Re: [Intel-gfx] [PATCH v15 1/1] drm/i915: Allow user to set cache at BO creation
Quoting Yang, Fei (2023-06-06 09:51:06) > >> On 31/05/2023 18:10, fei.y...@intel.com wrote: > >>> From: Fei Yang > >>> > >>> 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. > >>> > >>> Note: The detailed description of PAT index is missing in current PRM > >>> even for older hardware and will be added by the next PRM update under > >>> chapter name "Memory Views". > >>> > >>> BSpec: 45101 > >>> > >>> Mesa support has been submitted in this merge request: > >>> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878 > >>> > >>> The media driver is supported by the following commits: > >>> https://github.com/intel/media-driver/commit/ > 92c00a857433ebb34ec575e9834f473c6fcb6341 > >>> https://github.com/intel/media-driver/commit/ > fd375cf2c5e1f6bf6b43258ff797b3134aadc9fd > >>> https://github.com/intel/media-driver/commit/ > 08dd244b22484770a33464c2c8ae85430e548000 We absolutely should not have merged this code to master branch yet. These should be reverted immediately and any releases that include the code must be pulled back. This is clearly explained in: https://www.kernel.org/doc/html/latest/gpu/drm-uapi.html#open-source-userspace-requirements "The kernel patch can only be merged after all the above requirements are met, but it *must* be merged to either drm-next or drm-misc-next *before* the userspace patches land. uAPI always flows from the kernel, doing things the other way round risks divergence of the uAPI definitions and header files." > >> On which platforms will media-driver use the uapi? I couldn't easily > >> figure out myself from the links above and also in the master branch I > >> couldn't find the implementation of CachePolicyGetPATIndex. > > > > These commits look like platform independent. Carl, could you chime in here? > > Confirmed with Carl and Lihao offline that the media driver is calling set_pat > extension in common code path, so the use of set_pat extension is platform > independent. The only problem right now is that the gmm library is not > returning > correct PAT index for all hardware platforms, so on some platforms the call > would > be bypassed and fall back to the old way. That means the code is unused for older platforms. The fact that there is potential to be used is not alone a reason for merging it. So I agree with Tvrtko that we should only limit this to the newer platforms where we have actual use that is ready and reviewed. We can extend to older platforms later, but in order not to block the progress please move the code for older platform to later series and only apply to platforms where this is needed. > I think this is the correct implementation. It should be platform independent > as > long as the application knows what PAT index to set. Updating the gmm library > to > understand PAT index for each hardware platform is a separate issue. If we don't have userspace ready, we don't merge the code. Regards, Joonas > >> Now that PRMs for Tigerlake have been published and Meteorlake situation > >> is documented indirectly in Mesa code, my only remaining concern is with > >> the older platforms. So if there is no particular reason to have the > >> extension working on those, I would strongly suggest we disable there. > > > > What's the concern? There is no change required for older platforms, > > existing > > user space code should continue to work. And this extension should be made > > available for any new development because the cache settings for BO's need > > to be immutable. And that is platform independent. > > > >> For a precedent see I915_CONTEXT_PARAM_SSEU and how it allows the > >> extension only on Gen11 and only for a very specific usecase (see > >> restrictions in set_sseu() and i915_gem_user_to_context_sseu()). > >> > >> Regards, > >> > >> Tvrtko > >> > >>> > >>> The IGT test related to this change is > >>> igt@gem_create@create-ext-set-pat > >>> > >>> Signed-off-by: Fei Yang > >>> Cc: Chris Wilson > >>> Cc: Matt Roper > >>> Cc: Andi Shyti > >>> Reviewed-by: Andi Shyti > >>> Acked-by: Jordan Justen > >>> Tested-by: Jordan Justen > >>> Acked-by: Carl Zhang > >>> Tested-by: Lihao Gu > >>> --- > >>> drivers/gpu/drm/i915/gem/i915_gem_create.c | 36 +++ > >>> drivers/gpu/drm/i915/gem/i915_gem_object.c
