Linaro-mm-sig

linaro-mm-sig@lists.linaro.org

8 participants
2915 discussions

Start a nNew thread

ACCOUNT WARNING: Undelivered mails linaro-mm-sig@lists.linaro.org Kindly rectify

by Noreply＠lists.linaro.org

3 years, 4 months

[RFC v2 0/6] Proposal for a GPU cgroup controller

by T.J. Mercier

This patch series revisits the proposal for a GPU cgroup controller to track and limit memory allocations by various device/allocator subsystems. The patch series also contains a simple prototype to illustrate how Android intends to implement DMA-BUF allocator attribution using the GPU cgroup controller. The prototype does not include resource limit enforcements. Changelog: v2: See the previous revision of this change submitted by Hridya Valsaraju at: https://lore.kernel.org/all/20220115010622.3185921-1-hridya@google.com/ Move dma-buf cgroup charge transfer from a dma_buf_op defined by every heap to a single dma-buf function for all heaps per Daniel Vetter and Christian König. Pointers to struct gpucg and struct gpucg_device tracking the current associations were added to the dma_buf struct to achieve this. Fix incorrect Kconfig help section indentation per Randy Dunlap. History of the GPU cgroup controller ==================================== The GPU/DRM cgroup controller came into being when a consensus[1] was reached that the resources it tracked were unsuitable to be integrated into memcg. Originally, the proposed controller was specific to the DRM subsystem and was intended to track GEM buffers and GPU-specific resources[2]. In order to help establish a unified memory accounting model for all GPU and all related subsystems, Daniel Vetter put forth a suggestion to move it out of the DRM subsystem so that it can be used by other DMA-BUF exporters as well[3]. This RFC proposes an interface that does the same. [1]: https://patchwork.kernel.org/project/dri-devel/cover/20190501140438.9506-1-… [2]: https://lore.kernel.org/amd-gfx/20210126214626.16260-1-brian.welty@intel.co… [3]: https://lore.kernel.org/amd-gfx/YCVOl8%2F87bqRSQei@phenom.ffwll.local/ T.J. Mercier (6): gpu: rfc: Proposal for a GPU cgroup controller cgroup: gpu: Add a cgroup controller for allocator attribution of GPU memory dmabuf: Use the GPU cgroup charge/uncharge APIs dmabuf: heaps: export system_heap buffers with GPU cgroup charging dmabuf: Add gpu cgroup charge transfer function android: binder: Add a buffer flag to relinquish ownership of fds Documentation/gpu/rfc/gpu-cgroup.rst | 195 +++++++++++++++++ Documentation/gpu/rfc/index.rst | 4 + drivers/android/binder.c | 26 +++ drivers/dma-buf/dma-buf.c | 100 +++++++++ drivers/dma-buf/dma-heap.c | 27 +++ drivers/dma-buf/heaps/system_heap.c | 3 + include/linux/cgroup_gpu.h | 127 +++++++++++ include/linux/cgroup_subsys.h | 4 + include/linux/dma-buf.h | 22 +- include/linux/dma-heap.h | 11 + include/uapi/linux/android/binder.h | 1 + init/Kconfig | 7 + kernel/cgroup/Makefile | 1 + kernel/cgroup/gpu.c | 304 +++++++++++++++++++++++++++ 14 files changed, 830 insertions(+), 2 deletions(-) create mode 100644 Documentation/gpu/rfc/gpu-cgroup.rst create mode 100644 include/linux/cgroup_gpu.h create mode 100644 kernel/cgroup/gpu.c -- 2.35.1.265.g69c8d7142f-goog

3 years, 4 months

completely rework the dma_resv semantic

by Christian König

Hi guys, by now that should be a rather well known set of changes. The basic idea is that instead of the fixed exclusive/shared classes we now attach an usage to each fence in the dma_resv object describing how the operation represented by the fence is using the resources protected by the dma_resv. I've addressed quite a bunch of comments already and I think this set has already been discussed quite well now. As improvement to the last version I've now added CCs for all the relevant maintainers to the patches changing some functionality inside drivers. Please review and comment, Christian.

