From 98830d91da082b0285d35bdf5b5ae98decac7df6 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Wed, 7 Jun 2017 17:13:35 -0700 Subject: drm/vc4: Add T-format scanout support. The T tiling format is what V3D uses for textures, with no raster support at all until later revisions of the hardware (and always at a large 3D performance penalty). If we can't scan out V3D's format, then we often need to do a relayout at some stage of the pipeline, either right before texturing from the scanout buffer (common in X11 without a compositor) or between a tiled screen buffer right before scanout (an option I've considered in trying to resolve this inconsistency, but which means needing to use the dirty fb ioctl and having some update policy). T-format scanout lets us avoid either of those shadow copies, for a massive, obvious performance improvement to X11 window dragging without a compositor. Unfortunately, enabling a compositor to work around the discrepancy has turned out to be too costly in memory consumption for the Raspbian distribution. Because the HVS operates a scanline at a time, compositing from T does increase the memory bandwidth cost of scanout. On my 1920x1080@32bpp display on a RPi3, we go from about 15% of system memory bandwidth with linear to about 20% with tiled. However, for X11 this still ends up being a huge performance win in active usage. This patch doesn't yet handle src_x/src_y offsetting within the tiled buffer. However, we fail to do so for untiled buffers already. Signed-off-by: Eric Anholt Link: http://patchwork.freedesktop.org/patch/msgid/20170608001336.12842-1-eric@anholt.net Reviewed-by: Boris Brezillon --- drivers/gpu/drm/vc4/vc4_plane.c | 31 +++++++++++++++++++++++++++---- drivers/gpu/drm/vc4/vc4_regs.h | 19 +++++++++++++++++++ 2 files changed, 46 insertions(+), 4 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/vc4/vc4_plane.c b/drivers/gpu/drm/vc4/vc4_plane.c index da18dec21696..fa6809d8b0fe 100644 --- a/drivers/gpu/drm/vc4/vc4_plane.c +++ b/drivers/gpu/drm/vc4/vc4_plane.c @@ -500,8 +500,8 @@ static int vc4_plane_mode_set(struct drm_plane *plane, u32 ctl0_offset = vc4_state->dlist_count; const struct hvs_format *format = vc4_get_hvs_format(fb->format->format); int num_planes = drm_format_num_planes(format->drm); - u32 scl0, scl1; - u32 lbm_size; + u32 scl0, scl1, pitch0; + u32 lbm_size, tiling; unsigned long irqflags; int ret, i; @@ -542,11 +542,31 @@ static int vc4_plane_mode_set(struct drm_plane *plane, scl1 = vc4_get_scl_field(state, 0); } + switch (fb->modifier) { + case DRM_FORMAT_MOD_LINEAR: + tiling = SCALER_CTL0_TILING_LINEAR; + pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH); + break; + case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: + tiling = SCALER_CTL0_TILING_256B_OR_T; + + pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET), + VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L), + VC4_SET_FIELD((vc4_state->src_w[0] + 31) >> 5, + SCALER_PITCH0_TILE_WIDTH_R)); + break; + default: + DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx", + (long long)fb->modifier); + return -EINVAL; + } + /* Control word */ vc4_dlist_write(vc4_state, SCALER_CTL0_VALID | (format->pixel_order << SCALER_CTL0_ORDER_SHIFT) | (format->hvs << SCALER_CTL0_PIXEL_FORMAT_SHIFT) | + VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) | (vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) | VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) | VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1)); @@ -600,8 +620,11 @@ static int vc4_plane_mode_set(struct drm_plane *plane, for (i = 0; i < num_planes; i++) vc4_dlist_write(vc4_state, 0xc0c0c0c0); - /* Pitch word 0/1/2 */ - for (i = 0; i < num_planes; i++) { + /* Pitch word 0 */ + vc4_dlist_write(vc4_state, pitch0); + + /* Pitch word 1/2 */ + for (i = 1; i < num_planes; i++) { vc4_dlist_write(vc4_state, VC4_SET_FIELD(fb->pitches[i], SCALER_SRC_PITCH)); } diff --git a/drivers/gpu/drm/vc4/vc4_regs.h b/drivers/gpu/drm/vc4/vc4_regs.h index 932093936178..d382c34c1b9e 100644 --- a/drivers/gpu/drm/vc4/vc4_regs.h +++ b/drivers/gpu/drm/vc4/vc4_regs.h @@ -709,6 +709,13 @@ enum hvs_pixel_format { #define SCALER_CTL0_SIZE_MASK VC4_MASK(29, 24) #define SCALER_CTL0_SIZE_SHIFT 24 +#define SCALER_CTL0_TILING_MASK VC4_MASK(21, 20) +#define SCALER_CTL0_TILING_SHIFT 20 +#define SCALER_CTL0_TILING_LINEAR 0 +#define SCALER_CTL0_TILING_64B 1 +#define SCALER_CTL0_TILING_128B 2 +#define SCALER_CTL0_TILING_256B_OR_T 3 + #define SCALER_CTL0_HFLIP BIT(16) #define SCALER_CTL0_VFLIP BIT(15) @@ -838,7 +845,19 @@ enum hvs_pixel_format { #define SCALER_PPF_KERNEL_OFFSET_SHIFT 0 #define SCALER_PPF_KERNEL_UNCACHED BIT(31) +/* PITCH0/1/2 fields for raster. */ #define SCALER_SRC_PITCH_MASK VC4_MASK(15, 0) #define SCALER_SRC_PITCH_SHIFT 0 +/* PITCH0 fields for T-tiled. */ +#define SCALER_PITCH0_TILE_WIDTH_L_MASK VC4_MASK(22, 16) +#define SCALER_PITCH0_TILE_WIDTH_L_SHIFT 16 +#define SCALER_PITCH0_TILE_LINE_DIR BIT(15) +#define SCALER_PITCH0_TILE_INITIAL_LINE_DIR BIT(14) +/* Y offset within a tile. */ +#define SCALER_PITCH0_TILE_Y_OFFSET_MASK VC4_MASK(13, 7) +#define SCALER_PITCH0_TILE_Y_OFFSET_SHIFT 7 +#define SCALER_PITCH0_TILE_WIDTH_R_MASK VC4_MASK(6, 0) +#define SCALER_PITCH0_TILE_WIDTH_R_SHIFT 0 + #endif /* VC4_REGS_H */ -- cgit v1.2.3 From 83753117f1de4f6ef7588fac9545065eed1e85e2 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Wed, 7 Jun 2017 17:13:36 -0700 Subject: drm/vc4: Add get/set tiling ioctls. This allows mesa to set the tiling format for a BO and have that tiling format be respected by mesa on the other side of an import/export (and by vc4 scanout in the kernel), without defining a protocol to pass the tiling through userspace. Signed-off-by: Eric Anholt Link: http://patchwork.freedesktop.org/patch/msgid/20170608001336.12842-2-eric@anholt.net Acked-by: Dave Airlie --- drivers/gpu/drm/vc4/vc4_bo.c | 83 +++++++++++++++++++++++++++++++++++++++++++ drivers/gpu/drm/vc4/vc4_drv.c | 2 ++ drivers/gpu/drm/vc4/vc4_drv.h | 6 ++++ drivers/gpu/drm/vc4/vc4_kms.c | 41 ++++++++++++++++++++- include/uapi/drm/vc4_drm.h | 16 +++++++++ 5 files changed, 147 insertions(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/vc4/vc4_bo.c b/drivers/gpu/drm/vc4/vc4_bo.c index 590c0912afc1..487f96412d35 100644 --- a/drivers/gpu/drm/vc4/vc4_bo.c +++ b/drivers/gpu/drm/vc4/vc4_bo.c @@ -343,6 +343,7 @@ void vc4_free_object(struct