From b76c14c8fb2af1e481d51a4eeab8e0c0594824c0 Mon Sep 17 00:00:00 2001 From: Daniele Ceraolo Spurio Date: Tue, 27 Sep 2022 17:41:43 -0700 Subject: drm/i915/huc: better define HuC status getparam possible return values. The current HuC status getparam return values are a bit confusing in regards to what happens in some scenarios. In particular, most of the error cases cause the ioctl to return an error, but a couple of them, INIT_FAIL and LOAD_FAIL, are not explicitly handled and neither is their expected return value documented; these 2 error cases therefore end up into the catch-all umbrella of the "HuC not loaded" case, with this case therefore including both some error scenarios and the load in progress one. The updates included in this patch change the handling so that all error cases behave the same way, i.e. return an errno code, and so that the HuC load in progress case is unambiguous. The patch also includes a small change to the FW init path to make sure we always transition to an error state if something goes wrong. Signed-off-by: Daniele Ceraolo Spurio Cc: Tvrtko Ursulin Cc: Tony Ye Acked-by: Tvrtko Ursulin Acked-by: Tony Ye Reviewed-by: Alan Previn Link: https://patchwork.freedesktop.org/patch/msgid/20220928004145.745803-14-daniele.ceraolospurio@intel.com --- include/uapi/drm/i915_drm.h | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) (limited to 'include/uapi') diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index 520ad2691a99..629198f1d8d8 100644 --- a/include/uapi/drm/i915_drm.h +++ b/include/uapi/drm/i915_drm.h @@ -645,6 +645,22 @@ typedef struct drm_i915_irq_wait { */ #define I915_SCHEDULER_CAP_STATIC_PRIORITY_MAP (1ul << 5) +/* + * Query the status of HuC load. + * + * The query can fail in the following scenarios with the listed error codes: + * -ENODEV if HuC is not present on this platform, + * -EOPNOTSUPP if HuC firmware usage is disabled, + * -ENOPKG if HuC firmware fetch failed, + * -ENOEXEC if HuC firmware is invalid or mismatched, + * -ENOMEM if i915 failed to prepare the FW objects for transfer to the uC, + * -EIO if the FW transfer or the FW authentication failed. + * + * If the IOCTL is successful, the returned parameter will be set to one of the + * following values: + * * 0 if HuC firmware load is not complete, + * * 1 if HuC firmware is authenticated and running. + */ #define I915_PARAM_HUC_STATUS 42 /* Query whether DRM_I915_GEM_EXECBUFFER2 supports the ability to opt-out of -- cgit v1.2.3 From 8133a6daad4e72748e239a02775a853ca7ed798b Mon Sep 17 00:00:00 2001 From: Matthew Auld Date: Tue, 4 Oct 2022 12:49:14 +0100 Subject: drm/i915: enable PS64 support for DG2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit It turns out that on production DG2/ATS HW we should have support for PS64. This feature allows to provide a 64K TLB hint at the PTE level, which is a lot more flexible than the current method of enabling 64K GTT pages for the entire page-table, since that leads to all kinds of annoying restrictions, as documented in: commit caa574ffc4aaf4f29b890223878c63e2e7772f62 Author: Matthew Auld Date: Sat Feb 19 00:17:49 2022 +0530 drm/i915/uapi: document behaviour for DG2 64K support On discrete platforms like DG2, we need to support a minimum page size of 64K when dealing with device local-memory. This is quite tricky for various reasons, so try to document the new implicit uapi for this. With PS64, we can now drop the 2M GTT alignment restriction, and instead only require 64K or larger when dealing with lmem. We still use the compact-pt layout when possible, but only when we are certain that this doesn't interfere with userspace. Note that this is a change in uAPI behaviour, but hopefully shouldn't be a concern (IGT is at least able to autodetect the alignment), since we are only making the GTT alignment constraint less restrictive. Based on a patch from CQ Tang. v2: update the comment wrt scratch page v3: (Nirmoy) - Fix the selftest to actually use the random size, plus some comment improvements, also drop the rem stuff. Reported-by: Michal Mrozek Signed-off-by: Matthew Auld Cc: Lionel Landwerlin Cc: Thomas Hellström Cc: Stuart Summers Cc: Jordan Justen Cc: Yang A Shi Cc: Nirmoy Das Cc: Niranjana Vishwanathapura Reviewed-by: Nirmoy Das Acked-by: Michal Mrozek Link: https://patchwork.freedesktop.org/patch/msgid/20221004114915.221708-1-matthew.auld@intel.com --- drivers/gpu/drm/i915/gem/selftests/huge_pages.c | 157 +++++++++++++++++++++++- drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 