3 years, 4 months

[PATCH] drm/i915/gt: fix unsigned integer to signed assignment

by Jiapeng Chong

Eliminate the follow smatch warning: drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c:4640 guc_create_virtual() warn: assigning (-2) to unsigned variable 've->base.instance'. drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c:4641 guc_create_virtual() warn: assigning (-2) to unsigned variable 've->base.uabi_instance'. Reported-by: Abaci Robot <abaci(a)linux.alibaba.com> Signed-off-by: Jiapeng Chong <jiapeng.chong(a)linux.alibaba.com> --- drivers/gpu/drm/i915/gt/intel_engine_types.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h index 36365bdbe1ee..dc7cc06c68e7 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_types.h +++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h @@ -328,10 +328,10 @@ struct intel_engine_cs { intel_engine_mask_t logical_mask; u8 class; - u8 instance; + s8 instance; u16 uabi_class; - u16 uabi_instance; + s16 uabi_instance; u32 uabi_capabilities; u32 context_size; -- 2.20.1.7.g153144c

3 years, 4 months

[PATCH v3] kfence: Make test case compatible with run time set sample interval

by Peng Liu

The parameter kfence_sample_interval can be set via boot parameter and late shell command, which is convenient for automated tests and KFENCE parameter optimization. However, KFENCE test case just uses compile-time CONFIG_KFENCE_SAMPLE_INTERVAL, which will make KFENCE test case not run as users desired. Export kfence_sample_interval, so that KFENCE test case can use run-time-set sample interval. Signed-off-by: Peng Liu <liupeng256(a)huawei.com> --- v2->v3: - Revise change log description v1->v2: - Use EXPORT_SYMBOL_GPL replace EXPORT_SYMBOL include/linux/kfence.h | 2 ++ mm/kfence/core.c | 3 ++- mm/kfence/kfence_test.c | 8 ++++---- 3 files changed, 8 insertions(+), 5 deletions(-) diff --git a/include/linux/kfence.h b/include/linux/kfence.h index 4b5e3679a72c..f49e64222628 100644 --- a/include/linux/kfence.h +++ b/include/linux/kfence.h @@ -17,6 +17,8 @@ #include <linux/atomic.h> #include <linux/static_key.h> +extern unsigned long kfence_sample_interval; + /* * We allocate an even number of pages, as it simplifies calculations to map * address to metadata indices; effectively, the very first page serves as an diff --git a/mm/kfence/core.c b/mm/kfence/core.c index 5ad40e3add45..13128fa13062 100644 --- a/mm/kfence/core.c +++ b/mm/kfence/core.c @@ -47,7 +47,8 @@ static bool kfence_enabled __read_mostly; -static unsigned long kfence_sample_interval __read_mostly = CONFIG_KFENCE_SAMPLE_INTERVAL; +unsigned long kfence_sample_interval __read_mostly = CONFIG_KFENCE_SAMPLE_INTERVAL; +EXPORT_SYMBOL_GPL(kfence_sample_interval); /* Export for test modules. */ #ifdef MODULE_PARAM_PREFIX #undef MODULE_PARAM_PREFIX diff --git a/mm/kfence/kfence_test.c b/mm/kfence/kfence_test.c index a22b1af85577..50dbb815a2a8 100644 --- a/mm/kfence/kfence_test.c +++ b/mm/kfence/kfence_test.c @@ -268,13 +268,13 @@ static void *test_alloc(struct kunit *test, size_t size, gfp_t gfp, enum allocat * 100x the sample interval should be more than enough to ensure we get * a KFENCE allocation eventually. */ - timeout = jiffies + msecs_to_jiffies(100 * CONFIG_KFENCE_SAMPLE_INTERVAL); + timeout = jiffies + msecs_to_jiffies(100 * kfence_sample_interval); /* * Especially for non-preemption kernels, ensure the allocation-gate * timer can catch up: after @resched_after, every failed allocation * attempt yields, to ensure the allocation-gate timer is scheduled. */ - resched_after = jiffies + msecs_to_jiffies(CONFIG_KFENCE_SAMPLE_INTERVAL); + resched_after = jiffies + msecs_to_jiffies(kfence_sample_interval); do { if (test_cache) alloc = kmem_cache_alloc(test_cache, gfp); @@ -608,7 +608,7 @@ static void test_gfpzero(struct kunit *test) int i; /* Skip if we think it'd take too long. */ - KFENCE_TEST_REQUIRES(test, CONFIG_KFENCE_SAMPLE_INTERVAL <= 100); + KFENCE_TEST_REQUIRES(test, kfence_sample_interval <= 100); setup_test_cache(test, size, 0, NULL); buf1 = test_alloc(test, size, GFP_KERNEL, ALLOCATE_ANY); @@ -739,7 +739,7 @@ static void test_memcache_alloc_bulk(struct kunit *test) * 100x the sample interval should be more than enough to ensure we get * a KFENCE allocation eventually. */ - timeout = jiffies + msecs_to_jiffies(100 * CONFIG_KFENCE_SAMPLE_INTERVAL); + timeout = jiffies + msecs_to_jiffies(100 * kfence_sample_interval); do { void *objects[100]; int i, num = kmem_cache_alloc_bulk(test_cache, GFP_ATOMIC, ARRAY_SIZE(objects), -- 2.18.0.huawei.25

