| Message ID | 20190724105626.53552-1-steven.price@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/panfrost: Export all GPU feature registers | expand |
Adding Alyssa's Collabora email. On Wed, Jul 24, 2019 at 4:56 AM Steven Price <steven.price@arm.com> wrote: > > Midgard/Bifrost GPUs have a bunch of feature registers providing details > of what the hardware supports. Panfrost already reads these, this patch > exports them all to user space so that the jobs created by the user space > driver can be tuned for the particular hardware implementation. > > Signed-off-by: Steven Price <steven.price@arm.com> > --- > drivers/gpu/drm/panfrost/panfrost_device.h | 1 + > drivers/gpu/drm/panfrost/panfrost_drv.c | 38 +++++++++++++++++++-- > drivers/gpu/drm/panfrost/panfrost_gpu.c | 2 ++ > include/uapi/drm/panfrost_drm.h | 39 ++++++++++++++++++++++ > 4 files changed, 77 insertions(+), 3 deletions(-) LGTM. I'll give it a bit more time to see if there are any comments before I apply it. Rob
This is definitely helpful! My one concern is, supposing userspace really does need all of this information, is it wasteful to have to do 30+ ioctls just to get this? kbase had a single ioctl to grab all of the properties, whether userspace wanted them or not. I'm not sure if that's better -- the two approaches are rather polar opposites. Granted this would be on driver init so not a critical path. On Wed, Jul 24, 2019 at 10:27:03AM -0600, Rob Herring wrote: > Adding Alyssa's Collabora email. > > On Wed, Jul 24, 2019 at 4:56 AM Steven Price <steven.price@arm.com> wrote: > > > > Midgard/Bifrost GPUs have a bunch of feature registers providing details > > of what the hardware supports. Panfrost already reads these, this patch > > exports them all to user space so that the jobs created by the user space > > driver can be tuned for the particular hardware implementation. > > > > Signed-off-by: Steven Price <steven.price@arm.com> > > --- > > drivers/gpu/drm/panfrost/panfrost_device.h | 1 + > > drivers/gpu/drm/panfrost/panfrost_drv.c | 38 +++++++++++++++++++-- > > drivers/gpu/drm/panfrost/panfrost_gpu.c | 2 ++ > > include/uapi/drm/panfrost_drm.h | 39 ++++++++++++++++++++++ > > 4 files changed, 77 insertions(+), 3 deletions(-) > > LGTM. I'll give it a bit more time to see if there are any comments > before I apply it. > > Rob
On Wed, Jul 24, 2019 at 10:40 AM Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com> wrote: > > This is definitely helpful! > > My one concern is, supposing userspace really does need all of this > information, is it wasteful to have to do 30+ ioctls just to get this? > kbase had a single ioctl to grab all of the properties, whether > userspace wanted them or not. I'm not sure if that's better -- the two > approaches are rather polar opposites. I think this ship already sailed when we added the first one with GPU_ID. Also, at least etnaviv works the same way. > > Granted this would be on driver init so not a critical path. Exactly. > > On Wed, Jul 24, 2019 at 10:27:03AM -0600, Rob Herring wrote: > > Adding Alyssa's Collabora email. > > > > On Wed, Jul 24, 2019 at 4:56 AM Steven Price <steven.price@arm.com> wrote: > > > > > > Midgard/Bifrost GPUs have a bunch of