Re: [Intel-gfx] [PATCH v15 1/1] drm/i915: Allow user to set cache at BO creation
>> On 31/05/2023 18:10, fei.y...@intel.com wrote: >>> From: Fei Yang >>> >>> 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. >>> >>> Note: The detailed description of PAT index is missing in current PRM >>> even for older hardware and will be added by the next PRM update under >>> chapter name "Memory Views". >>> >>> BSpec: 45101 >>> >>> Mesa support has been submitted in this merge request: >>> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878 >>> >>> The media driver is supported by the following commits: >>> https://github.com/intel/media-driver/commit/92c00a857433ebb34ec575e9834f473c6fcb6341 >>> https://github.com/intel/media-driver/commit/fd375cf2c5e1f6bf6b43258ff797b3134aadc9fd >>> https://github.com/intel/media-driver/commit/08dd244b22484770a33464c2c8ae85430e548000 >> >> On which platforms will media-driver use the uapi? I couldn't easily >> figure out myself from the links above and also in the master branch I >> couldn't find the implementation of CachePolicyGetPATIndex. > > These commits look like platform independent. Carl, could you chime in here? Confirmed with Carl and Lihao offline that the media driver is calling set_pat extension in common code path, so the use of set_pat extension is platform independent. The only problem right now is that the gmm library is not returning correct PAT index for all hardware platforms, so on some platforms the call would be bypassed and fall back to the old way. I think this is the correct implementation. It should be platform independent as long as the application knows what PAT index to set. Updating the gmm library to understand PAT index for each hardware platform is a separate issue. >> Now that PRMs for Tigerlake have been published and Meteorlake situation >> is documented indirectly in Mesa code, my only remaining concern is with >> the older platforms. So if there is no particular reason to have the >> extension working on those, I would strongly suggest we disable there. > > What's the concern? There is no change required for older platforms, existing > user space code should continue to work. And this extension should be made > available for any new development because the cache settings for BO's need > to be immutable. And that is platform independent. > >> For a precedent see I915_CONTEXT_PARAM_SSEU and how it allows the >> extension only on Gen11 and only for a very specific usecase (see >> restrictions in set_sseu() and i915_gem_user_to_context_sseu()). >> >> Regards, >> >> Tvrtko >> >>> >>> The IGT test related to this change is >>> igt@gem_create@create-ext-set-pat >>> >>> Signed-off-by: Fei Yang >>> Cc: Chris Wilson >>> Cc: Matt Roper >>> Cc: Andi Shyti >>> Reviewed-by: Andi Shyti >>> Acked-by: Jordan Justen >>> Tested-by: Jordan Justen >>> Acked-by: Carl Zhang >>> Tested-by: Lihao Gu >>> --- >>> drivers/gpu/drm/i915/gem/i915_gem_create.c | 36 +++ >>> drivers/gpu/drm/i915/gem/i915_gem_object.c | 6 >>> include/uapi/drm/i915_drm.h| 41 ++ >>> 3 files changed, 83 insertions(+) >>> >>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_create.c >>> b/drivers/gpu/drm/i915/gem/i915_gem_create.c >>> index bfe1dbda4cb7..644a936248ad 100644 >>> --- a/drivers/gpu/drm/i915/gem/i915_gem_create.c >>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_create.c >>> @@ -245,6 +245,7 @@ struct create_ext { >>>unsigned int n_placements; >>>unsigned int placement_mask; >>>unsigned long flags; >>> + unsigned int pat_index; >>> }; >>> >>> static void repr_placements(char *buf, size_t size, >>> @@ -394,11 +395,39 @@ static int ext_set_protected(struct >>> i915_user_extension __user *base, void *data >>>return 0; >>> } >>> >>> +static int ext_set_pat(struct i915_user_extension __user *base, void *data) >>> +{ >>> + struct create_ext *ext_data = data; >>> + struct drm_i915_private *i915 = ext_data->i915; >>> + struct drm_i915_gem_create_ext_set_pat ext; >>> + unsigned int max_pat_index; >>> + >>> + BUILD_BUG_ON(sizeof(struct drm_i915_gem_create_ext_set_pat) != >>> + offsetofend(struct drm_i915_gem_create_ext_set_pat, >>> rsvd)); >>> + >>> + if (copy_from_user(, base, sizeof(ext))) >>> + return -EFAULT;
