From 863a8eb3f27098b42772f668e3977ff4cae10b04 Mon Sep 17 00:00:00 2001 From: "Matthew Wilcox (Oracle)" Date: Tue, 19 Sep 2023 20:48:55 +0100 Subject: i915: Limit the length of an sg list to the requested length The folio conversion changed the behaviour of shmem_sg_alloc_table() to put the entire length of the last folio into the sg list, even if the sg list should have been shorter. gen8_ggtt_insert_entries() relied on the list being the right length and would overrun the end of the page tables. Other functions may also have been affected. Clamp the length of the last entry in the sg list to be the expected length. Signed-off-by: Matthew Wilcox (Oracle) Fixes: 0b62af28f249 ("i915: convert shmem_sg_free_table() to use a folio_batch") Cc: stable@vger.kernel.org # 6.5.x Link: https://gitlab.freedesktop.org/drm/intel/-/issues/9256 Link: https://lore.kernel.org/lkml/6287208.lOV4Wx5bFT@natalenko.name/ Reported-by: Oleksandr Natalenko Tested-by: Oleksandr Natalenko Reviewed-by: Andrzej Hajda Signed-off-by: Andrzej Hajda Link: https://patchwork.freedesktop.org/patch/msgid/20230919194855.347582-1-willy@infradead.org (cherry picked from commit 26a8e32e6d77900819c0c730fbfb393692dbbeea) Signed-off-by: Rodrigo Vivi --- drivers/gpu/drm/i915/gem/i915_gem_shmem.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c index 8f1633c3fb93..73a4a4eb29e0 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c @@ -100,6 +100,7 @@ int shmem_sg_alloc_table(struct drm_i915_private *i915, struct sg_table *st, st->nents = 0; for (i = 0; i < page_count; i++) { struct folio *folio; + unsigned long nr_pages; const unsigned int shrink[] = { I915_SHRINK_BOUND | I915_SHRINK_UNBOUND, 0, @@ -150,6 +151,8 @@ int shmem_sg_alloc_table(struct drm_i915_private *i915, struct sg_table *st, } } while (1); + nr_pages = min_t(unsigned long, + folio_nr_pages(folio), page_count - i); if (!i || sg->length >= max_segment || folio_pfn(folio) != next_pfn) { @@ -157,13 +160,13 @@ int shmem_sg_alloc_table(struct drm_i915_private *i915, struct sg_table *st, sg = sg_next(sg); st->nents++; - sg_set_folio(sg, folio, folio_size(folio), 0); + sg_set_folio(sg, folio, nr_pages * PAGE_SIZE, 0); } else { /* XXX: could overflow? */ - sg->length += folio_size(folio); + sg->length += nr_pages * PAGE_SIZE; } - next_pfn = folio_pfn(folio) + folio_nr_pages(folio); - i += folio_nr_pages(folio) - 1; + next_pfn = folio_pfn(folio) + nr_pages; + i += nr_pages - 1; /* Check that the i965g/gm workaround works. */ GEM_BUG_ON(gfp & __GFP_DMA32 && next_pfn >= 0x00100000UL); -- cgit v1.2.3 From b7599d241778d0b10cdf7a5c755aa7db9b83250c Mon Sep 17 00:00:00 2001 From: Javier Pello Date: Sat, 2 Sep 2023 17:10:39 +0200 Subject: drm/i915/gt: Fix reservation address in ggtt_reserve_guc_top There is an assertion in ggtt_reserve_guc_top that the global GTT is of size at least GUC_GGTT_TOP, which is not the case on a 32-bit platform; see commit 562d55d991b39ce376c492df2f7890fd6a541ffc ("drm/i915/bdw: Only use 2g GGTT for 32b platforms"). If GEM_BUG_ON is enabled, this triggers a BUG(); if GEM_BUG_ON is disabled, the subsequent reservation fails and the driver fails to initialise the device: i915 0000:00:02.0: [drm:i915_init_ggtt [i915]] Failed to reserve top of GGTT for GuC i915 0000:00:02.0: Device initialization failed (-28) i915 0000:00:02.0: Please file a bug on drm/i915; see https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs for details. i915: probe of 0000:00:02.0 failed with error -28 Make