drm_gem_object *gem_bo) bo->validated_shader = NULL; } + bo->t_format = false; bo->free_time = jiffies; list_add(&bo->size_head, cache_list); list_add(&bo->unref_head, &vc4->bo_cache.time_list); @@ -568,6 +569,88 @@ vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data, return ret; } +/** + * vc4_set_tiling_ioctl() - Sets the tiling modifier for a BO. + * @dev: DRM device + * @data: ioctl argument + * @file_priv: DRM file for this fd + * + * The tiling state of the BO decides the default modifier of an fb if + * no specific modifier was set by userspace, and the return value of + * vc4_get_tiling_ioctl() (so that userspace can treat a BO it + * received from dmabuf as the same tiling format as the producer + * used). + */ +int vc4_set_tiling_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_vc4_set_tiling *args = data; + struct drm_gem_object *gem_obj; + struct vc4_bo *bo; + bool t_format; + + if (args->flags != 0) + return -EINVAL; + + switch (args->modifier) { + case DRM_FORMAT_MOD_NONE: + t_format = false; + break; + case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: + t_format = true; + break; + default: + return -EINVAL; + } + + gem_obj = drm_gem_object_lookup(file_priv, args->handle); + if (!gem_obj) { + DRM_ERROR("Failed to look up GEM BO %d\n", args->handle); + return -ENOENT; + } + bo = to_vc4_bo(gem_obj); + bo->t_format = t_format; + + drm_gem_object_unreference_unlocked(gem_obj); + + return 0; +} + +/** + * vc4_get_tiling_ioctl() - Gets the tiling modifier for a BO. + * @dev: DRM device + * @data: ioctl argument + * @file_priv: DRM file for this fd + * + * Returns the tiling modifier for a BO as set by vc4_set_tiling_ioctl(). + */ +int vc4_get_tiling_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_vc4_get_tiling *args = data; + struct drm_gem_object *gem_obj; + struct vc4_bo *bo; + + if (args->flags != 0 || args->modifier != 0) + return -EINVAL; + + gem_obj = drm_gem_object_lookup(file_priv, args->handle); + if (!gem_obj) { + DRM_ERROR("Failed to look up GEM BO %d\n", args->handle); + return -ENOENT; + } + bo = to_vc4_bo(gem_obj); + + if (bo->t_format) + args->modifier = DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED; + else + args->modifier = DRM_FORMAT_MOD_NONE; + + drm_gem_object_unreference_unlocked(gem_obj); + + return 0; +} + void vc4_bo_cache_init(struct drm_device *dev) { struct vc4_dev *vc4 = to_vc4_dev(dev); diff --git a/drivers/gpu/drm/vc4/vc4_drv.c b/drivers/gpu/drm/vc4/vc4_drv.c index 136bb4213dc0..c6b487c3d2b7 100644 --- a/drivers/gpu/drm/vc4/vc4_drv.c +++ b/drivers/gpu/drm/vc4/vc4_drv.c @@ -138,6 +138,8 @@ static const struct drm_ioctl_desc vc4_drm_ioctls[] = { DRM_IOCTL_DEF_DRV(VC4_GET_HANG_STATE, vc4_get_hang_state_ioctl, DRM_ROOT_ONLY), DRM_IOCTL_DEF_DRV(VC4_GET_PARAM, vc4_get_param_ioctl, DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(VC4_SET_TILING, vc4_set_tiling_ioctl, DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(VC4_GET_TILING, vc4_get_tiling_ioctl, DRM_RENDER_ALLOW), }; static struct drm_driver vc4_drm_driver = { diff --git a/drivers/gpu/drm/vc4/vc4_drv.h b/drivers/gpu/drm/vc4/vc4_drv.h index a5bf2e5e0b57..df22698d62ee 100644 --- a/drivers/gpu/drm/vc4/vc4_drv.h +++ b/drivers/gpu/drm/vc4/vc4_drv.h @@ -148,6 +148,8 @@ struct vc4_bo { */ uint64_t write_seqno; + bool t_format; + /* List entry for the BO's position in either * vc4_exec_info->unref_list or vc4_dev->bo_cache.time_list */ @@ -470,6 +472,10 @@ int vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); +int vc4_set_tiling_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int vc4_get_tiling_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); int vc4_get_hang_state_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); int vc4_mmap(struct file *filp, struct vm_area_struct *vma); diff --git a/drivers/gpu/drm/vc4/vc4_kms.c b/drivers/gpu/drm/vc4/vc4_kms.c index 928d191ef90f..202f7ebf5a7b 100644 --- a/drivers/gpu/drm/vc4/vc4_kms.c +++ b/drivers/gpu/drm/vc4/vc4_kms.c @@ -202,11 +202,50 @@ static int vc4_atomic_commit(struct drm_device *dev, return 0; } +static struct drm_framebuffer *vc4_fb_create(struct drm_device *dev, + struct drm_file *file_priv, + const struct drm_mode_fb_cmd2 *mode_cmd) +{ + struct drm_mode_fb_cmd2 mode_cmd_local; + + /* If the user didn't specify a modifier, use the + * vc4_set_tiling_ioctl() state for the BO. + */ + if (!(mode_cmd->flags & DRM_MODE_FB_MODIFIERS)) { + struct drm_gem_object *gem_obj; + struct vc4_bo *bo; + + gem_obj = drm_gem_object_lookup(file_priv, + mode_cmd->handles[0]); + if (!gem_obj) { + DRM_ERROR("Failed to look up GEM BO %d\n", + mode_cmd->handles[0]); + return ERR_PTR(-ENOENT); + } + bo = to_vc4_bo(gem_obj); + + mode_cmd_local = *mode_cmd; + + if (bo->t_format) { + mode_cmd_local.modifier[0] = + DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED; + } else { + mode_cmd_local.modifier[0] = DRM_FORMAT_MOD_NONE; + } + + drm_gem_object_unreference_unlocked(gem_obj); + + mode_cmd = &mode_cmd_local; + } + + return drm_fb_cma_create(dev, file_priv, mode_cmd); +} + static const struct drm_mode_config_funcs vc4_mode_funcs = { .output_poll_changed = vc4_output_poll_changed, .atomic_check = drm_atomic_helper_check, .atomic_commit = vc4_atomic_commit, - .fb_create = drm_fb_cma_create, + .fb_create = vc4_fb_create, }; int vc4_kms_load(struct drm_device *dev) diff --git a/include/uapi/drm/vc4_drm.h b/include/uapi/drm/vc4_drm.h index f07a09016726..6ac4c5c014cb 100644 --- a/include/uapi/drm/vc4_drm.h +++ b/include/uapi/drm/vc4_drm.h @@ -38,6 +38,8 @@ extern "C" { #define DRM_VC4_CREATE_SHADER_BO 0x05 #define DRM_VC4_GET_HANG_STATE 0x06 #define DRM_VC4_GET_PARAM 0x07 +#define DRM_VC4_SET_TILING 0x08 +#define DRM_VC4_GET_TILING 0x09 #define DRM_IOCTL_VC4_SUBMIT_CL DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SUBMIT_CL, struct drm_vc4_submit_cl) #define DRM_IOCTL_VC4_WAIT_SEQNO DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_SEQNO, struct drm_vc4_wait_seqno) @@ -47,6 +49,8 @@ extern "C" { #define DRM_IOCTL_VC4_CREATE_SHADER_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_SHADER_BO, struct drm_vc4_create_shader_bo) #define DRM_IOCTL_VC4_GET_HANG_STATE DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_HANG_STATE, struct drm_vc4_get_hang_state) #define DRM_IOCTL_VC4_GET_PARAM DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_PARAM, struct drm_vc4_get_param) +#define