Re: [Intel-gfx] [PATCH v15 1/1] drm/i915: Allow user to set cache at BO creation
> On 31/05/2023 18:10, fei.y...@intel.com wrote: >> From: Fei Yang >> >> 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. >> >> Note: The detailed description of PAT index is missing in current PRM >> even for older hardware and will be added by the next PRM update under >> chapter name "Memory Views". >> >> BSpec: 45101 >> >> Mesa support has been submitted in this merge request: >> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878 >> >> The media driver is supported by the following commits: >> https://github.com/intel/media-driver/commit/92c00a857433ebb34ec575e9834f473c6fcb6341 >> https://github.com/intel/media-driver/commit/fd375cf2c5e1f6bf6b43258ff797b3134aadc9fd >> https://github.com/intel/media-driver/commit/08dd244b22484770a33464c2c8ae85430e548000 > > On which platforms will media-driver use the uapi? I couldn't easily > figure out myself from the links above and also in the master branch I > couldn't find the implementation of CachePolicyGetPATIndex. These commits look like platform independent. Carl, could you chime in here? > Now that PRMs for Tigerlake have been published and Meteorlake situation > is documented indirectly in Mesa code, my only remaining concern is with > the older platforms. So if there is no particular reason to have the > extension working on those, I would strongly suggest we disable there. What's the concern? There is no change required for older platforms, existing user space code should continue to work. And this extension should be made available for any new development because the cache settings for BO's need to be immutable. And that is platform independent. > For a precedent see I915_CONTEXT_PARAM_SSEU and how it allows the > extension only on Gen11 and only for a very specific usecase (see > restrictions in set_sseu() and i915_gem_user_to_context_sseu()). > > Regards, > > Tvrtko > >> >> The IGT test related to this change is >> igt@gem_create@create-ext-set-pat >> >> Signed-off-by: Fei Yang >> Cc: Chris Wilson >> Cc: Matt Roper >> Cc: Andi Shyti >> Reviewed-by: Andi Shyti >> Acked-by: Jordan Justen >> Tested-by: Jordan Justen >> Acked-by: Carl Zhang >> Tested-by: Lihao Gu >> --- >> drivers/gpu/drm/i915/gem/i915_gem_create.c | 36 +++ >> drivers/gpu/drm/i915/gem/i915_gem_object.c | 6 >> include/uapi/drm/i915_drm.h| 41 ++ >> 3 files changed, 83 insertions(+) >> >> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_create.c >> b/drivers/gpu/drm/i915/gem/i915_gem_create.c >> index bfe1dbda4cb7..644a936248ad 100644 >> --- a/drivers/gpu/drm/i915/gem/i915_gem_create.c >> +++ b/drivers/gpu/drm/i915/gem/i915_gem_create.c >> @@ -245,6 +245,7 @@ struct create_ext { >>unsigned int n_placements; >>unsigned int placement_mask; >>unsigned long flags; >> + unsigned int pat_index; >> }; >> >> static void repr_placements(char *buf, size_t size, >> @@ -394,11 +395,39 @@ static int ext_set_protected(struct >> i915_user_extension __user *base, void *data >>return 0; >> } >> >> +static int ext_set_pat(struct i915_user_extension __user *base, void *data) >> +{ >> + struct create_ext *ext_data = data; >> + struct drm_i915_private *i915 = ext_data->i915; >> + struct drm_i915_gem_create_ext_set_pat ext; >> + unsigned int max_pat_index; >> + >> + BUILD_BUG_ON(sizeof(struct drm_i915_gem_create_ext_set_pat) != >> + offsetofend(struct drm_i915_gem_create_ext_set_pat, >> rsvd)); >> + >> + if (copy_from_user(, base, sizeof(ext))) >> + return -EFAULT; >> + >> + max_pat_index = INTEL_INFO(i915)->max_pat_index; >> + >> + if (ext.pat_index > max_pat_index) { >> + drm_dbg(>drm, "PAT index is invalid: %u\n", >> + ext.pat_index); >> + return -EINVAL; >> + } >> + >> + ext_data->pat_index = ext.pat_index; >> + >> + return 0; >> +} >> + >> static const i915_user_extension_fn create_extensions[] = { >>[I915_GEM_CREATE_EXT_MEMORY_REGIONS] = ext_set_placements, >>[I915_GEM_CREATE_EXT_PROTECTED_CONTENT] = ext_set_protected, >> + [I915_GEM_CREATE_EXT_SET_PAT] = ext_set_pat, >> }; >> >> +#define PAT_INDEX_NOT_SET0x >> /** >>* i915_gem_create_ext_ioctl
