From 7154d76fedf549607afbc0d13db9aaf02da5cebf Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Thu, 3 Nov 2016 18:53:10 -0700 Subject: drm/vc4: Add support for rendering with ETC1 textures. The validation for it ends up being quite simple, but I hadn't got around to it before merging the driver. For backwards compatibility, we also need to add a flag so that the userspace GL driver can easily tell if the kernel will allow ETC1 textures (on an old kernel, it will continue to convert to RGBA8) Signed-off-by: Eric Anholt --- include/uapi/drm/vc4_drm.h | 1 + 1 file changed, 1 insertion(+) (limited to 'include/uapi/drm') diff --git a/include/uapi/drm/vc4_drm.h b/include/uapi/drm/vc4_drm.h index ad7edc3edf7c..69caa21f0cb2 100644 --- a/include/uapi/drm/vc4_drm.h +++ b/include/uapi/drm/vc4_drm.h @@ -286,6 +286,7 @@ struct drm_vc4_get_hang_state { #define DRM_VC4_PARAM_V3D_IDENT1 1 #define DRM_VC4_PARAM_V3D_IDENT2 2 #define DRM_VC4_PARAM_SUPPORTS_BRANCHES 3 +#define DRM_VC4_PARAM_SUPPORTS_ETC1 4 struct drm_vc4_get_param { __u32 param; -- cgit v1.3 From 91d5ee0451607d2a33c2b811ef3cc4a3e4578ed8 Mon Sep 17 00:00:00 2001 From: Ville Syrjälä Date: Thu, 3 Nov 2016 16:10:01 +0200 Subject: drm/uapi: Add a warning that mode flags must match the xrandr definitions MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Existing userspace expected the mode flags to match the xrandr definitions 1:1, and even adding new flags in he previously unused bits is likely to break existing userspace. Add a comment warning people about this potential trap. Cc: Shashank Sharma Cc: "Lin, Jia" Cc: Akashdeep Sharma Cc: Jim Bride Cc: Jose Abreu Cc: Daniel Vetter Cc: Emil Velikov Signed-off-by: Ville Syrjälä Signed-off-by: Daniel Vetter Link: http://patchwork.freedesktop.org/patch/msgid/1478182201-26086-1-git-send-email-ville.syrjala@linux.intel.com --- include/uapi/drm/drm_mode.h | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) (limited to 'include/uapi/drm') diff --git a/include/uapi/drm/drm_mode.h b/include/uapi/drm/drm_mode.h index 084b50a02dc5..01000c9f7c2c 100644 --- a/include/uapi/drm/drm_mode.h +++ b/include/uapi/drm/drm_mode.h @@ -47,7 +47,15 @@ extern "C" { #define DRM_MODE_TYPE_DRIVER (1<<6) /* Video mode flags */ -/* bit compatible with the xorg definitions. */ +/* bit compatible with the xrandr RR_ definitions (bits 0-13) + * + * ABI warning: Existing userspace really expects + * the mode flags to match the xrandr definitions. Any + * changes that don't match the xrandr definitions will + * likely need a new client cap or some other mechanism + * to avoid breaking existing userspace. This includes + * allocating new flags in the previously unused bits! + */ #define DRM_MODE_FLAG_PHSYNC (1<<0) #define DRM_MODE_FLAG_NHSYNC (1<<1) #define DRM_MODE_FLAG_PVSYNC (1<<2) -- cgit v1.3 From eef18a827a9ec58aa9fc1ccfb7e65ff04ebc25f3 Mon Sep 17 00:00:00 2001 From: Junwei Zhang Date: Fri, 4 Nov 2016 16:16:10 -0400 Subject: drm/amdgpu: add the interface of waiting multiple fences (v4) v2: agd: rebase and squash in all the previous optimizations and changes so everything compiles. v3: