- Linux-stable-mirror - lists.linaro.org

FAILED: patch "[PATCH] Revert "net: skb: introduce and use a single page frag cache"" failed to apply to 6.12-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 6.12-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.12.y git checkout FETCH_HEAD git cherry-pick -x 011b0335903832facca86cd8ed05d7d8d94c9c76 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2025021859-renewal-onto-1877@gregkh' --subject-prefix 'PATCH 6.12.y' HEAD^.. Possible dependencies: thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 011b0335903832facca86cd8ed05d7d8d94c9c76 Mon Sep 17 00:00:00 2001 From: Paolo Abeni <pabeni(a)redhat.com> Date: Thu, 6 Feb 2025 22:28:48 +0100 Subject: [PATCH] Revert "net: skb: introduce and use a single page frag cache" This reverts commit dbae2b062824 ("net: skb: introduce and use a single page frag cache"). The intended goal of such change was to counter a performance regression introduced by commit 3226b158e67c ("net: avoid 32 x truesize under-estimation for tiny skbs"). Unfortunately, the blamed commit introduces another regression for the virtio_net driver. Such a driver calls napi_alloc_skb() with a tiny size, so that the whole head frag could fit a 512-byte block. The single page frag cache uses a 1K fragment for such allocation, and the additional overhead, under small UDP packets flood, makes the page allocator a bottleneck. Thanks to commit bf9f1baa279f ("net: add dedicated kmem_cache for typical/small skb->head"), this revert does not re-introduce the original regression. Actually, in the relevant test on top of this revert, I measure a small but noticeable positive delta, just above noise level. The revert itself required some additional mangling due to the introduction of the SKB_HEAD_ALIGN() helper and local lock infra in the affected code. Suggested-by: Eric Dumazet <edumazet(a)google.com> Fixes: dbae2b062824 ("net: skb: introduce and use a single page frag cache") Signed-off-by: Paolo Abeni <pabeni(a)redhat.com> Link: https://patch.msgid.link/e649212fde9f0fdee23909ca0d14158d32bb7425.173887729… Signed-off-by: Jakub Kicinski <kuba(a)kernel.org> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h index c0a86afb85da..365f0e2098d1 100644 --- a/include/linux/netdevice.h +++ b/include/linux/netdevice.h @@ -4115,7 +4115,6 @@ void netif_receive_skb_list(struct list_head *head); gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb); void napi_gro_flush(struct napi_struct *napi, bool flush_old); struct sk_buff *napi_get_frags(struct napi_struct *napi); -void napi_get_frags_check(struct napi_struct *napi); gro_result_t napi_gro_frags(struct napi_struct *napi); static inline void napi_free_frags(struct napi_struct *napi) diff --git a/net/core/dev.c b/net/core/dev.c index b91658e8aedb..55e356a68db6 100644 --- a/net/core/dev.c +++ b/net/core/dev.c @@ -6920,6 +6920,23 @@ netif_napi_dev_list_add(struct net_device *dev, struct napi_struct *napi) list_add_rcu(&napi->dev_list, higher); /* adds after higher */ } +/* Double check that napi_get_frags() allocates skbs with + * skb->head being backed by slab, not a page fragment. + * This is to make sure bug fixed in 3226b158e67c + * ("net: avoid 32 x truesize under-estimation for tiny skbs") + * does not accidentally come back. + */ +static void napi_get_frags_check(struct napi_struct *napi) +{ + struct sk_buff *skb; + + local_bh_disable(); + skb = napi_get_frags(napi); + WARN_ON_ONCE(skb && skb->head_frag); + napi_free_frags(napi); + local_bh_enable(); +} + void netif_napi_add_weight_locked(struct net_device *dev, struct napi_struct *napi, int (*poll)(struct napi_struct *, int), diff --git a/net/core/skbuff.c b/net/core/skbuff.c index a441613a1e6c..6a99c453397f 100644 --- a/net/core/skbuff.c +++ b/net/core/skbuff.c @@ -220,67 +220,9 @@ static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) #define NAPI_SKB_CACHE_BULK 16 #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2) -#if PAGE_SIZE == SZ_4K - -#define NAPI_HAS_SMALL_PAGE_FRAG 1 -#define NAPI_SMALL_PAGE_PFMEMALLOC(nc) ((nc).pfmemalloc) - -/* specialized page frag allocator using a single order 0 page - * and slicing it into 1K sized fragment. Constrained to systems - * with a very limited amount of 1K fragments fitting a single - * page - to avoid excessive truesize underestimation - */ - -struct page_frag_1k { - void *va; - u16 offset; - bool pfmemalloc; -}; - -static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp) -{ - struct page *page; - int offset; - - offset = nc->offset - SZ_1K; - if (likely(offset >= 0)) - goto use_frag; - - page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); - if (!page) - return NULL; - - nc->va = page_address(page); - nc->pfmemalloc = page_is_pfmemalloc(page); - offset = PAGE_SIZE - SZ_1K; - page_ref_add(page, offset / SZ_1K); - -use_frag: - nc->offset = offset; - return nc->va + offset; -} -#else - -/* the small page is actually unused in this build; add dummy helpers - * to please the compiler and avoid later preprocessor's conditionals - */ -#define NAPI_HAS_SMALL_PAGE_FRAG 0 -#define NAPI_SMALL_PAGE_PFMEMALLOC(nc) false - -struct page_frag_1k { -}; - -static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask) -{ - return NULL; -} - -#endif - struct napi_alloc_cache { local_lock_t bh_lock; struct page_frag_cache page; - struct page_frag_1k page_small; unsigned int skb_count; void *skb_cache[NAPI_SKB_CACHE_SIZE]; }; @@ -290,23 +232,6 @@ static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache) = { .bh_lock = INIT_LOCAL_LOCK(bh_lock), }; -/* Double check that napi_get_frags() allocates skbs with - * skb->head being backed by slab, not a page fragment. - * This is to make sure bug fixed in 3226b158e67c - * ("net: avoid 32 x truesize under-estimation for tiny skbs") - * does not accidentally come back. - */ -void napi_get_frags_check(struct napi_struct *napi) -{ - struct sk_buff *skb; - - local_bh_disable(); - skb = napi_get_frags(napi); - WARN_ON_ONCE(!NAPI_HAS_SMALL_PAGE_FRAG && skb && skb->head_frag); - napi_free_frags(napi); - local_bh_enable(); -} - void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) { struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); @@ -813,10 +738,8 @@ struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len) /* If requested length is either too small or too big, * we use kmalloc() for skb->head allocation. - * When the small frag allocator is available, prefer it over kmalloc - * for small fragments */ - if ((!NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) || + if (len <= SKB_WITH_OVERHEAD(1024) || len > SKB_WITH_OVERHEAD(PAGE_SIZE) || (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI, @@ -826,32 +749,16 @@ struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len) goto skb_success; } + len = SKB_HEAD_ALIGN(len); + if (sk_memalloc_socks()) gfp_mask |= __GFP_MEMALLOC; local_lock_nested_bh(&napi_alloc_cache.bh_lock); nc = this_cpu_ptr(&napi_alloc_cache); - if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) { - /* we are artificially inflating the allocation size, but - * that is not as bad as it may look like, as: - * - 'len' less than GRO_MAX_HEAD makes little sense - * - On most systems, larger 'len' values lead to fragment - * size above 512 bytes - * - kmalloc would use the kmalloc-1k slab for such values - * - Builds with smaller GRO_MAX_HEAD will very likely do - * little networking, as that implies no WiFi and no - * tunnels support, and 32 bits arches. - */ - len = SZ_1K; - data = page_frag_alloc_1k(&nc->page_small, gfp_mask); - pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small); - } else { - len = SKB_HEAD_ALIGN(len); - - data = page_frag_alloc(&nc->page, len, gfp_mask); - pfmemalloc = page_frag_cache_is_pfmemalloc(&nc->page); - } + data = page_frag_alloc(&nc->page, len, gfp_mask); + pfmemalloc = page_frag_cache_is_pfmemalloc(&nc->page); local_unlock_nested_bh(&napi_alloc_cache.bh_lock); if (unlikely(!data))