81 ++++++------ drivers/gpu/drm/i915/gt/intel_gtt.c | 21 +--- drivers/gpu/drm/i915/gt/intel_gtt.h | 1 + drivers/gpu/drm/i915/i915_drv.h | 7 -- drivers/gpu/drm/i915/i915_pci.c | 2 - drivers/gpu/drm/i915/i915_vma.c | 9 +- drivers/gpu/drm/i915/intel_device_info.h | 1 - drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 9 +- include/uapi/drm/i915_drm.h | 36 ++---- 10 files changed, 218 insertions(+), 106 deletions(-) (limited to 'include/uapi') diff --git a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c index c570cf780079..0cb99e75b0bc 100644 --- a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c +++ b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c @@ -1161,7 +1161,8 @@ static int igt_write_huge(struct drm_i915_private *i915, GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj)); size = obj->base.size; - if (obj->mm.page_sizes.sg & I915_GTT_PAGE_SIZE_64K) + if (obj->mm.page_sizes.sg & I915_GTT_PAGE_SIZE_64K && + !HAS_64K_PAGES(i915)) size = round_up(size, I915_GTT_PAGE_SIZE_2M); n = 0; @@ -1214,6 +1215,10 @@ static int igt_write_huge(struct drm_i915_private *i915, * size and ensure the vma offset is at the start of the pt * boundary, however to improve coverage we opt for testing both * aligned and unaligned offsets. + * + * With PS64 this is no longer the case, but to ensure we + * sometimes get the compact layout for smaller objects, apply + * the round_up anyway. */ if (obj->mm.page_sizes.sg & I915_GTT_PAGE_SIZE_64K) offset_low = round_down(offset_low, @@ -1411,6 +1416,7 @@ static int igt_ppgtt_sanity_check(void *arg) { SZ_2M + SZ_4K, SZ_64K | SZ_4K }, { SZ_2M + SZ_4K, SZ_2M | SZ_4K }, { SZ_2M + SZ_64K, SZ_2M | SZ_64K }, + { SZ_2M + SZ_64K, SZ_64K }, }; int i, j; int err; @@ -1540,6 +1546,154 @@ out_put: return err; } +static int igt_ppgtt_mixed(void *arg) +{ + struct drm_i915_private *i915 = arg; + const unsigned long flags = PIN_OFFSET_FIXED | PIN_USER; + struct drm_i915_gem_object *obj, *on; + struct i915_gem_engines *engines; + struct i915_gem_engines_iter it; + struct i915_address_space *vm; + struct i915_gem_context *ctx; + struct intel_context *ce; + struct file *file; + I915_RND_STATE(prng); + LIST_HEAD(objects); + struct intel_memory_region *mr; + struct i915_vma *vma; + unsigned int count; + u32 i, addr; + int *order; + int n, err; + + /* + * Sanity check mixing 4K and 64K pages within the same page-table via + * the new PS64 TLB hint. + */ + + if (!HAS_64K_PAGES(i915)) { + pr_info("device lacks PS64, skipping\n"); + return 0; + } + + file = mock_file(i915); + if (IS_ERR(file)) + return PTR_ERR(file); + + ctx = hugepage_ctx(i915, file); + if (IS_ERR(ctx)) { + err = PTR_ERR(ctx); + goto out; + } + vm = i915_gem_context_get_eb_vm(ctx); + + i = 0; + addr = 0; + do { + u32 sz; + + sz = i915_prandom_u32_max_state(SZ_4M, &prng); + sz = max_t(u32, sz, SZ_4K); + + mr = i915->mm.regions[INTEL_REGION_LMEM_0]; + if (i & 1) + mr = i915->mm.regions[INTEL_REGION_SMEM]; + + obj = i915_gem_object_create_region(mr, sz, 0, 0); + if (IS_ERR(obj)) { + err = PTR_ERR(obj); + goto out_vm; + } + + list_add_tail(&obj->st_link, &objects); + + vma = i915_vma_instance(obj, vm, NULL); + if (IS_ERR(vma)) { + err = PTR_ERR(vma); + goto err_put; + } + + addr = round_up(addr, mr->min_page_size); + err = i915_vma_pin(vma, 0, 0, addr | flags); + if (err) + goto err_put; + + if (mr->type == INTEL_MEMORY_LOCAL && + (vma->resource->page_sizes_gtt & I915_GTT_PAGE_SIZE_4K)) { + err = -EINVAL; + goto err_put; + } + + addr += obj->base.size; + i++; + } while (addr <= SZ_16M); + + n = 0; + count = 0; + for_each_gem_engine(ce, i915_gem_context_lock_engines(ctx), it) { + count++; + if (!intel_engine_can_store_dword(ce->engine)) + continue; + + n++; + } + i915_gem_context_unlock_engines(ctx); + if (!n) + goto err_put; + + order = i915_random_order(count * count, &prng); + if (!order) { + err = -ENOMEM; + goto err_put; + } + + i = 0; + addr = 0; + engines = i915_gem_context_lock_engines(ctx); + list_for_each_entry(obj, &objects, st_link) { + u32 rnd = i915_prandom_u32_max_state(UINT_MAX, &prng); + + addr = round_up(addr, obj->mm.region->min_page_size); + + ce = engines->engines[order[i] % engines->num_engines]; + i = (i + 1) % (count * count); + if (!ce || !intel_engine_can_store_dword(ce->engine)) + continue; + + err = __igt_write_huge(ce, obj, obj->base.size, addr, 0, rnd); + if (err) + break; + + err = __igt_write_huge(ce, obj, obj->base.size, addr, + offset_in_page(rnd) / sizeof(u32), rnd + 1); + if (err) + break; + + err = __igt_write_huge(ce, obj, obj->base.size, addr, + (PAGE_SIZE / sizeof(u32)) - 1, + rnd + 2); + if (err) + break; + + addr += obj->base.size; + + cond_resched(); + } + + i915_gem_context_unlock_engines(ctx); + kfree(order); +err_put: + list_for_each_entry_safe(obj, on, &objects, st_link) { + list_del(&obj->st_link); + i915_gem_object_put(obj); + } +out_vm: + i915_vm_put(vm); +out: + fput(file); + return err; +} + static int igt_tmpfs_fallback(void *arg) { struct drm_i915_private *i915 = arg; @@ -1803,6 +1957,7 @@ int i915_gem_huge_page_live_selftests(struct drm_i915_private *i915) SUBTEST(igt_ppgtt_smoke_huge), SUBTEST(igt_ppgtt_sanity_check), SUBTEST(igt_ppgtt_compact), + SUBTEST(igt_ppgtt_mixed), }; if (!HAS_PPGTT(i915)) { diff --git a/drivers/gpu/drm/i915/gt/gen8_ppgtt.c b/drivers/gpu/drm/i915/gt/gen8_ppgtt.c index 8dd60d5ef905..4daaa6f55668 100644 --- a/drivers/gpu/drm/i915/gt/gen8_ppgtt.c +++ b/drivers/gpu/drm/i915/gt/gen8_ppgtt.c @@ -476,6 +476,7 @@ xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm, const gen8_pte_t pte_encode = vm->pte_encode(0, cache_level, flags); unsigned int rem = sg_dma_len(iter->sg); u64 start = vma_res->start; + u64 end = start + vma_res->vma_size; GEM_BUG_ON(!i915_vm_is_4lvl(vm)); @@ -489,9 +490,10 @@ xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm, gen8_pte_t encode = pte_encode; unsigned int page_size; gen8_pte_t *vaddr; - u16 index, max; + u16 index, max, nent, i; max = I915_PDES; + nent = 1; if (vma_res->bi.page_sizes.sg & I915_GTT_PAGE_SIZE_2M && IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_2M) && @@ -503,25 +505,37 @@ xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm, vaddr = px_vaddr(pd); } else { - if (encode & GEN12_PPGTT_PTE_LM) { - GEM_BUG_ON(__gen8_pte_index(start, 0) % 16); - GEM_BUG_ON(rem < I915_GTT_PAGE_SIZE_64K); - GEM_BUG_ON(!IS_ALIGNED(iter->dma, - I915_GTT_PAGE_SIZE_64K)); - - index = __gen8_pte_index(start, 0) / 16; - page_size = I915_GTT_PAGE_SIZE_64K; - - max /= 16; - - vaddr = px_vaddr(pd); - vaddr[__gen8_pte_index(start, 1)] |= GEN12_PDE_64K; + index = __gen8_pte_index(start, 0); + page_size = I915_GTT_PAGE_SIZE; - pt->is_compact = true; - } else { - GEM_BUG_ON(pt->is_compact); - index = __gen8_pte_index(start, 0); - page_size = I915_GTT_PAGE_SIZE; + if (vma_res->bi.page_sizes.sg & I915_GTT_PAGE_SIZE_64K) { + /* + * Device local-memory on these platforms should + * always use 64K pages or larger (including GTT + * alignment), therefore if we know the whole + * page-table needs to be filled we can always + * safely use the compact-layout. Otherwise fall + * back to the TLB hint with PS64. If this is + * system memory we only bother with PS64. + */ + if ((encode & GEN12_PPGTT_PTE_LM) && + end - start >= SZ_2M && !index) { + index = __gen8_pte_index(start, 0) / 16; + page_size = I915_GTT_PAGE_SIZE_64K; + + max /= 16; + + vaddr = px_vaddr(pd); + vaddr[__gen8_pte_index(start, 1)] |= GEN12_PDE_64K; + + pt->is_compact = true; + } else if (IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_64K) && + rem >= I915_GTT_PAGE_SIZE_64K && + !(index % 16)) { + encode |= GEN12_PTE_PS64; + page_size = I915_GTT_PAGE_SIZE_64K; + nent = 16; + } } vaddr = px_vaddr(pt); @@ -529,7 +543,12 @@ xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm, do { GEM_BUG_ON(rem < page_size); - vaddr[index++] = encode | iter->dma; + + for (i = 0; i < nent; i++) { + vaddr[index++] = + encode | (iter->dma + i * + I915_GTT_PAGE_SIZE); + } start += page_size; iter->dma += page_size; @@ -745,6 +764,8 @@ static void __xehpsdv_ppgtt_insert_entry_lm(struct i915_address_space *vm, GEM_BUG_ON(!IS_ALIGNED(addr, SZ_64K)); GEM_BUG_ON(!IS_ALIGNED(offset, SZ_64K)); + /* XXX: we don't strictly need to use this layout */ + if (!pt->is_compact) { vaddr = px_vaddr(pd); vaddr[gen8_pd_index(idx, 1)] |= GEN12_PDE_64K; @@ -935,22 +956,10 @@ struct i915_ppgtt *gen8_ppgtt_create(struct intel_gt *gt, ppgtt->vm.alloc_pt_dma = alloc_pt_dma; /* - * On some platforms the hw has dropped support for 4K GTT pages - * when dealing with LMEM, and due to the design of 64K GTT - * pages in the hw, we can only mark the *entire* page-table as - * operating in 64K GTT mode, since the enable bit is still on - * the pde, and not the pte. And since we still need to allow - * 4K GTT pages for SMEM objects, we can't have a "normal" 4K - * page-table with scratch pointing to LMEM, since that's - * undefined from the hw pov. The simplest solution is to just - * move