the reservation at the top of the available space, whatever that is, instead of assuming that the top will be GUC_GGTT_TOP. Fixes: 911800765ef6 ("drm/i915/uc: Reserve upper range of GGTT") Link: https://gitlab.freedesktop.org/drm/intel/-/issues/9080 Signed-off-by: Javier Pello Reviewed-by: Daniele Ceraolo Spurio Cc: Fernando Pacheco Cc: Chris Wilson Cc: Jani Nikula Cc: Joonas Lahtinen Cc: Rodrigo Vivi Cc: Tvrtko Ursulin Cc: intel-gfx@lists.freedesktop.org Cc: stable@vger.kernel.org # v5.3+ Signed-off-by: John Harrison Link: https://patchwork.freedesktop.org/patch/msgid/20230902171039.2229126186d697dbcf62d6d8@otheo.eu (cherry picked from commit 0f3fa942d91165c2702577e9274d2ee1c7212afc) Signed-off-by: Rodrigo Vivi --- drivers/gpu/drm/i915/gt/intel_ggtt.c | 23 +++++++++++++++++------ 1 file changed, 17 insertions(+), 6 deletions(-) diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c index dd0ed941441a..da21f2786b5d 100644 --- a/drivers/gpu/drm/i915/gt/intel_ggtt.c +++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c @@ -511,20 +511,31 @@ void intel_ggtt_unbind_vma(struct i915_address_space *vm, vm->clear_range(vm, vma_res->start, vma_res->vma_size); } +/* + * Reserve the top of the GuC address space for firmware images. Addresses + * beyond GUC_GGTT_TOP in the GuC address space are inaccessible by GuC, + * which makes for a suitable range to hold GuC/HuC firmware images if the + * size of the GGTT is 4G. However, on a 32-bit platform the size of the GGTT + * is limited to 2G, which is less than GUC_GGTT_TOP, but we reserve a chunk + * of the same size anyway, which is far more than needed, to keep the logic + * in uc_fw_ggtt_offset() simple. + */ +#define GUC_TOP_RESERVE_SIZE (SZ_4G - GUC_GGTT_TOP) + static int ggtt_reserve_guc_top(struct i915_ggtt *ggtt) { - u64 size; + u64 offset; int ret; if (!intel_uc_uses_guc(&ggtt->vm.gt->uc)) return 0; - GEM_BUG_ON(ggtt->vm.total <= GUC_GGTT_TOP); - size = ggtt->vm.total - GUC_GGTT_TOP; + GEM_BUG_ON(ggtt->vm.total <= GUC_TOP_RESERVE_SIZE); + offset = ggtt->vm.total - GUC_TOP_RESERVE_SIZE; - ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, &ggtt->uc_fw, size, - GUC_GGTT_TOP, I915_COLOR_UNEVICTABLE, - PIN_NOEVICT); + ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, &ggtt->uc_fw, + GUC_TOP_RESERVE_SIZE, offset, + I915_COLOR_UNEVICTABLE, PIN_NOEVICT); if (ret) drm_dbg(&ggtt->vm.i915->drm, "Failed to reserve top of GGTT for GuC\n"); -- cgit v1.2.3 From 907ef0398c938be8232b77c61cfcf50fbfd95554 Mon Sep 17 00:00:00 2001 From: Umesh Nerlige Ramappa Date: Mon, 25 Sep 2023 12:21:17 -0700 Subject: i915/guc: Get runtime pm in busyness worker only if already active Ideally the busyness worker should take a gt pm wakeref because the worker only needs to be active while gt is awake. However, the gt_park path cancels the worker synchronously and this complicates the flow if the worker is also running at the same time. The cancel waits for the worker and when the worker releases the wakeref, that would call gt_park and would lead to a deadlock. The resolution is to take the global pm wakeref if runtime pm is already active. If not, we don't need to update the busyness stats as the stats would already be updated when the gt was parked. Note: - We do not requeue the worker if we cannot take a reference to runtime pm since intel_guc_busyness_unpark would requeue the worker in the resume path. - If the gt was parked longer than time taken for GT timestamp to roll over, we