8 months, 4 weeks

2
2
0 0

[PATCH 0/2] vsock: fix use-after free and null-ptr-deref

by Luigi Leonardi

Hi all, This series contains two patches that are already available upstream: - The first commit fixes a use-after-free[1], but introduced a null-ptr-deref[2]. - The second commit fixes it. [3] I suggested waiting for both of them to be merged upstream and then applying them togheter to stable[4]. It should be applied to: - 6.13.y - 6.12.y - 6.6.y I will send another series for - 6.1.y - 5.15.y - 5.10.y because of conflicts. [1]https://lore.kernel.org/all/20250128-vsock-transport-vs-autobind-v3-0-1cf… [2]https://lore.kernel.org/all/67a09300.050a0220.d7c5a.008b.GAE@google.com/ [3]https://lore.kernel.org/all/20250210-vsock-linger-nullderef-v3-0-ef6244d0… [4]https://lore.kernel.org/all/2025020644-unwitting-scary-3c0d@gregkh/ Thanks, Luigi --- Michal Luczaj (2): vsock: Keep the binding until socket destruction vsock: Orphan socket after transport release net/vmw_vsock/af_vsock.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) --- base-commit: a1856aaa2ca74c88751f7d255dfa0c8c50fcc1ca change-id: 20250214-linux-rolling-stable-d73f0bed815d Best regards, -- Luigi Leonardi <leonardi(a)redhat.com>

8 months, 4 weeks

3
7
0 0

[PATCH 6.6.y] drm/amd/display: Add null check for head_pipe in dcn201_acquire_free_pipe_for_layer

by Xiangyu Chen

From: Srinivasan Shanmugam <srinivasan.shanmugam(a)amd.com> [ Upstream commit f22f4754aaa47d8c59f166ba3042182859e5dff7 ] This commit addresses a potential null pointer dereference issue in the `dcn201_acquire_free_pipe_for_layer` function. The issue could occur when `head_pipe` is null. The fix adds a check to ensure `head_pipe` is not null before asserting it. If `head_pipe` is null, the function returns NULL to prevent a potential null pointer dereference. Reported by smatch: drivers/gpu/drm/amd/amdgpu/../display/dc/resource/dcn201/dcn201_resource.c:1016 dcn201_acquire_free_pipe_for_layer() error: we previously assumed 'head_pipe' could be null (see line 1010) Cc: Tom Chung <chiahsuan.chung(a)amd.com> Cc: Rodrigo Siqueira <Rodrigo.Siqueira(a)amd.com> Cc: Roman Li <roman.li(a)amd.com> Cc: Alex Hung <alex.hung(a)amd.com> Cc: Aurabindo Pillai <aurabindo.pillai(a)amd.com> Cc: Harry Wentland <harry.wentland(a)amd.com> Cc: Hamza Mahfooz <hamza.mahfooz(a)amd.com> Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam(a)amd.com> Reviewed-by: Tom Chung <chiahsuan.chung(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> [dcn201 was moved from drivers/gpu/drm/amd/display/dc to drivers/gpu/drm/amd/display/dc/resource since 8b8eed05a1c6 ("drm/amd/display: Refactor resource into component directory"). The path is changed accordingly to apply the patch on 6.6.y.] Signed-off-by: Xiangyu Chen <xiangyu.chen(a)windriver.com> Signed-off-by: He Zhe <zhe.he(a)windriver.com> --- Verified the build test only due to we don't have DCN201 device. --- drivers/gpu/drm/amd/display/dc/dcn201/dcn201_resource.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/display/dc/dcn201/dcn201_resource.c b/drivers/gpu/drm/amd/display/dc/dcn201/dcn201_resource.c index 2dc4d2c1410b..8efe3f32a0e7 100644 --- a/drivers/gpu/drm/amd/display/dc/dcn201/dcn201_resource.c +++ b/drivers/gpu/drm/amd/display/dc/dcn201/dcn201_resource.c @@ -1002,8 +1002,10 @@ static struct pipe_ctx *dcn201_acquire_free_pipe_for_layer( struct pipe_ctx *head_pipe = resource_get_otg_master_for_stream(res_ctx, opp_head_pipe->stream); struct pipe_ctx *idle_pipe = resource_find_free_secondary_pipe_legacy(res_ctx, pool, head_pipe); - if (!head_pipe) + if (!head_pipe) { ASSERT(0); + return NULL; + } if (!idle_pipe) return NULL; -- 2.25.1