the 64K scratch page to SMEM on all platforms and call - * it a day, since that should work for all configurations. - * - * Using SMEM instead of LMEM has the additional advantage of - * not reserving high performance memory for a "never" used - * filler page. It also removes the device access that would - * be required to initialise the scratch page, reducing pressure - * on an even scarcer resource. + * Using SMEM here instead of LMEM has the advantage of not reserving + * high performance memory for a "never" used filler page. It also + * removes the device access that would be required to initialise the + * scratch page, reducing pressure on an even scarcer resource. */ ppgtt->vm.alloc_scratch_dma = alloc_pt_dma; diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c b/drivers/gpu/drm/i915/gt/intel_gtt.c index 2eaeba14319e..13e411187fd5 100644 --- a/drivers/gpu/drm/i915/gt/intel_gtt.c +++ b/drivers/gpu/drm/i915/gt/intel_gtt.c @@ -269,11 +269,7 @@ void i915_address_space_init(struct i915_address_space *vm, int subclass) memset64(vm->min_alignment, I915_GTT_MIN_ALIGNMENT, ARRAY_SIZE(vm->min_alignment)); - if (HAS_64K_PAGES(vm->i915) && NEEDS_COMPACT_PT(vm->i915) && - subclass == VM_CLASS_PPGTT) { - vm->min_alignment[INTEL_MEMORY_LOCAL] = I915_GTT_PAGE_SIZE_2M; - vm->min_alignment[INTEL_MEMORY_STOLEN_LOCAL] = I915_GTT_PAGE_SIZE_2M; - } else if (HAS_64K_PAGES(vm->i915)) { + if (HAS_64K_PAGES(vm->i915)) { vm->min_alignment[INTEL_MEMORY_LOCAL] = I915_GTT_PAGE_SIZE_64K; vm->min_alignment[INTEL_MEMORY_STOLEN_LOCAL] = I915_GTT_PAGE_SIZE_64K; } @@ -343,7 +339,8 @@ int setup_scratch_page(struct i915_address_space *vm) */ size = I915_GTT_PAGE_SIZE_4K; if (i915_vm_is_4lvl(vm) && - HAS_PAGE_SIZES(vm->i915, I915_GTT_PAGE_SIZE_64K)) + HAS_PAGE_SIZES(vm->i915, I915_GTT_PAGE_SIZE_64K) && + !HAS_64K_PAGES(vm->i915)) size = I915_GTT_PAGE_SIZE_64K; do { @@ -385,18 +382,6 @@ skip: if (size == I915_GTT_PAGE_SIZE_4K) return -ENOMEM; - /* - * If we need 64K minimum GTT pages for device local-memory, - * like on XEHPSDV, then we need to fail the allocation here, - * otherwise we can't safely support the insertion of - * local-memory pages for this vm, since the HW expects the - * correct physical alignment and size when the page-table is - * operating in 64K GTT mode, which includes any scratch PTEs, - * since userspace can still touch them. - */ - if (HAS_64K_PAGES(vm->i915)) - return -ENOMEM; - size = I915_GTT_PAGE_SIZE_4K; } while (1); } diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.h b/drivers/gpu/drm/i915/gt/intel_gtt.h index c0ca53cba9f0..062b78333fb2 100644 --- a/drivers/gpu/drm/i915/gt/intel_gtt.h +++ b/drivers/gpu/drm/i915/gt/intel_gtt.h @@ -93,6 +93,7 @@ typedef u64 gen8_pte_t; #define GEN12_GGTT_PTE_LM BIT_ULL(1) #define GEN12_PDE_64K BIT(6) +#define GEN12_PTE_PS64 BIT(8) /* * Cacheability Control is a 4-bit value. The low three bits are stored in bits diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index ca0609dc5fb0..e390da969be5 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -905,13 +905,6 @@ IS_SUBPLATFORM(const struct drm_i915_private *i915, */ #define HAS_64K_PAGES(dev_priv) (INTEL_INFO(dev_priv)->has_64k_pages) -/* - * Set this flag when platform doesn't allow both 64k pages and 4k pages in - * the same PT. this flag means we need to support compact PT layout for the - * ppGTT when using the 64K GTT pages. - */ -#define NEEDS_COMPACT_PT(dev_priv) (INTEL_INFO(dev_priv)->needs_compact_pt) - #define HAS_IPC(dev_priv) (INTEL_INFO(dev_priv)->display.has_ipc) #define HAS_REGION(i915, i) (RUNTIME_INFO(i915)->memory_regions & (i)) diff --git a/drivers/gpu/drm/i915/i915_pci.c b/drivers/gpu/drm/i915/i915_pci.c index cd4487a1d3be..f1344db626f5 100644 --- a/drivers/gpu/drm/i915/i915_pci.c +++ b/drivers/gpu/drm/i915/i915_pci.c @@ -1042,7 +1042,6 @@ static const struct intel_device_info xehpsdv_info = { PLATFORM(INTEL_XEHPSDV), NO_DISPLAY, .has_64k_pages = 1, - .needs_compact_pt = 1, .has_media_ratio_mode = 1, .__runtime.platform_engine_mask = BIT(RCS0) | BIT(BCS0) | @@ -1064,7 +1063,6 @@ static const struct intel_device_info xehpsdv_info = { .has_64k_pages = 1, \ .has_guc_deprivilege = 1, \ .has_heci_pxp = 1, \ - .needs_compact_pt = 1, \ .has_media_ratio_mode = 1, \ .