3 years, 4 months

[PATCH v2] kfence: Make test case compatible with run time set sample interval

by Peng Liu

The parameter kfence_sample_interval can be set via boot parameter and late shell command, which is convenient for automatical tests and KFENCE parameter optimation. However, KFENCE test case just use compile time CONFIG_KFENCE_SAMPLE_INTERVAL, this will make KFENCE test case not run as user desired. This patch will make KFENCE test case compatible with run-time-set sample interval. v1->v2: - Use EXPORT_SYMBOL_GPL replace EXPORT_SYMBOL Signed-off-by: Peng Liu <liupeng256(a)huawei.com> --- include/linux/kfence.h | 2 ++ mm/kfence/core.c | 3 ++- mm/kfence/kfence_test.c | 8 ++++---- 3 files changed, 8 insertions(+), 5 deletions(-) diff --git a/include/linux/kfence.h b/include/linux/kfence.h index 4b5e3679a72c..f49e64222628 100644 --- a/include/linux/kfence.h +++ b/include/linux/kfence.h @@ -17,6 +17,8 @@ #include <linux/atomic.h> #include <linux/static_key.h> +extern unsigned long kfence_sample_interval; + /* * We allocate an even number of pages, as it simplifies calculations to map * address to metadata indices; effectively, the very first page serves as an diff --git a/mm/kfence/core.c b/mm/kfence/core.c index 5ad40e3add45..13128fa13062 100644 --- a/mm/kfence/core.c +++ b/mm/kfence/core.c @@ -47,7 +47,8 @@ static bool kfence_enabled __read_mostly; -static unsigned long kfence_sample_interval __read_mostly = CONFIG_KFENCE_SAMPLE_INTERVAL; +unsigned long kfence_sample_interval __read_mostly = CONFIG_KFENCE_SAMPLE_INTERVAL; +EXPORT_SYMBOL_GPL(kfence_sample_interval); /* Export for test modules. */ #ifdef MODULE_PARAM_PREFIX #undef MODULE_PARAM_PREFIX diff --git a/mm/kfence/kfence_test.c b/mm/kfence/kfence_test.c index a22b1af85577..50dbb815a2a8 100644 --- a/mm/kfence/kfence_test.c +++ b/mm/kfence/kfence_test.c @@ -268,13 +268,13 @@ static void *test_alloc(struct kunit *test, size_t size, gfp_t gfp, enum allocat * 100x the sample interval should be more than enough to ensure we get * a KFENCE allocation eventually. */ - timeout = jiffies + msecs_to_jiffies(100 * CONFIG_KFENCE_SAMPLE_INTERVAL); + timeout = jiffies + msecs_to_jiffies(100 * kfence_sample_interval); /* * Especially for non-preemption kernels, ensure the allocation-gate * timer can catch up: after @resched_after, every failed allocation * attempt yields, to ensure the allocation-gate timer is scheduled. */ - resched_after = jiffies + msecs_to_jiffies(CONFIG_KFENCE_SAMPLE_INTERVAL); + resched_after = jiffies + msecs_to_jiffies(kfence_sample_interval); do { if (test_cache) alloc = kmem_cache_alloc(test_cache, gfp); @@ -608,7 +608,7 @@ static void test_gfpzero(struct kunit *test) int i; /* Skip if we think it'd take too long. */ - KFENCE_TEST_REQUIRES(test, CONFIG_KFENCE_SAMPLE_INTERVAL <= 100); + KFENCE_TEST_REQUIRES(test, kfence_sample_interval <= 100); setup_test_cache(test, size, 0, NULL); buf1 = test_alloc(test, size, GFP_KERNEL, ALLOCATE_ANY); @@ -739,7 +739,7 @@ static void test_memcache_alloc_bulk(struct kunit *test) * 100x the sample interval should be more than enough to ensure we get * a KFENCE allocation eventually. */ - timeout = jiffies + msecs_to_jiffies(100 * CONFIG_KFENCE_SAMPLE_INTERVAL); + timeout = jiffies + msecs_to_jiffies(100 * kfence_sample_interval); do { void *objects[100]; int i, num = kmem_cache_alloc_bulk(test_cache, GFP_ATOMIC, ARRAY_SIZE(objects), -- 2.18.0.huawei.25