feature registers providing details > > > of what the hardware supports. Panfrost already reads these, this patch > > > exports them all to user space so that the jobs created by the user space > > > driver can be tuned for the particular hardware implementation. > > > > > > Signed-off-by: Steven Price <steven.price@arm.com> > > > --- > > > drivers/gpu/drm/panfrost/panfrost_device.h | 1 + > > > drivers/gpu/drm/panfrost/panfrost_drv.c | 38 +++++++++++++++++++-- > > > drivers/gpu/drm/panfrost/panfrost_gpu.c | 2 ++ > > > include/uapi/drm/panfrost_drm.h | 39 ++++++++++++++++++++++ > > > 4 files changed, 77 insertions(+), 3 deletions(-) > > > > LGTM. I'll give it a bit more time to see if there are any comments > > before I apply it. > > > > Rob
> I think this ship already sailed when we added the first one with > GPU_ID. Also, at least etnaviv works the same way. Fair enough then! Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
diff --git a/drivers/gpu/drm/panfrost/panfrost_device.h b/drivers/gpu/drm/panfrost/panfrost_device.h index 83cc01cafde1..ea5948ff3647 100644 --- a/drivers/gpu/drm/panfrost/panfrost_device.h +++ b/drivers/gpu/drm/panfrost/panfrost_device.h @@ -43,6 +43,7 @@ struct panfrost_features { u32 js_features[16]; u32 nr_core_groups; + u32 thread_tls_alloc; unsigned long hw_features[64 / BITS_PER_LONG]; unsigned long hw_issues[64 / BITS_PER_LONG]; diff --git a/drivers/gpu/drm/panfrost/panfrost_drv.c b/drivers/gpu/drm/panfrost/panfrost_drv.c index 85b4b51b6a0d..4b554c8d56da 100644 --- a/drivers/gpu/drm/panfrost/panfrost_drv.c +++ b/drivers/gpu/drm/panfrost/panfrost_drv.c @@ -32,10 +32,42 @@ static int panfrost_ioctl_get_param(struct drm_device *ddev, void *data, struct if (param->pad != 0) return -EINVAL; +#define PANFROST_FEATURE(name, member) \ + case DRM_PANFROST_PARAM_ ## name: \ + param->value = pfdev->features.member; \ + break +#define PANFROST_FEATURE_ARRAY(name, member, max) \ + case DRM_PANFROST_PARAM_ ## name ## 0 ... \ + DRM_PANFROST_PARAM_ ## name ## max: \ + param->value = pfdev->features.member[param->param - \ + DRM_PANFROST_PARAM_ ## name ## 0]; \ + break + switch (param->param) { - case DRM_PANFROST_PARAM_GPU_PROD_ID: - param->value = pfdev->features.id; - break; + PANFROST_FEATURE(GPU_PROD_ID, id); + PANFROST_FEATURE(GPU_REVISION, revision); + PANFROST_FEATURE(SHADER_PRESENT, shader_present); + PANFROST_FEATURE(TILER_PRESENT, tiler_present); + PANFROST_FEATURE(L2_PRESENT, l2_present); + PANFROST_FEATURE(STACK_PRESENT, stack_present); + PANFROST_FEATURE(AS_PRESENT, as_present); + PANFROST_FEATURE(JS_PRESENT, js_present); + PANFROST_FEATURE(L2_FEATURES, l2_features); + PANFROST_FEATURE(CORE_FEATURES, core_features); + PANFROST_FEATURE(TILER_FEATURES, tiler_features); + PANFROST_FEATURE(MEM_FEATURES, mem_features); + PANFROST_FEATURE(MMU_FEATURES, mmu_features); + PANFROST_FEATURE(THREAD_FEATURES, thread_features); + PANFROST_FEATURE(MAX_THREADS, max_threads); + PANFROST_FEATURE(THREAD_MAX_WORKGROUP_SZ, + thread_max_workgroup_sz); + PANFROST_FEATURE(THREAD_MAX_BARRIER_SZ, + thread_max_barrier_sz); + PANFROST_FEATURE(COHERENCY_FEATURES, coherency_features); + PANFROST_FEATURE_ARRAY(TEXTURE_FEATURES, texture_features, 