__runtime.platform_engine_mask = \ BIT(RCS0) | BIT(BCS0) | \ diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c index f17c09ead7d7..c39488eb9eeb 100644 --- a/drivers/gpu/drm/i915/i915_vma.c +++ b/drivers/gpu/drm/i915/i915_vma.c @@ -776,12 +776,6 @@ i915_vma_insert(struct i915_vma *vma, struct i915_gem_ww_ctx *ww, GEM_BUG_ON(!IS_ALIGNED(end, I915_GTT_PAGE_SIZE)); alignment = max(alignment, i915_vm_obj_min_alignment(vma->vm, vma->obj)); - /* - * for compact-pt we round up the reservation to prevent - * any smaller pages being used within the same PDE - */ - if (NEEDS_COMPACT_PT(vma->vm->i915)) - size = round_up(size, alignment); /* If binding the object/GGTT view requires more space than the entire * aperture has, reject it early before evicting everything in a vain @@ -820,7 +814,8 @@ i915_vma_insert(struct i915_vma *vma, struct i915_gem_ww_ctx *ww, * forseeable future. See also i915_ggtt_offset(). */ if (upper_32_bits(end - 1) && - vma->page_sizes.sg > I915_GTT_PAGE_SIZE) { + vma->page_sizes.sg > I915_GTT_PAGE_SIZE && + !HAS_64K_PAGES(vma->vm->i915)) { /* * We can't mix 64K and 4K PTEs in the same page-table * (2M block), and so to avoid the ugliness and diff --git a/drivers/gpu/drm/i915/intel_device_info.h b/drivers/gpu/drm/i915/intel_device_info.h index d638235e1d26..da833267a782 100644 --- a/drivers/gpu/drm/i915/intel_device_info.h +++ b/drivers/gpu/drm/i915/intel_device_info.h @@ -146,7 +146,6 @@ enum intel_ppgtt_type { /* Keep has_* in alphabetical order */ \ func(has_64bit_reloc); \ func(has_64k_pages); \ - func(needs_compact_pt); \ func(gpu_reset_clobbers_display); \ func(has_reset_engine); \ func(has_3d_pipeline); \ diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c index ea2cf1080979..27c733b00976 100644 --- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c +++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c @@ -1114,15 +1114,8 @@ static int misaligned_case(struct i915_address_space *vm, struct intel_memory_re expected_node_size = expected_vma_size; if (HAS_64K_PAGES(vm->i915) && i915_gem_object_is_lmem(obj)) { - /* - * The compact-pt should expand lmem node to 2MB for the ppGTT, - * for all other cases we should only expect 64K. - */ expected_vma_size = round_up(size, I915_GTT_PAGE_SIZE_64K); - if (NEEDS_COMPACT_PT(vm->i915) && !i915_is_ggtt(vm)) - expected_node_size = round_up(size, I915_GTT_PAGE_SIZE_2M); - else - expected_node_size = round_up(size, I915_GTT_PAGE_SIZE_64K); + expected_node_size = round_up(size, I915_GTT_PAGE_SIZE_64K); } if (vma->size != expected_vma_size || vma->node.size != expected_node_size) { diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index 629198f1d8d8..08d69e36fb66 100644 --- a/include/uapi/drm/i915_drm.h +++ b/include/uapi/drm/i915_drm.h @@ -3509,27 +3509,13 @@ struct drm_i915_gem_create_ext { * * The (page-aligned) allocated size for the object will be returned. * - * DG2 64K min page size implications: + * On platforms like DG2/ATS the kernel will always use 64K or larger + * pages for I915_MEMORY_CLASS_DEVICE. The kernel also requires a + * minimum of 64K GTT alignment for such objects. * - * On discrete platforms, starting from DG2, we have to contend with GTT - * page size restrictions when dealing with I915_MEMORY_CLASS_DEVICE - * objects. Specifically the hardware only supports 64K or larger GTT - * page sizes for such memory. The kernel will already ensure that all - * I915_MEMORY_CLASS_DEVICE memory is allocated using 64K or larger page - * sizes underneath. - * - * Note that the returned size here will always reflect any required - * rounding up done by the kernel, i.e 4K will now become 64K on devices - * such as DG2. The kernel will always select the largest minimum - * page-size for the set of possible placements as the value to use when - * rounding up the @size. - * - * Special DG2 GTT address alignment requirement: - * - * The GTT alignment will also need to be at least 2M for such objects. - * - * Note that due to how the hardware implements 64K GTT page support, we - * have some further complications: + * NOTE: Previously the ABI here required a minimum GTT alignment of 2M + * on DG2/ATS, due to how the hardware implemented 64K GTT page support, + * where we had the following complications: * * 1) The entire PDE (which covers a 2MB virtual address range), must * contain only 64K PTEs, i.e mixing 4K and 64K PTEs in the same @@ -3538,12 +3524,10 @@ struct drm_i915_gem_create_ext { * 2) We still need to support 4K PTEs for I915_MEMORY_CLASS_SYSTEM * objects. * - * To keep things simple for userland, we mandate that any GTT mappings - * must be aligned to and rounded up to 2MB. The kernel will internally - * pad them out to the next 2MB boundary. As this only wastes