Re: [Intel-gfx] [PATCH v15 1/1] drm/i915: Allow user to set cache at BO creation
On 31/05/2023 18:10, fei.y...@intel.com wrote: From: Fei Yang 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. Note: The detailed description of PAT index is missing in current PRM even for older hardware and will be added by the next PRM update under chapter name "Memory Views". BSpec: 45101 Mesa support has been submitted in this merge request: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878 The media driver is supported by the following commits: https://github.com/intel/media-driver/commit/92c00a857433ebb34ec575e9834f473c6fcb6341 https://github.com/intel/media-driver/commit/fd375cf2c5e1f6bf6b43258ff797b3134aadc9fd https://github.com/intel/media-driver/commit/08dd244b22484770a33464c2c8ae85430e548000 On which platforms will media-driver use the uapi? I couldn't easily figure out myself from the links above and also in the master branch I couldn't find the implementation of CachePolicyGetPATIndex. Now that PRMs for Tigerlake have been published and Meteorlake situation is documented indirectly in Mesa code, my only remaining concern is with the older platforms. So if there is no particular reason to have the extension working on those, I would strongly suggest we disable there. For a precedent see I915_CONTEXT_PARAM_SSEU and how it allows the extension only on Gen11 and only for a very specific usecase (see restrictions in set_sseu() and i915_gem_user_to_context_sseu()). Regards, Tvrtko The IGT test related to this change is igt@gem_create@create-ext-set-pat Signed-off-by: Fei Yang Cc: Chris Wilson Cc: Matt Roper Cc: Andi Shyti Reviewed-by: Andi Shyti Acked-by: Jordan Justen Tested-by: Jordan Justen Acked-by: Carl Zhang Tested-by: Lihao Gu --- drivers/gpu/drm/i915/gem/i915_gem_create.c | 36 +++ drivers/gpu/drm/i915/gem/i915_gem_object.c | 6 include/uapi/drm/i915_drm.h| 41 ++ 3 files changed, 83 insertions(+) diff --git a/drivers/gpu/drm/i915/gem/i915_gem_create.c b/drivers/gpu/drm/i915/gem/i915_gem_create.c index bfe1dbda4cb7..644a936248ad 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_create.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_create.c @@ -245,6 +245,7 @@ struct create_ext { unsigned int n_placements; unsigned int placement_mask; unsigned long flags; + unsigned int pat_index; }; static void repr_placements(char *buf, size_t size, @@ -394,11 +395,39 @@ static int ext_set_protected(struct i915_user_extension __user *base, void *data return 0; } +static int ext_set_pat(struct i915_user_extension __user *base, void *data) +{ + struct create_ext *ext_data = data; + struct drm_i915_private *i915 = ext_data->i915; + struct drm_i915_gem_create_ext_set_pat ext; + unsigned int max_pat_index; + + BUILD_BUG_ON(sizeof(struct drm_i915_gem_create_ext_set_pat) != +offsetofend(struct drm_i915_gem_create_ext_set_pat, rsvd)); + + if (copy_from_user(, base, sizeof(ext))) + return -EFAULT; + + max_pat_index = INTEL_INFO(i915)->max_pat_index; + + if (ext.pat_index > max_pat_index) { + drm_dbg(>drm, "PAT index is invalid: %u\n", + ext.pat_index); + return -EINVAL; + } + + ext_data->pat_index = ext.pat_index; + + return 0; +} + static const i915_user_extension_fn create_extensions[] = { [I915_GEM_CREATE_EXT_MEMORY_REGIONS] = ext_set_placements, [I915_GEM_CREATE_EXT_PROTECTED_CONTENT] = ext_set_protected, + [I915_GEM_CREATE_EXT_SET_PAT] = ext_set_pat, }; +#define PAT_INDEX_NOT_SET 0x /** * i915_gem_create_ext_ioctl - Creates a new mm object and returns a handle to it. * @dev: drm device pointer @@ -418,6 +447,7 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data, if (args->flags & ~I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS) return -EINVAL; + ext_data.pat_index = PAT_INDEX_NOT_SET; ret = i915_user_extensions(u64_to_user_ptr(args->extensions), create_extensions, ARRAY_SIZE(create_extensions), @@ -454,5 +484,11 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data, if