3); + PANFROST_FEATURE_ARRAY(JS_FEATURES, js_features, 15); + PANFROST_FEATURE(NR_CORE_GROUPS, nr_core_groups); + PANFROST_FEATURE(THREAD_TLS_ALLOC, thread_tls_alloc); default: return -EINVAL; } diff --git a/drivers/gpu/drm/panfrost/panfrost_gpu.c b/drivers/gpu/drm/panfrost/panfrost_gpu.c index 20ab333fc925..f67ed925c0ef 100644 --- a/drivers/gpu/drm/panfrost/panfrost_gpu.c +++ b/drivers/gpu/drm/panfrost/panfrost_gpu.c @@ -232,6 +232,8 @@ static void panfrost_gpu_init_features(struct panfrost_device *pfdev) pfdev->features.stack_present = gpu_read(pfdev, GPU_STACK_PRESENT_LO); pfdev->features.stack_present |= (u64)gpu_read(pfdev, GPU_STACK_PRESENT_HI) << 32; + pfdev->features.thread_tls_alloc = gpu_read(pfdev, GPU_THREAD_TLS_ALLOC); + gpu_id = gpu_read(pfdev, GPU_ID); pfdev->features.revision = gpu_id & 0xffff; pfdev->features.id = gpu_id >> 16; diff --git a/include/uapi/drm/panfrost_drm.h b/include/uapi/drm/panfrost_drm.h index b5d370638846..cb577fb96b38 100644 --- a/include/uapi/drm/panfrost_drm.h +++ b/include/uapi/drm/panfrost_drm.h @@ -127,6 +127,45 @@ struct drm_panfrost_mmap_bo { enum drm_panfrost_param { DRM_PANFROST_PARAM_GPU_PROD_ID, + DRM_PANFROST_PARAM_GPU_REVISION, + DRM_PANFROST_PARAM_SHADER_PRESENT, + DRM_PANFROST_PARAM_TILER_PRESENT, + DRM_PANFROST_PARAM_L2_PRESENT, + DRM_PANFROST_PARAM_STACK_PRESENT, + DRM_PANFROST_PARAM_AS_PRESENT, + DRM_PANFROST_PARAM_JS_PRESENT, + DRM_PANFROST_PARAM_L2_FEATURES, + DRM_PANFROST_PARAM_CORE_FEATURES, + DRM_PANFROST_PARAM_TILER_FEATURES, + DRM_PANFROST_PARAM_MEM_FEATURES, + DRM_PANFROST_PARAM_MMU_FEATURES, + DRM_PANFROST_PARAM_THREAD_FEATURES, + DRM_PANFROST_PARAM_MAX_THREADS, + DRM_PANFROST_PARAM_THREAD_MAX_WORKGROUP_SZ, + DRM_PANFROST_PARAM_THREAD_MAX_BARRIER_SZ, + DRM_PANFROST_PARAM_COHERENCY_FEATURES, + DRM_PANFROST_PARAM_TEXTURE_FEATURES0, + DRM_PANFROST_PARAM_TEXTURE_FEATURES1, + DRM_PANFROST_PARAM_TEXTURE_FEATURES2, + DRM_PANFROST_PARAM_TEXTURE_FEATURES3, + DRM_PANFROST_PARAM_JS_FEATURES0, + DRM_PANFROST_PARAM_JS_FEATURES1, + DRM_PANFROST_PARAM_JS_FEATURES2, + DRM_PANFROST_PARAM_JS_FEATURES3, + DRM_PANFROST_PARAM_JS_FEATURES4, + DRM_PANFROST_PARAM_JS_FEATURES5, + DRM_PANFROST_PARAM_JS_FEATURES6, + DRM_PANFROST_PARAM_JS_FEATURES7, + DRM_PANFROST_PARAM_JS_FEATURES8, + DRM_PANFROST_PARAM_JS_FEATURES9, + DRM_PANFROST_PARAM_JS_FEATURES10, + DRM_PANFROST_PARAM_JS_FEATURES11, + DRM_PANFROST_PARAM_JS_FEATURES12, + DRM_PANFROST_PARAM_JS_FEATURES13, + DRM_PANFROST_PARAM_JS_FEATURES14, + DRM_PANFROST_PARAM_JS_FEATURES15, + DRM_PANFROST_PARAM_NR_CORE_GROUPS, + DRM_PANFROST_PARAM_THREAD_TLS_ALLOC, }; struct drm_panfrost_get_param {
Midgard/Bifrost GPUs have a bunch of feature registers providing details of what the hardware supports. Panfrost already reads these, this patch exports them all to user space so that the jobs created by the user space driver can be tuned for the particular hardware implementation. Signed-off-by: Steven Price <steven.price@arm.com> --- drivers/gpu/drm/panfrost/panfrost_device.h | 1 + drivers/gpu/drm/panfrost/panfrost_drv.c | 38 +++++++++++++++++++-- drivers/gpu/drm/panfrost/panfrost_gpu.c | 2 ++ include/uapi/drm/panfrost_drm.h | 39 ++++++++++++++++++++++ 4 files changed, 77 insertions(+), 3 deletions(-)