3 years, 4 months

[PATCH] dma-buf: heaps: Fix potential spectre v1 gadget

by Jordy Zomer

It appears like nr could be a Spectre v1 gadget as it's supplied by a user and used as an array index. Prevent the contents of kernel memory from being leaked to userspace via speculative execution by using array_index_nospec. Signed-off-by: Jordy Zomer <jordy(a)pwning.systems> --- drivers/dma-buf/dma-heap.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/drivers/dma-buf/dma-heap.c b/drivers/dma-buf/dma-heap.c index 56bf5ad01ad5..8f5848aa144f 100644 --- a/drivers/dma-buf/dma-heap.c +++ b/drivers/dma-buf/dma-heap.c @@ -14,6 +14,7 @@ #include <linux/xarray.h> #include <linux/list.h> #include <linux/slab.h> +#include <linux/nospec.h> #include <linux/uaccess.h> #include <linux/syscalls.h> #include <linux/dma-heap.h> @@ -135,6 +136,7 @@ static long dma_heap_ioctl(struct file *file, unsigned int ucmd, if (nr >= ARRAY_SIZE(dma_heap_ioctl_cmds)) return -EINVAL; + nr = array_index_nospec(nr, ARRAY_SIZE(dma_heap_ioctl_cmds)); /* Get the kernel ioctl cmd that matches */ kcmd = dma_heap_ioctl_cmds[nr]; -- 2.27.0