Re: [PATCH v15 1/1] drm/i915: Allow user to set cache at BO creation
> Hi Fei, > > On Wed, May 31, 2023 at 10:10:08AM -0700, fei.y...@intel.com wrote: >> From: Fei Yang >> >> 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. >> >> Note: The detailed description of PAT index is missing in current PRM >> even for older hardware and will be added by the next PRM update under >> chapter name "Memory Views". > > Documentation has been updated: > > https://www.intel.com/content/www/us/en/docs/graphics-for-linux/developer-reference/1-0/tiger-lake.html > > If it's OK with you, before pushing I can replace this Note with: > >" >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 >" Looks good to me. Thank you Andi and Tvrtko for all your help. -Fei >Thank you Tvrtko for the intution you had about the documentation >and for pushing for the update. It is greate to have this uAPI >well documented! > >Andi > >> BSpec: 45101 >> >> Mesa support has been submitted in this merge request: >> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878 >> >> The media driver is supported by the following commits: >> https://github.com/intel/media-driver/commit/92c00a857433ebb34ec575e9834f473c6fcb6341 >> https://github.com/intel/media-driver/commit/fd375cf2c5e1f6bf6b43258ff797b3134aadc9fd >> https://github.com/intel/media-driver/commit/08dd244b22484770a33464c2c8ae85430e548000 >> >> The IGT test related to this change is >> igt@gem_create@create-ext-set-pat >> >> Signed-off-by: Fei Yang >> Cc: Chris Wilson >> Cc: Matt Roper >> Cc: Andi Shyti >> Reviewed-by: Andi Shyti >> Acked-by: Jordan Justen >> Tested-by: Jordan Justen >> Acked-by: Carl Zhang >> Tested-by: Lihao Gu >> --- >> drivers/gpu/drm/i915/gem/i915_gem_create.c | 36 +++ >> drivers/gpu/drm/i915/gem/i915_gem_object.c | 6 >> include/uapi/drm/i915_drm.h| 41 ++ >> 3 files changed, 83 insertions(+) >> >> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_create.c >> b/drivers/gpu/drm/i915/gem/i915_gem_create.c >> index bfe1dbda4cb7..644a936248ad 100644 >> --- a/drivers/gpu/drm/i915/gem/i915_gem_create.c >> +++ b/drivers/gpu/drm/i915/gem/i915_gem_create.c >> @@ -245,6 +245,7 @@ struct create_ext { >>unsigned int n_placements; >>unsigned int placement_mask; >>unsigned long flags; >> + unsigned int pat_index; >> }; >> >> static void repr_placements(char *buf, size_t size, >> @@ -394,11 +395,39 @@ static int ext_set_protected(struct >> i915_user_extension __user *base, void *data >>return 0; >> } >> >> +static int ext_set_pat(struct i915_user_extension __user *base, void *data) >> +{ >> + struct create_ext *ext_data = data; >> + struct drm_i915_private *i915 = ext_data->i915; >> + struct drm_i915_gem_create_ext_set_pat ext; >> + unsigned int max_pat_index; >> + >> + BUILD_BUG_ON(sizeof(struct drm_i915_gem_create_ext_set_pat) != >> + offsetofend(struct drm_i915_gem_create_ext_set_pat, >> rsvd)); >> + >> + if (copy_from_user(, base, sizeof(ext))) >> + return -EFAULT; >> + >> + max_pat_index = INTEL_INFO(i915)->max_pat_index; >> + >> + if (ext.pat_index > max_pat_index) { >> + drm_dbg(>drm, "PAT index is invalid: %u\n", >> + ext.pat_index); >> + return -EINVAL; >> + } >> + >> + ext_data->pat_index = ext.pat_index; >> + >> + return 0; >> +} >> + >> static const i915_user_extension_fn create_extensions[] = { >>[I915_GEM_CREATE_EXT_MEMORY_REGIONS] = ext_set_placements, >>[I915_GEM_CREATE_EXT_PROTECTED_CONTENT] = ext_set_protected, >> + [I915_GEM_CREATE_EXT_SET_PAT] = ext_set_pat, >> }; >> >> +#define PAT_INDEX_NOT_SET0x >> /** >> * i915_gem_create_ext_ioctl - Creates a new mm object and returns a handle >> to it. >> * @dev: drm device pointer >> @@ -418,6 +447,7 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void >> *data, >>if (args->flags & ~I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS) >>return -EINVAL; >> >> + ext_data.pat_index = PAT_INDEX_NOT_SET; >>ret = i915_user_extensions(u64_to_user_ptr(args->extensions), >>