8 months, 4 weeks

3
2
0 0

Re: Patch "alpha: make stack 16-byte aligned (most cases)" has been added to the 6.6-stable tree

by Ivan Kokshaysky

On Tue, Feb 18, 2025 at 01:55:45PM +0100, gregkh(a)linuxfoundation.org wrote: > > This is a note to let you know that I've just added the patch titled > > alpha: make stack 16-byte aligned (most cases) Hi Greg, thanks for applying this! > Patches currently in stable-queue which might be from ink(a)unseen.parts are > > queue-6.6/alpha-make-stack-16-byte-aligned-most-cases.patch > queue-6.6/alpha-align-stack-for-page-fault-and-user-unaligned-trap-handlers.patch The third one (commit 77b823fa619f97d alpha: replace hardcoded stack offsets with autogenerated ones) is also needed, but it won't apply as-is to 6.6 and older kernels. Do you want me to provide the patches? Ivan.

8 months, 4 weeks

2
4
0 0

[PATCH 6.6.y] drm/amd/display: Pass non-null to dcn20_validate_apply_pipe_split_flags

by Xiangyu Chen

From: Alex Hung <alex.hung(a)amd.com> [ upstream commit 5559598742fb4538e4c51c48ef70563c49c2af23 ] [WHAT & HOW] "dcn20_validate_apply_pipe_split_flags" dereferences merge, and thus it cannot be a null pointer. Let's pass a valid pointer to avoid null dereference. This fixes 2 FORWARD_NULL issues reported by Coverity. Reviewed-by: Rodrigo Siqueira <rodrigo.siqueira(a)amd.com> Signed-off-by: Jerry Zuo <jerry.zuo(a)amd.com> Signed-off-by: Alex Hung <alex.hung(a)amd.com> Tested-by: Daniel Wheeler <daniel.wheeler(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> [dcn20 and dcn21 were moved from drivers/gpu/drm/amd/display/dc to drivers/gpu/drm/amd/display/dc/resource since 8b8eed05a1c6 ("drm/amd/display: Refactor resource into component directory"). The path is changed accordingly to apply the patch on 6.6.y.] Signed-off-by: Xiangyu Chen <xiangyu.chen(a)windriver.com> Signed-off-by: He Zhe <zhe.he(a)windriver.com> --- Verified the build test only due to we don't have dcn20/dcn21 device. --- drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c | 3 ++- drivers/gpu/drm/amd/display/dc/dcn21/dcn21_resource.c | 3 ++- 2 files changed, 4 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c index d587f807dfd7..294609557b73 100644 --- a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c +++ b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c @@ -2026,6 +2026,7 @@ bool dcn20_fast_validate_bw( { bool out = false; int split[MAX_PIPES] = { 0 }; + bool merge[MAX_PIPES] = { false }; int pipe_cnt, i, pipe_idx, vlevel; ASSERT(pipes); @@ -2050,7 +2051,7 @@ bool dcn20_fast_validate_bw( if (vlevel > context->bw_ctx.dml.soc.num_states) goto validate_fail; - vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, vlevel, split, NULL); + vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, vlevel, split, merge); /*initialize pipe_just_split_from to invalid idx*/ for (i = 0; i < MAX_PIPES; i++) diff --git a/drivers/gpu/drm/amd/display/dc/dcn21/dcn21_resource.c b/drivers/gpu/drm/amd/display/dc/dcn21/dcn21_resource.c index 8dffa5b6426e..24105a5b9f2a 100644 --- a/drivers/gpu/drm/amd/display/dc/dcn21/dcn21_resource.c +++ b/drivers/gpu/drm/amd/display/dc/dcn21/dcn21_resource.c @@ -800,6 +800,7 @@ bool dcn21_fast_validate_bw(struct dc *dc, { bool out = false; int split[MAX_PIPES] = { 0 }; + bool merge[MAX_PIPES] = { false }; int pipe_cnt, i, pipe_idx, vlevel; ASSERT(pipes); @@ -842,7 +843,7 @@ bool dcn21_fast_validate_bw(struct dc *dc, goto validate_fail; } - vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, vlevel, split, NULL); + vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, vlevel, split, merge); for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; -- 2.25.1