3 years, 5 months

[PATCH] dma-resv: some doc polish for iterators

by Daniel Vetter

Hammer it a bit more in that iterators can be restarted and when that matters, plus suggest to prefer the locked version whenver. Also delete the two leftover kerneldoc for static functions plus sprinkle some more links while at it. Signed-off-by: Daniel Vetter <daniel.vetter(a)intel.com> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: "Christian König" <christian.koenig(a)amd.com> Cc: linux-media(a)vger.kernel.org Cc: linaro-mm-sig(a)lists.linaro.org --- drivers/dma-buf/dma-resv.c | 26 ++++++++++++-------------- include/linux/dma-resv.h | 13 ++++++++++++- 2 files changed, 24 insertions(+), 15 deletions(-) diff --git a/drivers/dma-buf/dma-resv.c b/drivers/dma-buf/dma-resv.c index 9eb2baa387d4..1453b664c405 100644 --- a/drivers/dma-buf/dma-resv.c +++ b/drivers/dma-buf/dma-resv.c @@ -323,12 +323,6 @@ void dma_resv_add_excl_fence(struct dma_resv *obj, struct dma_fence *fence) } EXPORT_SYMBOL(dma_resv_add_excl_fence); -/** - * dma_resv_iter_restart_unlocked - restart the unlocked iterator - * @cursor: The dma_resv_iter object to restart - * - * Restart the unlocked iteration by initializing the cursor object. - */ static void dma_resv_iter_restart_unlocked(struct dma_resv_iter *cursor) { cursor->seq = read_seqcount_begin(&cursor->obj->seq); @@ -344,14 +338,6 @@ static void dma_resv_iter_restart_unlocked(struct dma_resv_iter *cursor) cursor->is_restarted = true; } -/** - * dma_resv_iter_walk_unlocked - walk over fences in a dma_resv obj - * @cursor: cursor to record the current position - * - * Return all the fences in the dma_resv object which are not yet signaled. - * The returned fence has an extra local reference so will stay alive. - * If a concurrent modify is detected the whole iteration is started over again. - */ static void dma_resv_iter_walk_unlocked(struct dma_resv_iter *cursor) { struct dma_resv *obj = cursor->obj; @@ -387,6 +373,12 @@ static void dma_resv_iter_walk_unlocked(struct dma_resv_iter *cursor) * dma_resv_iter_first_unlocked - first fence in an unlocked dma_resv obj. * @cursor: the cursor with the current position * + * Subsequent fences are iterated with dma_resv_iter_next_unlocked(). + * + * Beware that the iterator can be restarted. Code which accumulates statistics + * or similar needs to check for this with dma_resv_iter_is_restarted(). For + * this reason prefer the locked dma_resv_iter_first() whenver possible. + * * Returns the first fence from an unlocked dma_resv obj. */ struct dma_fence *dma_resv_iter_first_unlocked(struct dma_resv_iter *cursor) @@ -406,6 +398,10 @@ EXPORT_SYMBOL(dma_resv_iter_first_unlocked); * dma_resv_iter_next_unlocked - next fence in an unlocked dma_resv obj. * @cursor: the cursor with the current position * + * Beware that the iterator can be restarted. Code which accumulates statistics + * or similar needs to check for this with dma_resv_iter_is_restarted(). For + * this reason prefer the locked dma_resv_iter_next() whenver possible. + * * Returns the next fence from an unlocked dma_resv obj. */ struct dma_fence *dma_resv_iter_next_unlocked(struct dma_resv_iter *cursor) @@ -431,6 +427,8 @@ EXPORT_SYMBOL(dma_resv_iter_next_unlocked); * dma_resv_iter_first - first fence from a locked dma_resv object * @cursor: cursor to record the current position * + * Subsequent fences are iterated with dma_resv_iter_next_unlocked(). + * * Return the first fence in the dma_resv object while holding the * &dma_resv.lock. */ diff --git a/include/linux/dma-resv.h b/include/linux/dma-resv.h index dbd235ab447f..ebe908592ac3 100644 --- a/include/linux/dma-resv.h +++ b/include/linux/dma-resv.h @@ -153,6 +153,13 @@ struct dma_resv { * struct dma_resv_iter - current position into the dma_resv fences * * Don't touch this directly in the driver, use the accessor function instead. + * + * IMPORTANT + * + * When using the lockless iterators like dma_resv_iter_next_unlocked() or + * dma_resv_for_each_fence_unlocked() beware that the iterator can be restarted. + * Code which accumulates statistics or similar needs to check for this with + * dma_resv_iter_is_restarted(). */ struct dma_resv_iter { /** @obj: The dma_resv object we iterate over */ @@ -243,7 +250,11 @@ static inline bool dma_resv_iter_is_restarted(struct dma_resv_iter *cursor) * &dma_resv.lock and using RCU instead. The cursor needs to be initialized * with dma_resv_iter_begin() and cleaned up with dma_resv_iter_end(). Inside * the iterator a reference to the dma_fence is held and the RCU lock dropped. - * When the dma_resv is modified the iteration starts over again. + * + * Beware that the iterator can be restarted when the struct dma_resv for + * @cursor is modified. Code which accumulates statistics or similar needs to + * check for this with dma_resv_iter_is_restarted(). For this reason prefer the + * lock iterator dma_resv_for_each_fence() whenever possible. */ #define dma_resv_for_each_fence_unlocked(cursor, fence) \ for (fence = dma_resv_iter_first_unlocked(cursor); \ -- 2.33.0