Re: [PATCH v15 1/1] drm/i915: Allow user to set cache at BO creation
Hi Fei, On Wed, May 31, 2023 at 10:10:08AM -0700, fei.y...@intel.com wrote: > From: Fei Yang > > 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. > > Note: The detailed description of PAT index is missing in current PRM > even for older hardware and will be added by the next PRM update under > chapter name "Memory Views". Documentation has been updated: https://www.intel.com/content/www/us/en/docs/graphics-for-linux/developer-reference/1-0/tiger-lake.html If it's OK with you, before pushing I can replace this Note with: " 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 " Thank you Tvrtko for the intution you had about the documentation and for pushing for the update. It is greate to have this uAPI well documented! Andi > BSpec: 45101 > > Mesa support has been submitted in this merge request: > https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878 > > The media driver is supported by the following commits: > https://github.com/intel/media-driver/commit/92c00a857433ebb34ec575e9834f473c6fcb6341 > https://github.com/intel/media-driver/commit/fd375cf2c5e1f6bf6b43258ff797b3134aadc9fd > https://github.com/intel/media-driver/commit/08dd244b22484770a33464c2c8ae85430e548000 > > The IGT test related to this change is > igt@gem_create@create-ext-set-pat > > Signed-off-by: Fei Yang > Cc: Chris Wilson > Cc: Matt Roper > Cc: Andi Shyti > Reviewed-by: Andi Shyti > Acked-by: Jordan Justen > Tested-by: Jordan Justen > Acked-by: Carl Zhang > Tested-by: Lihao Gu > --- > drivers/gpu/drm/i915/gem/i915_gem_create.c | 36 +++ > drivers/gpu/drm/i915/gem/i915_gem_object.c | 6 > include/uapi/drm/i915_drm.h| 41 ++ > 3 files changed, 83 insertions(+) > > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_create.c > b/drivers/gpu/drm/i915/gem/i915_gem_create.c > index bfe1dbda4cb7..644a936248ad 100644 > --- a/drivers/gpu/drm/i915/gem/i915_gem_create.c > +++ b/drivers/gpu/drm/i915/gem/i915_gem_create.c > @@ -245,6 +245,7 @@ struct create_ext { > unsigned int n_placements; > unsigned int placement_mask; > unsigned long flags; > + unsigned int pat_index; > }; > > static void repr_placements(char *buf, size_t size, > @@ -394,11 +395,39 @@ static int ext_set_protected(struct i915_user_extension > __user *base, void *data > return 0; > } > > +static int ext_set_pat(struct i915_user_extension __user *base, void *data) > +{ > + struct create_ext *ext_data = data; > + struct drm_i915_private *i915 = ext_data->i915; > + struct drm_i915_gem_create_ext_set_pat ext; > + unsigned int max_pat_index; > + > + BUILD_BUG_ON(sizeof(struct drm_i915_gem_create_ext_set_pat) != > + offsetofend(struct drm_i915_gem_create_ext_set_pat, rsvd)); > + > + if (copy_from_user(, base, sizeof(ext))) > + return -EFAULT; > + > + max_pat_index = INTEL_INFO(i915)->max_pat_index; > + > + if (ext.pat_index > max_pat_index) { > + drm_dbg(>drm, "PAT index is invalid: %u\n", > + ext.pat_index); > + return -EINVAL; > + } > + > + ext_data->pat_index = ext.pat_index; > + > + return 0; > +} > + > static const i915_user_extension_fn create_extensions[] = { > [I915_GEM_CREATE_EXT_MEMORY_REGIONS] = ext_set_placements, > [I915_GEM_CREATE_EXT_PROTECTED_CONTENT] = ext_set_protected, > + [I915_GEM_CREATE_EXT_SET_PAT] = ext_set_pat, > }; > > +#define PAT_INDEX_NOT_SET0x > /** > * i915_gem_create_ext_ioctl - Creates a new mm object and returns a handle > to it. > * @dev: drm device pointer > @@ -418,6 +447,7 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void > *data, > if (args->flags & ~I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS) > return -EINVAL; > > + ext_data.pat_index = PAT_INDEX_NOT_SET; > ret = i915_user_extensions(u64_to_user_ptr(args->extensions), > create_extensions, > ARRAY_SIZE(create_extensions), > @@ -454,5 +484,11 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void > *data, >