8 months, 4 weeks

2
1
0 0

6.12.y stable backport request for bpf selftests fixes

by Alan Maguire

Please backport 42602e3a06f8e5b9a059344e305c9bee2dcc87c8 bpf: handle implicit declaration of function gettid in bpf_iter.c and 4b7c05598a644782b8451e415bb56f31e5c9d3ee selftests/bpf: Fix uprobe consumer test to 6.12.y to fix BPF selftest-related issues (compilation failure and test failure respectively). Both apply to linux-6.12.y cleanly. Thank you! Alan

8 months, 4 weeks

2
1
0 0

Patches for the stable branches

by Jürgen Groß

Greg, please add the following upstream commits to the stable branches: 5cc2db37124bb33914996d6fdbb2ddb3811f2945 98a5cfd2320966f40fe049a9855f8787f0126825 0bd797b801bd8ee06c822844e20d73aaea0878dd They are fixing boot failures when running as a Xen PVH guest for some kernel configurations. Thanks, Juergen

8 months, 4 weeks

2
1
0 0

Request to apply 6d3e0d8cc632 to stable kernels

by Brian Norris

Hi, May I request this be ported to stable kernels? 6d3e0d8cc632 ("kdb: Do not assume write() callback available") It landed in v6.6, and I've tested the trivial cherry-pick to v6.1.y. AFAICT, it makes sense and applies cleanly to at least v5.15.y and v5.10.y (and possibly more), although I didn't test those. It's a bit of an older (but trivial) fix, and I assume not many folks actually test out KDB/KGDB that often, but I wasted my time re-discovering it. Thanks, Brian