3 years, 5 months

[PATCH 1/9] dma-buf: consolidate dma_fence subclass checking

by Christian König

Consolidate the wrapper functions to check for dma_fence subclasses in the dma_fence header. This makes it easier to document and also check the different requirements for fence containers in the subclasses. Signed-off-by: Christian König <christian.koenig(a)amd.com> --- include/linux/dma-fence-array.h | 15 +------------ include/linux/dma-fence-chain.h | 3 +-- include/linux/dma-fence.h | 38 +++++++++++++++++++++++++++++++++ 3 files changed, 40 insertions(+), 16 deletions(-) diff --git a/include/linux/dma-fence-array.h b/include/linux/dma-fence-array.h index 303dd712220f..fec374f69e12 100644 --- a/include/linux/dma-fence-array.h +++ b/include/linux/dma-fence-array.h @@ -45,19 +45,6 @@ struct dma_fence_array { struct irq_work work; }; -extern const struct dma_fence_ops dma_fence_array_ops; - -/** - * dma_fence_is_array - check if a fence is from the array subsclass - * @fence: fence to test - * - * Return true if it is a dma_fence_array and false otherwise. - */ -static inline bool dma_fence_is_array(struct dma_fence *fence) -{ - return fence->ops == &dma_fence_array_ops; -} - /** * to_dma_fence_array - cast a fence to a dma_fence_array * @fence: fence to cast to a dma_fence_array @@ -68,7 +55,7 @@ static inline bool dma_fence_is_array(struct dma_fence *fence) static inline struct dma_fence_array * to_dma_fence_array(struct dma_fence *fence) { - if (fence->ops != &dma_fence_array_ops) + if (!fence || !dma_fence_is_array(fence)) return NULL; return container_of(fence, struct dma_fence_array, base); diff --git a/include/linux/dma-fence-chain.h b/include/linux/dma-fence-chain.h index 54fe3443fd2c..ee906b659694 100644 --- a/include/linux/dma-fence-chain.h +++ b/include/linux/dma-fence-chain.h @@ -49,7 +49,6 @@ struct dma_fence_chain { spinlock_t lock; }; -extern const struct dma_fence_ops dma_fence_chain_ops; /** * to_dma_fence_chain - cast a fence to a dma_fence_chain @@ -61,7 +60,7 @@ extern const struct dma_fence_ops dma_fence_chain_ops; static inline struct dma_fence_chain * to_dma_fence_chain(struct dma_fence *fence) { - if (!fence || fence->ops != &dma_fence_chain_ops) + if (!fence || !dma_fence_is_chain(fence)) return NULL; return container_of(fence, struct dma_fence_chain, base); diff --git a/include/linux/dma-fence.h b/include/linux/dma-fence.h index 1ea691753bd3..775cdc0b4f24 100644 --- a/include/linux/dma-fence.h +++ b/include/linux/dma-fence.h @@ -587,4 +587,42 @@ struct dma_fence *dma_fence_get_stub(void); struct dma_fence *dma_fence_allocate_private_stub(void); u64 dma_fence_context_alloc(unsigned num); +extern const struct dma_fence_ops dma_fence_array_ops; +extern const struct dma_fence_ops dma_fence_chain_ops; + +/** + * dma_fence_is_array - check if a fence is from the array subclass + * @fence: the fence to test + * + * Return true if it is a dma_fence_array and false otherwise. + */ +static inline bool dma_fence_is_array(struct dma_fence *fence) +{ + return fence->ops == &dma_fence_array_ops; +} + +/** + * dma_fence_is_chain - check if a fence is from the chain subclass + * @fence: the fence to test + * + * Return true if it is a dma_fence_chain and false otherwise. + */ +static inline bool dma_fence_is_chain(struct dma_fence *fence) +{ + return fence->ops == &dma_fence_chain_ops; +} + +/** + * dma_fence_is_container - check if a fence is a container for other fences + * @fence: the fence to test + * + * Return true if this fence is a container for other fences, false otherwise. + * This is important since we can't build up large fence structure or otherwise + * we run into recursion during operation on those fences. + */ +static inline bool dma_fence_is_container(struct dma_fence *fence) +{ + return dma_fence_is_array(fence) || dma_fence_is_chain(fence); +} + #endif /* __LINUX_DMA_FENCE_H */ -- 2.25.1