squash in Slava's 32bit build fix v4: rebase on drm-next (fence -> dma_fence), squash in Monk's ioctl update patch Signed-off-by: Junwei Zhang Reviewed-by: Monk Liu Reviewed-by: Jammy Zhou Signed-off-by: Alex Deucher Signed-off-by: Sumit Semwal [sumits: fix checkpatch warnings] Link: http://patchwork.freedesktop.org/patch/msgid/1478290570-30982-2-git-send-email-alexander.deucher@amd.com --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 + drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 174 ++++++++++++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c | 3 +- include/uapi/drm/amdgpu_drm.h | 28 +++++ 5 files changed, 207 insertions(+), 1 deletion(-) (limited to 'include/uapi/drm') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 2ec7b3baeec2..c2b8496cdf63 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1212,6 +1212,8 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); +int amdgpu_cs_wait_fences_ioctl(struct drm_device *dev, void *data, + struct drm_file *filp); int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index a024217896fd..b4bc83aa5999 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -1139,6 +1139,180 @@ int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data, return 0; } +/** + * amdgpu_cs_get_fence - helper to get fence from drm_amdgpu_fence + * + * @adev: amdgpu device + * @filp: file private + * @user: drm_amdgpu_fence copied from user space + */ +static struct dma_fence *amdgpu_cs_get_fence(struct amdgpu_device *adev, + struct drm_file *filp, + struct drm_amdgpu_fence *user) +{ + struct amdgpu_ring *ring; + struct amdgpu_ctx *ctx; + struct dma_fence *fence; + int r; + + r = amdgpu_cs_get_ring(adev, user->ip_type, user->ip_instance, + user->ring, &ring); + if (r) + return ERR_PTR(r); + + ctx = amdgpu_ctx_get(filp->driver_priv, user->ctx_id); + if (ctx == NULL) + return ERR_PTR(-EINVAL); + + fence = amdgpu_ctx_get_fence(ctx, ring, user->seq_no); + amdgpu_ctx_put(ctx); + + return fence; +} + +/** + * amdgpu_cs_wait_all_fence - wait on all fences to signal + * + * @adev: amdgpu device + * @filp: file private + * @wait: wait parameters + * @fences: array of drm_amdgpu_fence + */ +static int amdgpu_cs_wait_all_fences(struct amdgpu_device *adev, + struct drm_file *filp, + union drm_amdgpu_wait_fences *wait, + struct drm_amdgpu_fence *fences) +{ + uint32_t fence_count = wait->in.fence_count; + unsigned int i; + long r = 1; + + for (i = 0; i < fence_count; i++) { + struct dma_fence *fence; + unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout_ns); + + fence = amdgpu_cs_get_fence(adev, filp, &fences[i]); + if (IS_ERR(fence)) + return PTR_ERR(fence); + else if (!fence) + continue; + + r = dma_fence_wait_timeout(fence, true, timeout); + if (r < 0) + return r; + + if (r == 0) + break; + } + + memset(wait, 0, sizeof(*wait)); + wait->out.status = (r > 0); + + return 0; +} + +/** + * amdgpu_cs_wait_any_fence - wait on any fence to signal + * + * @adev: amdgpu device + * @filp: file private + * @wait: wait parameters + * @fences: array of drm_amdgpu_fence + */ +static int amdgpu_cs_wait_any_fence(struct