. + +When a HugeTLB cgroup goes offline with some reservations or faults still +charged to it, the behavior is as follows: + +- The fault charges are charged to the parent HugeTLB cgroup (reparented), +- the reservation charges remain on the offline HugeTLB cgroup. + +This means that if a HugeTLB cgroup gets offlined while there is still HugeTLB +reservations charged to it, that cgroup persists as a zombie until all HugeTLB +reservations are uncharged. HugeTLB reservations behave in this manner to match +the memory controller whose cgroups also persist as zombie until all charged +memory is uncharged. Also, the tracking of HugeTLB reservations is a bit more +complex compared to the tracking of HugeTLB faults, so it is significantly +harder to reparent reservations at offline time. -- 2.25.0.341.g760bfbb309-goog
On 2/3/20 3:22 PM, Mina Almasry wrote:
These counters will track hugetlb reservations rather than hugetlb memory faulted in. This patch only adds the counter, following patches add the charging and uncharging of the counter.
This is patch 1 of an 9 patch series.
Problem: Currently tasks attempting to reserve more hugetlb memory than is available get a failure at mmap/shmget time. This is thanks to Hugetlbfs Reservations [1]. However, if a task attempts to reserve more hugetlb memory than its hugetlb_cgroup limit allows, the kernel will allow the mmap/shmget call, but will SIGBUS the task when it attempts to fault in the excess memory.
We have users hitting their hugetlb_cgroup limits and thus we've been looking at this failure mode. We'd like to improve this behavior such that users violating the hugetlb_cgroup limits get an error on mmap/shmget time, rather than getting SIGBUS'd when they try to fault the excess memory in. This gives the user an opportunity to fallback more gracefully to non-hugetlbfs memory for example.
The underlying problem is that today's hugetlb_cgroup accounting happens at hugetlb memory *fault* time, rather than at *reservation* time. Thus, enforcing the hugetlb_cgroup limit only happens at fault time, and the offending task gets SIGBUS'd.
Proposed Solution: A new page counter named 'hugetlb.xMB.rsvd.[limit|usage|max_usage]_in_bytes'. This counter has slightly different semantics than 'hugetlb.xMB.[limit|usage|max_usage]_in_bytes':
- While usage_in_bytes tracks all *faulted* hugetlb memory,
rsvd.usage_in_bytes tracks all *reserved* hugetlb memory and hugetlb memory faulted in without a prior reservation.
- If a task attempts to reserve more memory than limit_in_bytes allows,
the kernel will allow it to do so. But if a task attempts to reserve more memory than rsvd.limit_in_bytes, the kernel will fail this reservation.
This proposal is implemented in this patch series, with tests to verify functionality and show the usage.
Alternatives considered:
A new cgroup, instead of only a new page_counter attached to the existing hugetlb_cgroup. Adding a new cgroup seemed like a lot of code duplication with hugetlb_cgroup. Keeping hugetlb related page counters under hugetlb_cgroup seemed cleaner as well.
Instead of adding a new counter, we considered adding a sysctl that modifies the behavior of hugetlb.xMB.[limit|usage]_in_bytes, to do accounting at reservation time rather than fault time. Adding a new page_counter seems better as userspace could, if it wants, choose to enforce different cgroups differently: one via limit_in_bytes, and another via rsvd.limit_in_bytes. This could be very useful if you're transitioning how hugetlb memory is partitioned on your system one cgroup at a time, for example. Also, someone may find usage for both limit_in_bytes and rsvd.limit_in_bytes concurrently, and this approach gives them the option to do so.
Testing:
- Added tests passing.
- Used libhugetlbfs for regression testing.
Signed-off-by: Mina Almasry almasrymina@google.com Acked-by: David Rientjes rientjes@google.com
Changes in v11:
- Renamed resv.* or 'reservation' or 'reserved' to rsvd.*
- Renamed hugetlb_cgroup_get_counter() to
Thanks! I was mostly concerned about using 'resv' in cgroup file names visible to users. Changing variable names is good as well.
Small nit, some lines of commit message wrap.
Reviewed-by: Mike Kravetz mike.kravetz@oracle.com
linux-kselftest-mirror@lists.linaro.org
-
Mike Kravetz -
Mina Almasry -
Sandipan Das