virtual - * address space and avoids userland having to copy any needlessly - * complicated PDE sharing scheme (coloring) and only affects DG2, this - * is deemed to be a good compromise. + * However on actual production HW this was completely changed to now + * allow setting a TLB hint at the PTE level (see PS64), which is a lot + * more flexible than the above. With this the 2M restriction was + * dropped where we now only require 64K. */ __u64 size; -- cgit v1.2.3 From d54576a074a29d4901d0a693cd84e1a89057f694 Mon Sep 17 00:00:00 2001 From: Matthew Auld Date: Tue, 4 Oct 2022 12:49:15 +0100 Subject: drm/i915/uapi: expose GTT alignment MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit On some platforms we potentially have different alignment restrictions depending on the memory type. We also now have different alignment restrictions for the same region across different kernel versions. Extend the region query to return the minimum required GTT alignment. Testcase: igt@gem_create@create-ext-placement-alignment Testcase: igt@i915_query@query-regions-sanity-check Suggested-by: Lionel Landwerlin Signed-off-by: Matthew Auld Cc: Michal Mrozek Cc: Thomas Hellström Cc: Stuart Summers Cc: Jordan Justen Cc: Yang A Shi Cc: Nirmoy Das Cc: Niranjana Vishwanathapura Reviewed-by: Nirmoy Das Acked-by: Jordan Justen Link: https://patchwork.freedesktop.org/patch/msgid/20221004114915.221708-2-matthew.auld@intel.com --- drivers/gpu/drm/i915/i915_query.c | 1 + include/uapi/drm/i915_drm.h | 29 +++++++++++++++++++++++++++-- 2 files changed, 28 insertions(+), 2 deletions(-) (limited to 'include/uapi') diff --git a/drivers/gpu/drm/i915/i915_query.c b/drivers/gpu/drm/i915/i915_query.c index 6ec9c9fb7b0d..111377f210ed 100644 --- a/drivers/gpu/drm/i915/i915_query.c +++ b/drivers/gpu/drm/i915/i915_query.c @@ -498,6 +498,7 @@ static int query_memregion_info(struct drm_i915_private *i915, info.region.memory_class = mr->type; info.region.memory_instance = mr->instance; info.probed_size = mr->total; + info.gtt_alignment = mr->min_page_size; if (mr->type == INTEL_MEMORY_LOCAL) info.probed_cpu_visible_size = mr->io_size; diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index 08d69e36fb66..2e613109356b 100644 --- a/include/uapi/drm/i915_drm.h +++ b/include/uapi/drm/i915_drm.h @@ -3346,8 +3346,33 @@ struct drm_i915_memory_region_info { /** @region: The class:instance pair encoding */ struct drm_i915_gem_memory_class_instance region; - /** @rsvd0: MBZ */ - __u32 rsvd0; + union { + /** @rsvd0: MBZ */ + __u32 rsvd0; + /** + * @gtt_alignment: + * + * The minimum required GTT alignment for this type of memory. + * When allocating a GTT address it must be aligned to this + * value or larger. On some platforms the kernel might opt to + * using 64K pages for I915_MEMORY_CLASS_DEVICE, where 64K GTT + * pages can then be used if we also use 64K GTT alignment. + * + * NOTE: If this is zero then this must be an older + * kernel which lacks support for this field. + * + * Side note: For larger objects (especially for + * I915_MEMORY_CLASS_DEVICE), like 2M+ in size, userspace should + * consider potentially bumping the GTT alignment to say 2M, + * which could potentially increase the likelihood of the kernel + * being able to utilise 2M GTT pages underneath, if the layout + * of the physical pages allows it. On some configurations we + * can then also use a more efficient page-table layout, if we + * can't use the more desirable 2M GTT page, so long as we know + * that the entire page-table will be used by this object. + */ + __u32 gtt_alignment; + }; /** * @probed_size: Memory probed by the driver -- cgit v1.2.3 From b0feda9ce756aa62dbfc29372f819734ffa195f9 Mon Sep 17 00:00:00 2001 From: Matthew Auld Date: Mon, 24 Oct 2022 11:19:46 +0100 Subject: Revert "drm/i915/uapi: expose GTT alignment" MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The process for merging uAPI is to have UMD side ready and reviewed and merged before merging. Revert for now until that is ready. This reverts commit d54576a074a29d4901d0a693cd84e1a89057f694. Reported-by: Joonas Lahtinen Signed-off-by: Matthew Auld Cc: Lionel Landwerlin Cc: Michal Mrozek Cc: Thomas Hellström Cc: Stuart Summers Cc: Jordan Justen Cc: Yang A Shi Cc: Nirmoy Das Cc: Niranjana Vishwanathapura Reviewed-by: Nirmoy Das Link: https://patchwork.freedesktop.org/patch/msgid/20221024101946.28974-1-matthew.auld@intel.com --- drivers/gpu/drm/i915/i915_query.c | 1 - include/uapi/drm/i915_drm.h | 29 ++--------------------------- 2 files changed, 2 insertions(+), 28 deletions(-) (limited to 'include/uapi') diff --git a/drivers/gpu/drm/i915/i915_query.c b/drivers/gpu/drm/i915/i915_query.c index 111377f210ed..6ec9c9fb7b0d 100644 --- a/drivers/gpu/drm/i915/i915_query.c +++ b/drivers/gpu/drm/i915/i915_query.c @@ -498,7 +498,6 @@ static int query_memregion_info(struct drm_i915_private *i915, info.region.memory_class = mr->type; info.region.memory_instance = mr->instance; info.probed_size = mr->total; - info.gtt_alignment = mr->min_page_size; if (mr->type == INTEL_MEMORY_LOCAL) info.probed_cpu_visible_size = mr->io_size; diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index 2e613109356b..08d69e36fb66 100644 --- a/include/uapi/drm/i915_drm.h +++ b/include/uapi/drm/i915_drm.h @@ -3346,33 +3346,8 @@ struct drm_i915_memory_region_info { /** @region: The class:instance pair encoding */ struct drm_i915_gem_memory_class_instance region; - union { - /** @rsvd0: MBZ */ - __u32 rsvd0; - /** - * @gtt_alignment: - * - * The minimum required GTT alignment for this type of memory. - * When allocating a GTT address it must be aligned to this - * value or larger. On some platforms the kernel might opt to - * using 64K pages for I915_MEMORY_CLASS_DEVICE, where 64K GTT - * pages can then be used if we also use 64K GTT alignment. - * - * NOTE: If this is zero then this must be an older - * kernel which lacks support for this field. - * - * Side note: For larger objects (especially for - * I915_MEMORY_CLASS_DEVICE), like 2M+ in size, userspace should - * consider potentially bumping the GTT alignment to say 2M, - * which could potentially increase the likelihood of the kernel - * being able to utilise 2M GTT pages underneath, if the layout - * of the physical pages allows it. On some configurations we - * can then also use a more efficient page-table layout, if we - * can't use the more desirable 2M GTT page, so long as we know - * that the entire page-table will be used by this object. - */ - __u32 gtt_alignment; - }; + /** @rsvd0: MBZ */ + __u32 rsvd0; /** * @probed_size: Memory probed by the driver -- cgit v1.2.3 From 81d5f7d91492aa3a362937926cdc094a7dc1e4b7 Mon Sep 17 00:00:00 2001 From: Umesh Nerlige Ramappa Date: Wed, 26 Oct 2022 22:20:48 +0000 Subject: drm/i915/perf: Add 32-bit OAG and OAR formats for DG2 Add new OA formats for DG2. MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18893 v2: - Update commit title (Ashutosh) - Coding style fixes (Lionel) - 64 bit OA formats need UMD changes in GPUvis, drop for now and send in a separate series with UMD changes v3: - Update commit message to drop 64 bit related description Signed-off-by: Umesh Nerlige Ramappa Reviewed-by: Lionel Landwerlin #1 Signed-off-by: John Harrison Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-3-umesh.nerlige.ramappa@intel.com --- drivers/gpu/drm/i915/i915_perf.c | 7 +++++++ include/uapi/drm/i915_drm.h | 4 ++++ 2 files changed, 11 insertions(+) (limited to 'include/uapi') diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c index 255335868b6a..2b772a6b1cd6 100644 --- a/drivers/gpu/drm/i915/i915_perf.c +++ b/drivers/gpu/drm/i915/i915_perf.c @@ -320,6 +320,8 @@ static const struct i915_oa_format oa_formats[I915_OA_FORMAT_MAX] = { [I915_OA_FORMAT_A12] = { 0, 64 }, [I915_OA_FORMAT_A12_B8_C8] = { 2, 128 }, [I915_OA_FORMAT_A32u40_A4u32_B8_C8] = { 5, 256 }, + [I915_OAR_FORMAT_A32u40_A4u32_B8_C8] = { 5, 256 }, + [I915_OA_FORMAT_A24u40_A14u32_B8_C8] = { 5, 256 }, }; #define SAMPLE_OA_REPORT (1<<0) @@ -4515,6 +4517,11 @@ static void oa_init_supported_formats(struct i915_perf *perf) oa_format_add(perf, I915_OA_FORMAT_C4_B8); break; + case INTEL_DG2: + oa_format_add(perf, I915_OAR_FORMAT_A32u40_A4u32_B8_C8); + oa_format_add(perf, I915_OA_FORMAT_A24u40_A14u32_B8_C8); + break; + default: MISSING_CASE(platform); } diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index 08d69e36fb66..b946674c68cc 100644 --- a/include/uapi/drm/i915_drm.h +++ b/include/uapi/drm/i915_drm.h @@ -2666,6 +2666,10 @@ enum drm_i915_oa_format { I915_OA_FORMAT_A12_B8_C8, I915_OA_FORMAT_A32u40_A4u32_B8_C8, + /* DG2 */ + I915_OAR_FORMAT_A32u40_A4u32_B8_C8, + I915_OA_FORMAT_A24u40_A14u32_B8_C8, + I915_OA_FORMAT_MAX /* non-ABI */ }; -- cgit v1.2.3 From bc7ed4d30815bc434c1e49dc6784164b352d167c Mon Sep 17 00:00:00 2001 From: Umesh Nerlige Ramappa Date: Wed, 26 Oct 2022 22:20:58 +0000 Subject: drm/i915/perf: Apply Wa_18013179988 OA reports in the OA buffer contain an OA timestamp field