amdgpu_device *adev, + struct drm_file *filp, + union drm_amdgpu_wait_fences *wait, + struct drm_amdgpu_fence *fences) +{ + unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout_ns); + uint32_t fence_count = wait->in.fence_count; + uint32_t first = ~0; + struct dma_fence **array; + unsigned int i; + long r; + + /* Prepare the fence array */ + array = kcalloc(fence_count, sizeof(struct dma_fence *), GFP_KERNEL); + + if (array == NULL) + return -ENOMEM; + + for (i = 0; i < fence_count; i++) { + struct dma_fence *fence; + + fence = amdgpu_cs_get_fence(adev, filp, &fences[i]); + if (IS_ERR(fence)) { + r = PTR_ERR(fence); + goto err_free_fence_array; + } else if (fence) { + array[i] = fence; + } else { /* NULL, the fence has been already signaled */ + r = 1; + goto out; + } + } + + r = dma_fence_wait_any_timeout(array, fence_count, true, timeout, + &first); + if (r < 0) + goto err_free_fence_array; + +out: + memset(wait, 0, sizeof(*wait)); + wait->out.status = (r > 0); + wait->out.first_signaled = first; + /* set return value 0 to indicate success */ + r = 0; + +err_free_fence_array: + for (i = 0; i < fence_count; i++) + dma_fence_put(array[i]); + kfree(array); + + return r; +} + +/** + * amdgpu_cs_wait_fences_ioctl - wait for multiple command submissions to finish + * + * @dev: drm device + * @data: data from userspace + * @filp: file private + */ +int amdgpu_cs_wait_fences_ioctl(struct drm_device *dev, void *data, + struct drm_file *filp) +{ + struct amdgpu_device *adev = dev->dev_private; + union drm_amdgpu_wait_fences *wait = data; + uint32_t fence_count = wait->in.fence_count; + struct drm_amdgpu_fence *fences_user; + struct drm_amdgpu_fence *fences; + int r; + + /* Get the fences from userspace */ + fences = kmalloc_array(fence_count, sizeof(struct drm_amdgpu_fence), + GFP_KERNEL); + if (fences == NULL) + return -ENOMEM; + + fences_user = (void __user *)(unsigned long)(wait->in.fences); + if (copy_from_user(fences, fences_user, + sizeof(struct drm_amdgpu_fence) * fence_count)) { + r = -EFAULT; + goto err_free_fences; + } + + if (wait->in.wait_all) + r = amdgpu_cs_wait_all_fences(adev, filp, wait, fences); + else + r = amdgpu_cs_wait_any_fence(adev, filp, wait, fences); + +err_free_fences: + kfree(fences); + + return r; +} + /** * amdgpu_cs_find_bo_va - find bo_va for VM address * diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c index 78392671046a..06e145b5afd7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c @@ -825,6 +825,7 @@ const struct drm_ioctl_desc amdgpu_ioctls_kms[] = { DRM_IOCTL_DEF_DRV(AMDGPU_CS, amdgpu_cs_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(AMDGPU_INFO, amdgpu_info_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(AMDGPU_WAIT_CS, amdgpu_cs_wait_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_WAIT_FENCES, amdgpu_cs_wait_fences_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(AMDGPU_GEM_METADATA, amdgpu_gem_metadata_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c index fd26c4b8d793..34a795463988 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c @@ -361,7 +361,8 @@ int amdgpu_sa_bo_new(struct