[PATCH v15 1/1] drm/i915: Allow user to set cache at BO creation
From: Fei Yang 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. Note: The detailed description of PAT index is missing in current PRM even for older hardware and will be added by the next PRM update under chapter name "Memory Views". BSpec: 45101 Mesa support has been submitted in this merge request: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878 The media driver is supported by the following commits: https://github.com/intel/media-driver/commit/92c00a857433ebb34ec575e9834f473c6fcb6341 https://github.com/intel/media-driver/commit/fd375cf2c5e1f6bf6b43258ff797b3134aadc9fd https://github.com/intel/media-driver/commit/08dd244b22484770a33464c2c8ae85430e548000 The IGT test related to this change is igt@gem_create@create-ext-set-pat Signed-off-by: Fei Yang Cc: Chris Wilson Cc: Matt Roper Cc: Andi Shyti Reviewed-by: Andi Shyti Acked-by: Jordan Justen Tested-by: Jordan Justen Acked-by: Carl Zhang Tested-by: Lihao Gu --- drivers/gpu/drm/i915/gem/i915_gem_create.c | 36 +++ drivers/gpu/drm/i915/gem/i915_gem_object.c | 6 include/uapi/drm/i915_drm.h| 41 ++ 3 files changed, 83 insertions(+) diff --git a/drivers/gpu/drm/i915/gem/i915_gem_create.c b/drivers/gpu/drm/i915/gem/i915_gem_create.c index bfe1dbda4cb7..644a936248ad 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_create.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_create.c @@ -245,6 +245,7 @@ struct create_ext { unsigned int n_placements; unsigned int placement_mask; unsigned long flags; + unsigned int pat_index; }; static void repr_placements(char *buf, size_t size, @@ -394,11 +395,39 @@ static int ext_set_protected(struct i915_user_extension __user *base, void *data return 0; } +static int ext_set_pat(struct i915_user_extension __user *base, void *data) +{ + struct create_ext *ext_data = data; + struct drm_i915_private *i915 = ext_data->i915; + struct drm_i915_gem_create_ext_set_pat ext; + unsigned int max_pat_index; + + BUILD_BUG_ON(sizeof(struct drm_i915_gem_create_ext_set_pat) != +offsetofend(struct drm_i915_gem_create_ext_set_pat, rsvd)); + + if (copy_from_user(, base, sizeof(ext))) + return -EFAULT; + + max_pat_index = INTEL_INFO(i915)->max_pat_index; + + if (ext.pat_index > max_pat_index) { + drm_dbg(>drm, "PAT index is invalid: %u\n", + ext.pat_index); + return -EINVAL; + } + + ext_data->pat_index = ext.pat_index; + + return 0; +} + static const i915_user_extension_fn create_extensions[] = { [I915_GEM_CREATE_EXT_MEMORY_REGIONS] = ext_set_placements, [I915_GEM_CREATE_EXT_PROTECTED_CONTENT] = ext_set_protected, + [I915_GEM_CREATE_EXT_SET_PAT] = ext_set_pat, }; +#define PAT_INDEX_NOT_SET 0x /** * i915_gem_create_ext_ioctl - Creates a new mm object and returns a handle to it. * @dev: drm device pointer @@ -418,6 +447,7 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data, if (args->flags & ~I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS) return -EINVAL; + ext_data.pat_index = PAT_INDEX_NOT_SET; ret = i915_user_extensions(u64_to_user_ptr(args->extensions), create_extensions, ARRAY_SIZE(create_extensions), @@ -454,5 +484,11 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data, if (IS_ERR(obj)) return PTR_ERR(obj); + if (ext_data.pat_index != PAT_INDEX_NOT_SET) { + i915_gem_object_set_pat_index(obj, ext_data.pat_index); + /* Mark pat_index is set by UMD */ + obj->pat_set_by_user = true; + } + return i915_gem_publish(obj, file, >size, >handle); } diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.c b/drivers/gpu/drm/i915/gem/i915_gem_object.c index 46a19b099ec8..97ac6fb37958 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_object.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_object.c @@ -208,6 +208,12 @@ bool i915_gem_object_can_bypass_llc(struct drm_i915_gem_object *obj) if (!(obj->flags & I915_BO_ALLOC_USER)) return false; + /* +*