8 months, 4 weeks

2
1
0 0

FAILED: patch "[PATCH] Revert "net: skb: introduce and use a single page frag cache"" failed to apply to 6.13-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 6.13-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.13.y git checkout FETCH_HEAD git cherry-pick -x 011b0335903832facca86cd8ed05d7d8d94c9c76 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2025021858-cobalt-scanner-666f@gregkh' --subject-prefix 'PATCH 6.13.y' HEAD^.. Possible dependencies: thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 011b0335903832facca86cd8ed05d7d8d94c9c76 Mon Sep 17 00:00:00 2001 From: Paolo Abeni <pabeni(a)redhat.com> Date: Thu, 6 Feb 2025 22:28:48 +0100 Subject: [PATCH] Revert "net: skb: introduce and use a single page frag cache" This reverts commit dbae2b062824 ("net: skb: introduce and use a single page frag cache"). The intended goal of such change was to counter a performance regression introduced by commit 3226b158e67c ("net: avoid 32 x truesize under-estimation for tiny skbs"). Unfortunately, the blamed commit introduces another regression for the virtio_net driver. Such a driver calls napi_alloc_skb() with a tiny size, so that the whole head frag could fit a 512-byte block. The single page frag cache uses a 1K fragment for such allocation, and the additional overhead, under small UDP packets flood, makes the page allocator a bottleneck. Thanks to commit bf9f1baa279f ("net: add dedicated kmem_cache for typical/small skb->head"), this revert does not re-introduce the original regression. Actually, in the relevant test on top of this revert, I measure a small but noticeable positive delta, just above noise level. The revert itself required some additional mangling due to the introduction of the SKB_HEAD_ALIGN() helper and local lock infra in the affected code. Suggested-by: Eric Dumazet <edumazet(a)google.com> Fixes: dbae2b062824 ("net: skb: introduce and use a single page frag cache") Signed-off-by: Paolo Abeni <pabeni(a)redhat.com> Link: https://patch.msgid.link/e649212fde9f0fdee23909ca0d14158d32bb7425.173887729… Signed-off-by: Jakub Kicinski <kuba(a)kernel.org> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h index c0a86afb85da..365f0e2098d1 100644 --- a/include/linux/netdevice.h +++ b/include/linux/netdevice.h @@ -4115,7 +4115,6 @@ void netif_receive_skb_list(struct list_head *head); gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb); void napi_gro_flush(struct napi_struct *napi, bool flush_old); struct sk_buff *napi_get_frags(struct napi_struct *napi); -void napi_get_frags_check(struct napi_struct *napi); gro_result_t napi_gro_frags(struct napi_struct *napi); static inline void napi_free_frags(struct napi_struct *napi) diff --git a/net/core/dev.c b/net/core/dev.c index b91658e8aedb..55e356a68db6 100644 --- a/net/core/dev.c +++ b/net/core/dev.c @@ -6920,6 +6920,23 @@ netif_napi_dev_list_add(struct net_device *dev, struct napi_struct *napi) list_add_rcu(&napi->dev_list, higher); /* adds after higher */ } +/* Double check that napi_get_frags() allocates skbs with + * skb->head being backed by slab, not a page fragment. + * This is to make sure bug fixed in 3226b158e67c + * ("net: avoid 32 x truesize under-estimation for tiny skbs") + * does not accidentally come back. + */ +static void napi_get_frags_check(struct napi_struct *napi) +{ + struct sk_buff *skb; + + local_bh_disable(); + skb = napi_get_frags(napi); + WARN_ON_ONCE(skb && skb->head_frag); + napi_free_frags(napi); + local_bh_enable(); +} + void netif_napi_add_weight_locked(struct net_device *dev, struct napi_struct *napi, int (*poll)(struct napi_struct *, int), diff --git a/net/core/skbuff.c b/net/core/skbuff.c index a441613a1e6c..6a99c453397f 100644 --- a/net/core/skbuff.c +++ b/net/core/skbuff.c @@ -220,67 +220,9 @@ static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) #define NAPI_SKB_CACHE_BULK 16 #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2) -#if PAGE_SIZE == SZ_4K - -#define