amdgpu_sa_manager *sa_manager, if (count) { spin_unlock(&sa_manager->wq.lock); t = dma_fence_wait_any_timeout(fences, count, false, - MAX_SCHEDULE_TIMEOUT); + MAX_SCHEDULE_TIMEOUT, + NULL); for (i = 0; i < count; ++i) dma_fence_put(fences[i]); diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h index 4684f378f046..2191a9e4f3db 100644 --- a/include/uapi/drm/amdgpu_drm.h +++ b/include/uapi/drm/amdgpu_drm.h @@ -50,6 +50,7 @@ extern "C" { #define DRM_AMDGPU_WAIT_CS 0x09 #define DRM_AMDGPU_GEM_OP 0x10 #define DRM_AMDGPU_GEM_USERPTR 0x11 +#define DRM_AMDGPU_WAIT_FENCES 0x12 #define DRM_IOCTL_AMDGPU_GEM_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_CREATE, union drm_amdgpu_gem_create) #define DRM_IOCTL_AMDGPU_GEM_MMAP DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_MMAP, union drm_amdgpu_gem_mmap) @@ -63,6 +64,7 @@ extern "C" { #define DRM_IOCTL_AMDGPU_WAIT_CS DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_WAIT_CS, union drm_amdgpu_wait_cs) #define DRM_IOCTL_AMDGPU_GEM_OP DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_OP, struct drm_amdgpu_gem_op) #define DRM_IOCTL_AMDGPU_GEM_USERPTR DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_USERPTR, struct drm_amdgpu_gem_userptr) +#define DRM_IOCTL_AMDGPU_WAIT_FENCES DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_WAIT_FENCES, union drm_amdgpu_wait_fences) #define AMDGPU_GEM_DOMAIN_CPU 0x1 #define AMDGPU_GEM_DOMAIN_GTT 0x2 @@ -307,6 +309,32 @@ union drm_amdgpu_wait_cs { struct drm_amdgpu_wait_cs_out out; }; +struct drm_amdgpu_fence { + __u32 ctx_id; + __u32 ip_type; + __u32 ip_instance; + __u32 ring; + __u64 seq_no; +}; + +struct drm_amdgpu_wait_fences_in { + /** This points to uint64_t * which points to fences */ + __u64 fences; + __u32 fence_count; + __u32 wait_all; + __u64 timeout_ns; +}; + +struct drm_amdgpu_wait_fences_out { + __u32 status; + __u32 first_signaled; +}; + +union drm_amdgpu_wait_fences { + struct drm_amdgpu_wait_fences_in in; + struct drm_amdgpu_wait_fences_out out; +}; + #define AMDGPU_GEM_OP_GET_GEM_CREATE_INFO 0 #define AMDGPU_GEM_OP_SET_PLACEMENT 1 -- cgit v1.3 From 7920232d54d97a35dbafc65cc6f6ffeefe9ecbbe Mon Sep 17 00:00:00 2001 From: Ville Syrjälä Date: Thu, 3 Nov 2016 14:31:43 +0200 Subject: Revert "drm: Add and handle new aspect ratios in DRM layer" MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This reverts commit a68362fe3e84fcbedd49939aa200519aa5410135. Adding new mode flags willy nilly breaks existing userspace. We need to coordinate this better, potentially with a new client cap that only exposes the aspect ratio flags when userspace is prepared for them (similar to what we do with stereo 3D modes). Cc: Shashank Sharma Cc: Lin, Jia Cc: Akashdeep Sharma Cc: Jim Bride Cc: Jose Abreu Cc: Daniel Vetter Cc: Emil Velikov Cc: Daniel Vetter Signed-off-by: Ville Syrjälä Signed-off-by: Daniel Vetter Link: http://patchwork.freedesktop.org/patch/msgid/1478176304-6743-1-git-send-email-ville.syrjala@linux.intel.com --- drivers/gpu/drm/drm_modes.c | 12 ------------ include/uapi/drm/drm_mode.h | 6 ------ 2 files changed, 18 deletions(-) (limited