To comply with the design that buffer objects shall have immutable cache setting through out their life cycle, {set, get}_caching ioctl's are no longer supported from MTL onward. With that change caching policy can only be set at object creation time. The current code applies a default (platform dependent) cache setting for all objects. However this is not optimal for performance tuning. The patch extends the existing gem_create uAPI to let user set PAT index for the object at creation time. The new extension is platform independent, so UMD's can switch to using this extension for older platforms as well, while {set, get}_caching are still supported on these legacy paltforms for compatibility reason. However, since PAT index was not clearly defined for platforms prior to GEN12 (TGL), so we are limiting this externsion to GEN12+ platforms only. See ext_set_pat() in for the implementation details. The documentation related to the PAT/MOCS tables is currently available for Tiger Lake here: https://www.intel.com/content/www/us/en/docs/graphics-for-linux/developer-reference/1-0/tiger-lake.html The documentation for other platforms is currently being updated. BSpec: 45101 Mesa support has been submitted in this merge request: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878 The media driver supprt has bin submitted in this merge request: https://github.com/intel/media-driver/pull/1680 The IGT test related to this change is igt@gem_create@create-ext-set-pat
Unified · Split
| @@ -245,6 +245,7 @@ struct create_ext { · 244 unmodified lines | |||
| 245 | unsigned int n_placements; | 245 | unsigned int n_placements; |
| 246 | unsigned int placement_mask; | 246 | unsigned int placement_mask; |
| 247 | unsigned long flags; | 247 | unsigned long flags; |
| 248 | unsigned int pat_index; | ||
| 248 | }; | 249 | }; |
| 249 | 250 | ||
| 250 | static void repr_placements(char *buf, size_t size, | 251 | static void repr_placements(char *buf, size_t size, |
| @@ -394,11 +395,43 @@ static int ext_set_protected(struct i915_user_extension __user *base, void *data · 143 unmodified lines | |||
| 394 | return 0; | 395 | return 0; |
| 395 | } | 396 | } |
| 396 | 397 | ||
| 398 | static int ext_set_pat(struct i915_user_extension __user *base, void *data) | ||
| 399 | { | ||
| 400 | struct create_ext *ext_data = data; | ||
| 401 | struct drm_i915_private *i915 = ext_data->i915; | ||
| 402 | struct drm_i915_gem_create_ext_set_pat ext; | ||
| 403 | unsigned int max_pat_index; | ||
| 404 | |||
| 405 | BUILD_BUG_ON(sizeof(struct drm_i915_gem_create_ext_set_pat) != | ||
| 406 | offsetofend(struct drm_i915_gem_create_ext_set_pat, rsvd)); | ||
| 407 | |||
| 408 | /* Limiting the extension only to Meteor Lake */ | ||
| 409 | if (!IS_METEORLAKE(i915)) | ||
| 410 | return -ENODEV; | ||
| 411 | |||
| 412 | if (copy_from_user(&ext, base, sizeof(ext))) | ||
| 413 | return -EFAULT; | ||
| 414 | |||
| 415 | max_pat_index = INTEL_INFO(i915)->max_pat_index; | ||
| 416 | |||
| 417 | if (ext.pat_index > max_pat_index) { | ||
| 418 | drm_dbg(&i915->drm, "PAT index is invalid: %u\n", | ||
| 419 | ext.pat_index); | ||
| 420 | return -EINVAL; | ||
| 421 | } | ||
| 422 | |||
| 423 | ext_data->pat_index = ext.pat_index; | ||
| 424 | |||
| 425 | return 0; | ||
| 426 | } | ||
| 427 | |||
| 397 | static const i915_user_extension_fn create_extensions[] = { | 428 | static const i915_user_extension_fn create_extensions[] = { |
| 398 | [I915_GEM_CREATE_EXT_MEMORY_REGIONS] = ext_set_placements, | 429 | [I915_GEM_CREATE_EXT_MEMORY_REGIONS] = ext_set_placements, |
| 399 | [I915_GEM_CREATE_EXT_PROTECTED_CONTENT] = ext_set_protected, | 430 | [I915_GEM_CREATE_EXT_PROTECTED_CONTENT] = ext_set_protected, |
| 431 | [I915_GEM_CREATE_EXT_SET_PAT] = ext_set_pat, | ||
| 400 | }; | 432 | }; |
| 401 | 433 | ||
| 434 | #define PAT_INDEX_NOT_SET 0xffff | ||
| 402 | /** | 435 | /** |
| 403 | * i915_gem_create_ext_ioctl - Creates a new mm object and returns a handle to it. | 436 | * i915_gem_create_ext_ioctl - Creates a new mm object and returns a handle to it. |
| 404 | * @dev: drm device pointer | 437 | * @dev: drm device pointer |
| @@ -418,6 +451,7 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data, · 13 unmodified lines | |||
| 418 | if (args->flags & ~I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS) | 451 | if (args->flags & ~I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS) |
| 419 | return -EINVAL; | 452 | return -EINVAL; |
| 420 | 453 | ||
| 454 | ext_data.pat_index = PAT_INDEX_NOT_SET; | ||
| 421 | ret = i915_user_extensions(u64_to_user_ptr(args->extensions), | 455 | ret = i915_user_extensions(u64_to_user_ptr(args->extensions), |