NAPI_HAS_SMALL_PAGE_FRAG 1 -#define NAPI_SMALL_PAGE_PFMEMALLOC(nc) ((nc).pfmemalloc) - -/* specialized page frag allocator using a single order 0 page - * and slicing it into 1K sized fragment. Constrained to systems - * with a very limited amount of 1K fragments fitting a single - * page - to avoid excessive truesize underestimation - */ - -struct page_frag_1k { - void *va; - u16 offset; - bool pfmemalloc; -}; - -static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp) -{ - struct page *page; - int offset; - - offset = nc->offset - SZ_1K; - if (likely(offset >= 0)) - goto use_frag; - - page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); - if (!page) - return NULL; - - nc->va = page_address(page); - nc->pfmemalloc = page_is_pfmemalloc(page); - offset = PAGE_SIZE - SZ_1K; - page_ref_add(page, offset / SZ_1K); - -use_frag: - nc->offset = offset; - return nc->va + offset; -} -#else - -/* the small page is actually unused in this build; add dummy helpers - * to please the compiler and avoid later preprocessor's conditionals - */ -#define NAPI_HAS_SMALL_PAGE_FRAG 0 -#define NAPI_SMALL_PAGE_PFMEMALLOC(nc) false - -struct page_frag_1k { -}; - -static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask) -{ - return NULL; -} - -#endif - struct napi_alloc_cache { local_lock_t bh_lock; struct page_frag_cache page; - struct page_frag_1k page_small; unsigned int skb_count; void *skb_cache[NAPI_SKB_CACHE_SIZE]; }; @@ -290,23 +232,6 @@ static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache) = { .bh_lock = INIT_LOCAL_LOCK(bh_lock), }; -/* Double check that napi_get_frags() allocates skbs with - * skb->head being backed by slab, not a page fragment. - * This is to make sure bug fixed in 3226b158e67c - * ("net: avoid 32 x truesize under-estimation for tiny skbs") - * does not accidentally come back. - */ -void napi_get_frags_check(struct napi_struct *napi) -{ - struct sk_buff *skb; - - local_bh_disable(); - skb = napi_get_frags(napi); - WARN_ON_ONCE(!NAPI_HAS_SMALL_PAGE_FRAG && skb && skb->head_frag); - napi_free_frags(napi); - local_bh_enable(); -} - void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) { struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); @@ -813,10 +738,8 @@ struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len) /* If requested length is either too small or too big, * we use kmalloc() for skb->head allocation. - * When the small frag allocator is available, prefer it over kmalloc - * for small fragments */ - if ((!NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) || + if (len <= SKB_WITH_OVERHEAD(1024) || len > SKB_WITH_OVERHEAD(PAGE_SIZE) || (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI, @@ -826,32 +749,16 @@ struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len) goto skb_success; } + len = SKB_HEAD_ALIGN(len); + if (sk_memalloc_socks()) gfp_mask |= __GFP_MEMALLOC; local_lock_nested_bh(&napi_alloc_cache.bh_lock); nc = this_cpu_ptr(&napi_alloc_cache); - if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) { - /* we are artificially inflating the allocation size, but - * that is not as bad as it may look like, as: - * - 'len' less than GRO_MAX_HEAD makes little sense - * - On most systems, larger 'len' values lead to fragment - * size above 512 bytes - * - kmalloc would use the kmalloc-1k slab for such values - * - Builds with smaller GRO_MAX_HEAD will very likely do - * little networking, as that implies no WiFi and no - * tunnels support, and 32 bits arches. - */ - len = SZ_1K; - data = page_frag_alloc_1k(&nc->page_small, gfp_mask); - pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small); - } else { - len = SKB_HEAD_ALIGN(len); - - data = page_frag_alloc(&nc->page, len, gfp_mask); - pfmemalloc = page_frag_cache_is_pfmemalloc(&nc->page); - } + data = page_frag_alloc(&nc->page, len, gfp_mask); + pfmemalloc = page_frag_cache_is_pfmemalloc(&nc->page); local_unlock_nested_bh(&napi_alloc_cache.bh_lock); if (unlikely(!data))