that helps user calculate delta between 2 OA reports. The calculation relies on the CS timestamp frequency to convert the timestamp value to nanoseconds. The CS timestamp frequency is a function of the CTC_SHIFT value in RPM_CONFIG0. In DG2, OA unit assumes that the CTC_SHIFT is 3, instead of using the actual value from RPM_CONFIG0. At the user level, this results in an error in calculating delta between 2 OA reports since the OA timestamp is not shifted in the same manner as CS timestamp. Also the periodicity of the reports is different from what the user configured because of mismatch in the CS and OA frequencies. The issue also affects MI_REPORT_PERF_COUNT command. To resolve this, return actual OA timestamp frequency to the user in i915_getparam_ioctl, so that user can calculate the right OA exponent as well as interpret the reports correctly. MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18893 v2: - Use REG_FIELD_GET (Ashutosh) - Update commit msg Signed-off-by: Umesh Nerlige Ramappa Reviewed-by: Ashutosh Dixit Signed-off-by: John Harrison Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-13-umesh.nerlige.ramappa@intel.com --- drivers/gpu/drm/i915/i915_getparam.c | 3 +++ drivers/gpu/drm/i915/i915_perf.c | 30 ++++++++++++++++++++++++++++-- drivers/gpu/drm/i915/i915_perf.h | 2 ++ include/uapi/drm/i915_drm.h | 6 ++++++ 4 files changed, 39 insertions(+), 2 deletions(-) (limited to 'include/uapi') diff --git a/drivers/gpu/drm/i915/i915_getparam.c b/drivers/gpu/drm/i915/i915_getparam.c index 342c8ca6414e..3047e80e1163 100644 --- a/drivers/gpu/drm/i915/i915_getparam.c +++ b/drivers/gpu/drm/i915/i915_getparam.c @@ -175,6 +175,9 @@ int i915_getparam_ioctl(struct drm_device *dev, void *data, case I915_PARAM_PERF_REVISION: value = i915_perf_ioctl_version(); break; + case I915_PARAM_OA_TIMESTAMP_FREQUENCY: + value = i915_perf_oa_timestamp_frequency(i915); + break; default: DRM_DEBUG("Unknown parameter %d\n", param->param); return -EINVAL; diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c index e14d16ac47de..b73d91b792df 100644 --- a/drivers/gpu/drm/i915/i915_perf.c +++ b/drivers/gpu/drm/i915/i915_perf.c @@ -3109,6 +3109,30 @@ get_sseu_config(struct intel_sseu *out_sseu, return i915_gem_user_to_context_sseu(engine->gt, drm_sseu, out_sseu); } +/* + * OA timestamp frequency = CS timestamp frequency in most platforms. On some + * platforms OA unit ignores the CTC_SHIFT and the 2 timestamps differ. In such + * cases, return the adjusted CS timestamp frequency to the user. + */ +u32 i915_perf_oa_timestamp_frequency(struct drm_i915_private *i915) +{ + /* Wa_18013179988:dg2 */ + if (IS_DG2(i915)) { + intel_wakeref_t wakeref; + u32 reg, shift; + + with_intel_runtime_pm(to_gt(i915)->uncore->rpm, wakeref) + reg = intel_uncore_read(to_gt(i915)->uncore, RPM_CONFIG0); + + shift = REG_FIELD_GET(GEN10_RPM_CONFIG0_CTC_SHIFT_PARAMETER_MASK, + reg); + + return to_gt(i915)->clock_frequency << (3 - shift); + } + + return to_gt(i915)->clock_frequency; +} + /** * i915_oa_stream_init - validate combined props for OA stream and init * @stream: An i915 perf stream @@ -3830,8 +3854,10 @@ err: static u64 oa_exponent_to_ns(struct i915_perf *perf, int exponent) { - return intel_gt_clock_interval_to_ns(to_gt(perf->i915), - 2ULL << exponent); + u64 nom = (2ULL << exponent) * NSEC_PER_SEC; + u32 den = i915_perf_oa_timestamp_frequency(perf->i915); + + return div_u64(nom + den - 1, den); } static __always_inline bool diff --git a/drivers/gpu/drm/i915/i915_perf.h b/drivers/gpu/drm/i915/i915_perf.h index 1d1329e5af3a..f96e09a4af04 100644 --- a/drivers/gpu/drm/i915/i915_perf.h +++ b/drivers/gpu/drm/i915/i915_perf.h @@ -57,4 +57,6 @@ static inline void i915_oa_config_put(struct i915_oa_config *oa_config) kref_put(&oa_config->ref, i915_oa_config_release); } +u32 i915_perf_oa_timestamp_frequency(struct drm_i915_private *i915); + #endif /* __I915_PERF_H__ */ diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index b946674c68cc..8df261c5ab9b 100644 --- a/include/uapi/drm/i915_drm.h +++ b/include/uapi/drm/i915_drm.h @@ -765,6 +765,12 @@ typedef struct drm_i915_irq_wait { /* Query if the kernel supports the I915_USERPTR_PROBE flag. */ #define I915_PARAM_HAS_USERPTR_PROBE 56 +/* + * Frequency of the timestamps in OA reports. This used to be the same as the CS + * timestamp frequency, but differs on some platforms. + */ +#define I915_PARAM_OA_TIMESTAMP_FREQUENCY 57 + /* Must be kept compact -- no holes and well documented */ /** -- cgit v1.2.3