| 422 | create_extensions, | 456 | create_extensions, |
| 423 | ARRAY_SIZE(create_extensions), | 457 | ARRAY_SIZE(create_extensions), |
| @@ -454,5 +488,11 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data, · 30 unmodified lines | |||
| 454 | if (IS_ERR(obj)) | 488 | if (IS_ERR(obj)) |
| 455 | return PTR_ERR(obj); | 489 | return PTR_ERR(obj); |
| 456 | 490 | ||
| 491 | if (ext_data.pat_index != PAT_INDEX_NOT_SET) { | ||
| 492 | i915_gem_object_set_pat_index(obj, ext_data.pat_index); | ||
| 493 | /* Mark pat_index is set by UMD */ | ||
| 494 | obj->pat_set_by_user = true; | ||
| 495 | } | ||
| 496 | |||
| 457 | return i915_gem_publish(obj, file, &args->size, &args->handle); | 497 | return i915_gem_publish(obj, file, &args->size, &args->handle); |
| 458 | } | 498 | } |
| @@ -208,6 +208,12 @@ bool i915_gem_object_can_bypass_llc(struct drm_i915_gem_object *obj) · 207 unmodified lines | |||
| 208 | if (!(obj->flags & I915_BO_ALLOC_USER)) | 208 | if (!(obj->flags & I915_BO_ALLOC_USER)) |
| 209 | return false; | 209 | return false; |
| 210 | 210 | ||
| 211 | /* | ||
| 212 | * Always flush cache for UMD objects at creation time. | ||
| 213 | */ | ||
| 214 | if (obj->pat_set_by_user) | ||
| 215 | return true; | ||
| 216 | |||
| 211 | /* | 217 | /* |
| 212 | * EHL and JSL add the 'Bypass LLC' MOCS entry, which should make it | 218 | * EHL and JSL add the 'Bypass LLC' MOCS entry, which should make it |
| 213 | * possible for userspace to bypass the GTT caching bits set by the | 219 | * possible for userspace to bypass the GTT caching bits set by the |
| @@ -3680,9 +3680,13 @@ struct drm_i915_gem_create_ext { · 3679 unmodified lines | |||
| 3680 | * | 3680 | * |
| 3681 | * For I915_GEM_CREATE_EXT_PROTECTED_CONTENT usage see | 3681 | * For I915_GEM_CREATE_EXT_PROTECTED_CONTENT usage see |
| 3682 | * struct drm_i915_gem_create_ext_protected_content. | 3682 | * struct drm_i915_gem_create_ext_protected_content. |
| 3683 | * | ||
| 3684 | * For I915_GEM_CREATE_EXT_SET_PAT usage see | ||
| 3685 | * struct drm_i915_gem_create_ext_set_pat. | ||
| 3683 | */ | 3686 | */ |
| 3684 | #define I915_GEM_CREATE_EXT_MEMORY_REGIONS 0 | 3687 | #define I915_GEM_CREATE_EXT_MEMORY_REGIONS 0 |
| 3685 | #define I915_GEM_CREATE_EXT_PROTECTED_CONTENT 1 | 3688 | #define I915_GEM_CREATE_EXT_PROTECTED_CONTENT 1 |
| 3689 | #define I915_GEM_CREATE_EXT_SET_PAT 2 | ||
| 3686 | __u64 extensions; | 3690 | __u64 extensions; |
| 3687 | }; | 3691 | }; |
| 3688 | 3692 | ||
| @@ -3797,6 +3801,43 @@ struct drm_i915_gem_create_ext_protected_content { · 108 unmodified lines | |||
| 3797 | __u32 flags; | 3801 | __u32 flags; |
| 3798 | }; | 3802 | }; |
| 3799 | 3803 | ||
| 3804 | /** | ||
| 3805 | * struct drm_i915_gem_create_ext_set_pat - The | ||
| 3806 | * I915_GEM_CREATE_EXT_SET_PAT extension. | ||
| 3807 | * | ||
| 3808 | * If this extension is provided, the specified caching policy (PAT index) is | ||
| 3809 | * applied to the buffer object. | ||
| 3810 | * | ||
| 3811 | * Below is an example on how to create an object with specific caching policy: | ||
| 3812 | * | ||
| 3813 | * .. code-block:: C | ||
| 3814 | * | ||
| 3815 | * struct drm_i915_gem_create_ext_set_pat set_pat_ext = { | ||
| 3816 | * .base = { .name = I915_GEM_CREATE_EXT_SET_PAT }, | ||
| 3817 | * .pat_index = 0, | ||
| 3818 | * }; | ||
| 3819 | * struct drm_i915_gem_create_ext create_ext = { | ||
| 3820 | * .size = PAGE_SIZE, | ||
| 3821 | * .extensions = (uintptr_t)&set_pat_ext, | ||
| 3822 | * }; | ||
| 3823 | * | ||
| 3824 | * int err = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE_EXT, &create_ext); | ||
| 3825 | * if (err) ... | ||
| 3826 | */ | ||
| 3827 | struct drm_i915_gem_create_ext_set_pat { | ||
| 3828 | /** @base: Extension link. See struct i915_user_extension. */ | ||
| 3829 | struct i915_user_extension base; | ||
| 3830 | /** | ||
| 3831 | * @pat_index: PAT index to be set | ||
| 3832 | * PAT index is a bit field in Page Table Entry to control caching | ||
| 3833 | * behaviors for GPU accesses. The definition of PAT index is | ||
| 3834 | * platform dependent and can be found in hardware specifications, | ||
| 3835 | */ | ||
| 3836 | __u32 pat_index; | ||
| 3837 | /** @rsvd: reserved for future use */ | ||
| 3838 | __u32 rsvd; | ||
| 3839 | }; | ||
| 3840 | |||
| 3800 | /* ID of the protected content session managed by i915 when PXP is active */ | 3841 | /* ID of the protected content session managed by i915 when PXP is active */ |
| 3801 | #define I915_PROTECTED_CONTENT_DEFAULT_SESSION 0xf | 3842 | #define I915_PROTECTED_CONTENT_DEFAULT_SESSION 0xf |
| 3802 | 3843 | ||