8 months, 4 weeks

2
1
0 0

[PATCH v2] KVM: PPC: Enable CAP_SPAPR_TCE_VFIO on pSeries KVM guests

by Amit Machhiwal

Currently on book3s-hv, the capability KVM_CAP_SPAPR_TCE_VFIO is only available for KVM Guests running on PowerNV and not for the KVM guests running on pSeries hypervisors. This prevents a pSeries L2 guest from leveraging the in-kernel acceleration for H_PUT_TCE_INDIRECT and H_STUFF_TCE hcalls that results in slow startup times for large memory guests. Fix this by enabling the CAP_SPAPR_TCE_VFIO on the pSeries hosts as well for the nested PAPR guests. With the patch, booting an L2 guest with 128G memory results in an average improvement of 11% in the startup times. Fixes: f431a8cde7f1 ("powerpc/iommu: Reimplement the iommu_table_group_ops for pSeries") Cc: stable(a)vger.kernel.org Signed-off-by: Amit Machhiwal <amachhiw(a)linux.ibm.com> --- Changes since v1: * Addressed review comments from Ritesh * v1: https://lore.kernel.org/all/20250109132053.158436-1-amachhiw@linux.ibm.com/ arch/powerpc/kvm/powerpc.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/arch/powerpc/kvm/powerpc.c b/arch/powerpc/kvm/powerpc.c index ce1d91eed231..a7138eb18d59 100644 --- a/arch/powerpc/kvm/powerpc.c +++ b/arch/powerpc/kvm/powerpc.c @@ -543,26 +543,23 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) r = !hv_enabled; break; #ifdef CONFIG_KVM_MPIC case KVM_CAP_IRQ_MPIC: r = 1; break; #endif #ifdef CONFIG_PPC_BOOK3S_64 case KVM_CAP_SPAPR_TCE: + fallthrough; case KVM_CAP_SPAPR_TCE_64: - r = 1; - break; case KVM_CAP_SPAPR_TCE_VFIO: - r = !!cpu_has_feature(CPU_FTR_HVMODE); - break; case KVM_CAP_PPC_RTAS: case KVM_CAP_PPC_FIXUP_HCALL: case KVM_CAP_PPC_ENABLE_HCALL: #ifdef CONFIG_KVM_XICS case KVM_CAP_IRQ_XICS: #endif case KVM_CAP_PPC_GET_CPU_CHAR: r = 1; break; #ifdef CONFIG_KVM_XIVE base-commit: 6d61a53dd6f55405ebcaea6ee38d1ab5a8856c2c -- 2.48.1

8 months, 4 weeks

2
2
0 0

2025

2024

2023

2022

2021

2020

2019

2018

2017

Linux-stable-mirror