to 'include/uapi/drm') diff --git a/drivers/gpu/drm/drm_modes.c b/drivers/gpu/drm/drm_modes.c index ce6eeda02acf..e82298e4c25b 100644 --- a/drivers/gpu/drm/drm_modes.c +++ b/drivers/gpu/drm/drm_modes.c @@ -1507,12 +1507,6 @@ void drm_mode_convert_to_umode(struct drm_mode_modeinfo *out, case HDMI_PICTURE_ASPECT_16_9: out->flags |= DRM_MODE_FLAG_PIC_AR_16_9; break; - case HDMI_PICTURE_ASPECT_64_27: - out->flags |= DRM_MODE_FLAG_PIC_AR_64_27; - break; - case DRM_MODE_PICTURE_ASPECT_256_135: - out->flags |= DRM_MODE_FLAG_PIC_AR_256_135; - break; case HDMI_PICTURE_ASPECT_RESERVED: default: out->flags |= DRM_MODE_FLAG_PIC_AR_NONE; @@ -1574,12 +1568,6 @@ int drm_mode_convert_umode(struct drm_display_mode *out, case DRM_MODE_FLAG_PIC_AR_16_9: out->picture_aspect_ratio |= HDMI_PICTURE_ASPECT_16_9; break; - case DRM_MODE_FLAG_PIC_AR_64_27: - out->picture_aspect_ratio |= HDMI_PICTURE_ASPECT_64_27; - break; - case DRM_MODE_FLAG_PIC_AR_256_135: - out->picture_aspect_ratio |= HDMI_PICTURE_ASPECT_256_135; - break; default: out->picture_aspect_ratio = HDMI_PICTURE_ASPECT_NONE; break; diff --git a/include/uapi/drm/drm_mode.h b/include/uapi/drm/drm_mode.h index 01000c9f7c2c..ebf622f5b238 100644 --- a/include/uapi/drm/drm_mode.h +++ b/include/uapi/drm/drm_mode.h @@ -89,8 +89,6 @@ extern "C" { #define DRM_MODE_PICTURE_ASPECT_NONE 0 #define DRM_MODE_PICTURE_ASPECT_4_3 1 #define DRM_MODE_PICTURE_ASPECT_16_9 2 -#define DRM_MODE_PICTURE_ASPECT_64_27 3 -#define DRM_MODE_PICTURE_ASPECT_256_135 4 /* Aspect ratio flag bitmask (4 bits 22:19) */ #define DRM_MODE_FLAG_PIC_AR_MASK (0x0F<<19) @@ -100,10 +98,6 @@ extern "C" { (DRM_MODE_PICTURE_ASPECT_4_3<<19) #define DRM_MODE_FLAG_PIC_AR_16_9 \ (DRM_MODE_PICTURE_ASPECT_16_9<<19) -#define DRM_MODE_FLAG_PIC_AR_64_27 \ - (DRM_MODE_PICTURE_ASPECT_64_27<<19) -#define DRM_MODE_FLAG_PIC_AR_256_135 \ - (DRM_MODE_PICTURE_ASPECT_256_135<<19) /* DPMS flags */ /* bit compatible with the xorg definitions. */ -- cgit v1.3 From c778cc5df944291dcdb1ca7a6bb781fbc22550c5 Mon Sep 17 00:00:00 2001 From: Jonas Pfeil Date: Tue, 8 Nov 2016 00:18:39 +0100 Subject: drm/vc4: Add fragment shader threading support FS threading brings performance improvements of 0-20% in glmark2. The validation code checks for thread switch signals and ensures that the registers of the other thread are not touched, and that our clamps are not live across thread switches. It also checks that the threading and branching instructions do not interfere. (Original patch by Jonas, changes by anholt for style cleanup, removing validation the kernel doesn't need to do, and adding the flag for userspace). v2: Minor style fixes from checkpatch. Signed-off-by: Jonas Pfeil Signed-off-by: Eric Anholt --- drivers/gpu/drm/vc4/vc4_drv.c | 1 + drivers/gpu/drm/vc4/vc4_drv.h | 2 + drivers/gpu/drm/vc4/vc4_validate.c | 17 +++++--- drivers/gpu/drm/vc4/vc4_validate_shaders.c | 63 ++++++++++++++++++++++++++++++ include/uapi/drm/vc4_drm.h | 1 + 5 files changed, 79 insertions(+), 5 deletions(-) (limited to 'include/uapi/drm') diff --git a/drivers/gpu/drm/vc4/vc4_drv.c b/drivers/gpu/drm/vc4/vc4_drv.c