3 years, 5 months

[PATCH 00/19] drm/i915/guc: Refactor ADS access to use dma_buf_map

by Lucas De Marchi

While porting i915 to arm64 we noticed some issues accessing lmem. Some writes were getting corrupted and the final state of the buffer didn't have exactly what we wrote. This became evident when enabling GuC submission: depending on the number of engines the ADS struct was being corrupted and GuC would reject it, refusin to initialize. From Documentation/core-api/bus-virt-phys-mapping.rst: This memory is called "PCI memory" or "shared memory" or "IO memory" or whatever, and there is only one way to access it: the readb/writeb and related functions. You should never take the address of such memory, because there is really nothing you can do with such an address: it's not conceptually in the same memory space as "real memory" at all, so you cannot just dereference a pointer. (Sadly, on x86 it **is** in the same memory space, so on x86 it actually works to just deference a pointer, but it's not portable). When reading or writing words directly to IO memory, in order to be portable the Linux kernel provides the abstraction detailed in section "Differences between I/O access functions" of Documentation/driver-api/device-io.rst. This limits our ability to simply overlay our structs on top a buffer and directly access it since that buffer may come from IO memory rather than system memory. Hence the approach taken in intel_guc_ads.c needs to be refactored. This is not the only place in i915 that neeed to be changed, but the one causing the most problems, with a real reproducer. This first set of patch focuses on fixing the gem object to pass the ADS After the addition of a few helpers in the dma_buf_map API, most of intel_guc_ads.c can be converted to use it. The exception is the regset initialization: we'd incur into a lot of extra indirection when reading/writting each register. So the regset is converted to use a temporary buffer allocated on probe, which is then copied to its final location when finishing the initialization or on gt reset. Testing on some discrete cards, after this change we can correctly pass the ADS struct to GuC and have it initialized correctly. thanks Lucas De Marchi Cc: linux-media(a)vger.kernel.org Cc: dri-devel(a)lists.freedesktop.org Cc: linaro-mm-sig(a)lists.linaro.org Cc: linux-kernel(a)vger.kernel.org Cc: Christian König <christian.koenig(a)amd.com> Cc: Daniel Vetter <daniel(a)ffwll.ch> Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio(a)intel.com> Cc: David Airlie <airlied(a)linux.ie> Cc: John Harrison <John.C.Harrison(a)Intel.com> Cc: Joonas Lahtinen <joonas.lahtinen(a)linux.intel.com> Cc: Maarten Lankhorst <maarten.lankhorst(a)linux.intel.com> Cc: Matt Roper <matthew.d.roper(a)intel.com> Cc: Matthew Auld <matthew.auld(a)intel.com> Cc: Matthew Brost <matthew.brost(a)intel.com> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: Thomas Hellström <thomas.hellstrom(a)linux.intel.com> Cc: Tvrtko Ursulin <tvrtko.ursulin(a)linux.intel.com> Lucas De Marchi (19): dma-buf-map: Add read/write helpers dma-buf-map: Add helper to initialize second map drm/i915/gt: Add helper for shmem copy to dma_buf_map drm/i915/guc: Keep dma_buf_map of ads_blob around drm/i915/guc: Add read/write helpers for ADS blob drm/i915/guc: Convert golden context init to dma_buf_map drm/i915/guc: Convert policies update to dma_buf_map drm/i915/guc: Convert engine record to dma_buf_map dma-buf-map: Add wrapper over memset drm/i915/guc: Convert guc_ads_private_data_reset to dma_buf_map drm/i915/guc: Convert golden context prep to dma_buf_map drm/i915/guc: Replace check for golden context size drm/i915/guc: Convert mapping table to dma_buf_map drm/i915/guc: Convert capture list to dma_buf_map drm/i915/guc: Prepare for error propagation drm/i915/guc: Use a single pass to calculate regset drm/i915/guc: Convert guc_mmio_reg_state_init to dma_buf_map drm/i915/guc: Convert __guc_ads_init to dma_buf_map drm/i915/guc: Remove plain ads_blob pointer drivers/gpu/drm/i915/gt/shmem_utils.c | 32 ++ drivers/gpu/drm/i915/gt/shmem_utils.h | 3 + drivers/gpu/drm/i915/gt/uc/intel_guc.h | 14 +- drivers/gpu/drm/i915/gt/uc/intel_guc_ads.c | 374 +++++++++++------- drivers/gpu/drm/i915/gt/uc/intel_guc_ads.h | 3 +- .../gpu/drm/i915/gt/uc/intel_guc_submission.c | 11 +- include/linux/dma-buf-map.h | 127 ++++++ 7 files changed, 405 insertions(+), 159 deletions(-) -- 2.35.0

3 years, 5 months

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-mm-sig