ignore those rollovers since we don't care about tracking the exact GT time. We only care about roll overs when the gt is active and running workloads. - There is a window of time between gt_park and runtime suspend, where the worker may run. This is acceptable since the worker will not find any new data to update busyness. v2: (Daniele) - Edit commit message and code comment - Use runtime pm in the worker - Put runtime pm after enabling the worker - Use Link tag and add Fixes tag v3: (Daniele) - Reword commit and comments and add details Link: https://gitlab.freedesktop.org/drm/intel/-/issues/7077 Fixes: 77cdd054dd2c ("drm/i915/pmu: Connect engine busyness stats from GuC to pmu") Signed-off-by: Umesh Nerlige Ramappa Reviewed-by: Daniele Ceraolo Spurio Link: https://patchwork.freedesktop.org/patch/msgid/20230925192117.2497058-1-umesh.nerlige.ramappa@intel.com (cherry picked from commit e2f99b79d4c594cdf7ab449e338d4947f5ea8903) Signed-off-by: Rodrigo Vivi --- drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 38 +++++++++++++++++++++-- 1 file changed, 35 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c index b5b7f2fe8c78..dc7b40e06e38 100644 --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c @@ -1432,6 +1432,36 @@ static void guc_timestamp_ping(struct work_struct *wrk) unsigned long index; int srcu, ret; + /* + * Ideally the busyness worker should take a gt pm wakeref because the + * worker only needs to be active while gt is awake. However, the + * gt_park path cancels the worker synchronously and this complicates + * the flow if the worker is also running at the same time. The cancel + * waits for the worker and when the worker releases the wakeref, that + * would call gt_park and would lead to a deadlock. + * + * The resolution is to take the global pm wakeref if runtime pm is + * already active. If not, we don't need to update the busyness stats as + * the stats would already be updated when the gt was parked. + * + * Note: + * - We do not requeue the worker if we cannot take a reference to runtime + * pm since intel_guc_busyness_unpark would requeue the worker in the + * resume path. + * + * - If the gt was parked longer than time taken for GT timestamp to roll + * over, we ignore those rollovers since we don't care about tracking + * the exact GT time. We only care about roll overs when the gt is + * active and running workloads. + * + * - There is a window of time between gt_park and runtime suspend, + * where the worker may run. This is acceptable since the worker will + * not find any new data to update busyness. + */ + wakeref = intel_runtime_pm_get_if_active(>->i915->runtime_pm); + if (!wakeref) + return; + /* * Synchronize with gt reset to make sure the worker does not * corrupt the engine/guc stats. NB: can't actually block waiting @@ -1440,10 +1470,9 @@ static void guc_timestamp_ping(struct work_struct *wrk) */ ret = intel_gt_reset_trylock(gt, &srcu); if (ret) - return; + goto err_trylock; - with_intel_runtime_pm(>->i915->runtime_pm, wakeref) - __update_guc_busyness_stats(guc); + __update_guc_busyness_stats(guc); /* adjust context stats for overflow */ xa_for_each(&guc->context_lookup, index, ce) @@ -1452,6 +1481,9 @@ static void guc_timestamp_ping(struct work_struct *wrk) intel_gt_reset_unlock(gt, srcu); guc_enable_busyness_worker(guc); + +err_trylock: + intel_runtime_pm_put(>->i915->runtime_pm, wakeref); } static int guc_action_enable_usage_stats(struct intel_guc *guc) -- cgit v1.2.3