index 7abfe088f2d1..86aabf6d0f79 100644 --- a/drivers/gpu/drm/vc4/vc4_drv.c +++ b/drivers/gpu/drm/vc4/vc4_drv.c @@ -82,6 +82,7 @@ static int vc4_get_param_ioctl(struct drm_device *dev, void *data, break; case DRM_VC4_PARAM_SUPPORTS_BRANCHES: case DRM_VC4_PARAM_SUPPORTS_ETC1: + case DRM_VC4_PARAM_SUPPORTS_THREADED_FS: args->value = true; break; default: diff --git a/drivers/gpu/drm/vc4/vc4_drv.h b/drivers/gpu/drm/vc4/vc4_drv.h index 7c1e4d97486f..fef172804345 100644 --- a/drivers/gpu/drm/vc4/vc4_drv.h +++ b/drivers/gpu/drm/vc4/vc4_drv.h @@ -381,6 +381,8 @@ struct vc4_validated_shader_info { uint32_t num_uniform_addr_offsets; uint32_t *uniform_addr_offsets; + + bool is_threaded; }; /** diff --git a/drivers/gpu/drm/vc4/vc4_validate.c b/drivers/gpu/drm/vc4/vc4_validate.c index e18f88203d32..9fd171c361c2 100644 --- a/drivers/gpu/drm/vc4/vc4_validate.c +++ b/drivers/gpu/drm/vc4/vc4_validate.c @@ -789,11 +789,6 @@ validate_gl_shader_rec(struct drm_device *dev, exec->shader_rec_v += roundup(packet_size, 16); exec->shader_rec_size -= packet_size; - if (!(*(uint16_t *)pkt_u & VC4_SHADER_FLAG_FS_SINGLE_THREAD)) { - DRM_ERROR("Multi-threaded fragment shaders not supported.\n"); - return -EINVAL; - } - for (i = 0; i < shader_reloc_count; i++) { if (src_handles[i] > exec->bo_count) { DRM_ERROR("Shader handle %d too big\n", src_handles[i]); @@ -810,6 +805,18 @@ validate_gl_shader_rec(struct drm_device *dev, return -EINVAL; } + if (((*(uint16_t *)pkt_u & VC4_SHADER_FLAG_FS_SINGLE_THREAD) == 0) != + to_vc4_bo(&bo[0]->base)->validated_shader->is_threaded) { + DRM_ERROR("Thread mode of CL and FS do not match\n"); + return -EINVAL; + } + + if (to_vc4_bo(&bo[1]->base)->validated_shader->is_threaded || + to_vc4_bo(&bo[2]->base)->validated_shader->is_threaded) { + DRM_ERROR("cs and vs cannot be threaded\n"); + return -EINVAL; + } + for (i = 0; i < shader_reloc_count; i++) { struct vc4_validated_shader_info *validated_shader; uint32_t o = shader_reloc_offsets[i]; diff --git a/drivers/gpu/drm/vc4/vc4_validate_shaders.c b/drivers/gpu/drm/vc4/vc4_validate_shaders.c index 917321ce832f..5dba13dd1e9b 100644 --- a/drivers/gpu/drm/vc4/vc4_validate_shaders.c +++ b/drivers/gpu/drm/vc4/vc4_validate_shaders.c @@ -83,6 +83,13 @@ struct vc4_shader_validation_state { * basic blocks. */ bool needs_uniform_address_for_loop; + + /* Set when we find an instruction writing the top half of the + * register files. If we allowed writing the unusable regs in + * a threaded shader, then the other shader running on our + * QPU's clamp validation would be invalid. + */ + bool all_registers_used; }; static uint32_t @@ -118,6 +125,13 @@ raddr_add_a_to_live_reg_index(uint64_t inst) return ~0; } +static bool +live_reg_is_upper_half(uint32_t lri) +{ + return (lri >= 16 && lri < 32) || + (lri >= 32 + 16 && lri < 32 + 32); +} + static bool is_tmu_submit(uint32_t waddr) { @@ -390,6 +404,9 @@ check_reg_write(struct vc4_validated_shader_info *validated_shader, } else { validation_state->live_immediates[lri] = ~0; } + + if (live_reg_is_upper_half(lri)) + validation_state->all_registers_used = true; } switch (waddr) { @@ -598,6 +615,11 @@ check_instruction_reads(struct vc4_validated_shader_info *validated_shader, } } + if ((raddr_a >= 16 && raddr_a < 32) || + (raddr_b >= 16 && raddr_b < 32 && sig != QPU_SIG_SMALL_IMM)) { + validation_state->all_registers_used = true; + } + return true; } @@ -753,6 +775,7 @@ vc4_validate_shader(struct drm_gem_cma_object *shader_obj) { bool found_shader_end = false; int shader_end_ip = 0; + uint32_t last_thread_switch_ip = -3; uint32_t ip; struct vc4_validated_shader_info *validated_shader = NULL; struct vc4_shader_validation_state validation_state; @@ -785,6 +808,17 @@ vc4_validate_shader(struct drm_gem_cma_object *shader_obj) if (!vc4_handle_branch_target(&validation_state)) goto fail; + if (ip == last_thread_switch_ip + 3) { + /* Reset r0-r3 live clamp data */ + int i; + + for (i = 64; i < LIVE_REG_COUNT; i++) { + validation_state.live_min_clamp_offsets[i] = ~0; + validation_state.live_max_clamp_regs[i] = false; + validation_state.live_immediates[i] = ~0; + } + } + switch (sig) { case QPU_SIG_NONE: case QPU_SIG_WAIT_FOR_SCOREBOARD: @@ -794,6 +828,8 @@ vc4_validate_shader(struct drm_gem_cma_object *shader_obj) case QPU_SIG_LOAD_TMU1: case QPU_SIG_PROG_END: case QPU_SIG_SMALL_IMM: + case QPU_SIG_THREAD_SWITCH: + case QPU_SIG_LAST_THREAD_SWITCH: if (!check_instruction_writes(validated_shader, &validation_state)) { DRM_ERROR("Bad write at ip %d\n", ip); @@ -809,6 +845,18 @@ vc4_validate_shader(struct drm_gem_cma_object *shader_obj) shader_end_ip = ip; } + if (sig == QPU_SIG_THREAD_SWITCH || + sig == QPU_SIG_LAST_THREAD_SWITCH) { + validated_shader->is_threaded = true; + + if (ip < last_thread_switch_ip + 3) { + DRM_ERROR("Thread switch too soon after " + "last switch at ip %d\n", ip); + goto fail; + } + last_thread_switch_ip = ip; + } + break; case QPU_SIG_LOAD_IMM: @@ -823,6 +871,13 @@ vc4_validate_shader(struct drm_gem_cma_object *shader_obj) if (!check_branch(inst, validated_shader, &validation_state, ip)) goto fail; + + if (ip < last_thread_switch_ip + 3) { + DRM_ERROR("Branch in thread switch at ip %d", + ip); + goto fail; + } + break; default: DRM_ERROR("Unsupported QPU signal %d at " @@ -844,6 +899,14 @@ vc4_validate_shader(struct drm_gem_cma_object *shader_obj) goto fail; } + /* Might corrupt other thread */ + if (validated_shader->is_threaded && + validation_state.all_registers_used) { + DRM_ERROR("Shader uses threading, but uses the upper " + "half of the registers, too\n"); + goto fail; + } + /* If we did a backwards branch and we haven't emitted a uniforms * reset since then, we still need the uniforms stream to have the * uniforms address available so that the backwards branch can do its diff --git a/include/uapi/drm/vc4_drm.h b/include/uapi/drm/vc4_drm.h index 69caa21f0cb2..f07a09016726 100644 --- a/include/uapi/drm/vc4_drm.h +++ b/include/uapi/drm/vc4_drm.h @@ -287,6 +287,7 @@ struct drm_vc4_get_hang_state { #define DRM_VC4_PARAM_V3D_IDENT2 2 #define DRM_VC4_PARAM_SUPPORTS_BRANCHES 3 #define DRM_VC4_PARAM_SUPPORTS_ETC1 4 +#define DRM_VC4_PARAM_SUPPORTS_THREADED_FS 5 struct drm_vc4_get_param { __u32 param; -- cgit v1.3