DRM_IOCTL_VC4_SET_TILING DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SET_TILING, struct drm_vc4_set_tiling) +#define DRM_IOCTL_VC4_GET_TILING DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_TILING, struct drm_vc4_get_tiling) struct drm_vc4_submit_rcl_surface { __u32 hindex; /* Handle index, or ~0 if not present. */ @@ -295,6 +299,18 @@ struct drm_vc4_get_param { __u64 value; }; +struct drm_vc4_get_tiling { + __u32 handle; + __u32 flags; + __u64 modifier; +}; + +struct drm_vc4_set_tiling { + __u32 handle; + __u32 flags; + __u64 modifier; +}; + #if defined(__cplusplus) } #endif -- cgit v1.2.3 From 34c8ea400ff6383b028f63df2453914163afc07c Mon Sep 17 00:00:00 2001 From: Boris Brezillon Date: Fri, 2 Jun 2017 10:32:08 +0200 Subject: drm/vc4: Mimic drm_atomic_helper_commit() behavior The VC4 KMS driver is implementing its own ->atomic_commit() but there are a few generic helpers we can use instead of open-coding the logic. Signed-off-by: Boris Brezillon Reviewed-by: Eric Anholt Link: http://patchwork.freedesktop.org/patch/msgid/1496392332-8722-4-git-send-email-boris.brezillon@free-electrons.com --- drivers/gpu/drm/vc4/vc4_kms.c | 38 ++++++++++++-------------------------- 1 file changed, 12 insertions(+), 26 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/vc4/vc4_kms.c b/drivers/gpu/drm/vc4/vc4_kms.c index 202f7ebf5a7b..bc6ecdc6f104 100644 --- a/drivers/gpu/drm/vc4/vc4_kms.c +++ b/drivers/gpu/drm/vc4/vc4_kms.c @@ -42,6 +42,10 @@ vc4_atomic_complete_commit(struct vc4_commit *c) struct drm_device *dev = state->dev; struct vc4_dev *vc4 = to_vc4_dev(dev); + drm_atomic_helper_wait_for_fences(dev, state, false); + + drm_atomic_helper_wait_for_dependencies(state); + drm_atomic_helper_commit_modeset_disables(dev, state); drm_atomic_helper_commit_planes(dev, state, 0); @@ -57,10 +61,14 @@ vc4_atomic_complete_commit(struct vc4_commit *c) */ state->legacy_cursor_update = false; + drm_atomic_helper_commit_hw_done(state); + drm_atomic_helper_wait_for_vblanks(dev, state); drm_atomic_helper_cleanup_planes(dev, state); + drm_atomic_helper_commit_cleanup_done(state); + drm_atomic_state_put(state); up(&vc4->async_modeset); @@ -117,32 +125,10 @@ static int vc4_atomic_commit(struct drm_device *dev, if (!c) return -ENOMEM; - /* Make sure that any outstanding modesets have finished. */ - if (nonblock) { - struct drm_crtc *crtc; - struct drm_crtc_state *crtc_state; - unsigned long flags; - bool busy = false; - - /* - * If there's an undispatched event to send then we're - * obviously still busy. If there isn't, then we can - * unconditionally wait for the semaphore because it - * shouldn't be contended (for long). - * - * This is to prevent a race where queuing a new flip - * from userspace immediately on receipt of an event - * beats our clean-up and returns EBUSY. - */ - spin_lock_irqsave(&dev->event_lock, flags); - for_each_crtc_in_state(state, crtc, crtc_state, i) - busy |= vc4_event_pending(crtc); - spin_unlock_irqrestore(&dev->event_lock, flags); - if (busy) { - kfree(c); - return -EBUSY; - } - } + ret = drm_atomic_helper_setup_commit(state, nonblock); + if (ret) + return ret; + ret = down_interruptible(&vc4->async_modeset); if (ret) { kfree(c); -- cgit v1.2.3