rpms
/
kernel
2
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Add vc4 T-Format support to improve performance on RPi

This commit is contained in:
Peter Robinson 2017-06-19 13:29:40 +01:00
parent 11323bafe4
commit eba478c4ed
2 changed files with 436 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
kernel.spec
View File

@ -551,9 +551,6 @@ Patch430: bcm2837-initial-support.patch
Patch431: arm-rk3288-tinker.patch
# http://www.spinics.net/lists/dri-devel/msg132235.html
Patch433: drm-vc4-Fix-OOPSes-from-trying-to-cache-a-partially-constructed-BO..patch
# bcm283x mmc for wifi http://www.spinics.net/lists/arm-kernel/msg567077.html
Patch434: bcm283x-mmc-bcm2835.patch
@ -575,6 +572,11 @@ Patch440: arm64-Add-option-of-13-for-FORCE_MAX_ZONEORDER.patch
Patch441: bcm2835-clk-audio-jitter-issues.patch
# http://www.spinics.net/lists/dri-devel/msg132235.html
Patch443: drm-vc4-Fix-OOPSes-from-trying-to-cache-a-partially-constructed-BO..patch
# https://lists.freedesktop.org/archives/dri-devel/2017-June/143688.html
Patch444: vc4-tformat-fixes.patch
Patch460: lib-cpumask-Make-CPUMASK_OFFSTACK-usable-without-deb.patch
Patch466: input-kill-stupid-messages.patch
@ -2230,6 +2232,9 @@ fi
#
#
%changelog
* Mon Jun 19 2017 Peter Robinson <pbrobinson@fedoraproject.org>
- Add vc4 T-Format support to improve performance
* Fri Jun 16 2017 Laura Abbott <labbott@fedoraproject.org>
- Fix an auditd race condition (rhbz 1459326)

428
vc4-tformat-fixes.patch Normal file
View File

@ -0,0 +1,428 @@
From ac4e55169649132123c4f2f39e0b02b5c849bae8 Mon Sep 17 00:00:00 2001
From: Peter Robinson <pbrobinson@gmail.com>
Date: Mon, 19 Jun 2017 13:20:48 +0100
Subject: [PATCH] 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.
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 <eric@anholt.net>
Reviewed-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Signed-off-by: Peter Robinson <pbrobinson@gmail.com>
---
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 +++++++++++++++++++-
drivers/gpu/drm/vc4/vc4_plane.c | 31 +++++++++++++--
drivers/gpu/drm/vc4/vc4_regs.h | 19 ++++++++++
include/uapi/drm/drm_fourcc.h | 22 +++++++++++
include/uapi/drm/vc4_drm.h | 16 ++++++++
8 files changed, 215 insertions(+), 5 deletions(-)
diff --git a/drivers/gpu/drm/vc4/vc4_bo.c b/drivers/gpu/drm/vc4/vc4_bo.c
index 3f6704cf6608..0918346c248e 100644
--- a/drivers/gpu/drm/vc4/vc4_bo.c
+++ b/drivers/gpu/drm/vc4/vc4_bo.c
@@ -325,6 +325,7 @@ 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);
@@ -525,6 +526,88 @@ 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 a459745e96f7..2edf2d4c5156 100644
--- a/drivers/gpu/drm/vc4/vc4_drv.c
+++ b/drivers/gpu/drm/vc4/vc4_drv.c
@@ -122,6 +122,8 @@ 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 0e59f3ee1b83..64f0cb1f889e 100644
--- a/drivers/gpu/drm/vc4/vc4_drv.h
+++ b/drivers/gpu/drm/vc4/vc4_drv.h
@@ -135,6 +135,8 @@ 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
*/
@@ -433,6 +435,10 @@ 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 ad7925a9e0ea..25be60016527 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/drivers/gpu/drm/vc4/vc4_plane.c b/drivers/gpu/drm/vc4/vc4_plane.c
index f7a229df572d..99f4d4b48015 100644
--- a/drivers/gpu/drm/vc4/vc4_plane.c
+++ b/drivers/gpu/drm/vc4/vc4_plane.c
@@ -498,8 +498,8 @@ 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;
@@ -540,11 +540,31 @@ 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));
@@ -598,8 +618,11 @@ 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 385405a2df05..362d8b7f8a5f 100644
--- a/drivers/gpu/drm/vc4/vc4_regs.h
+++ b/drivers/gpu/drm/vc4/vc4_regs.h
@@ -604,6 +604,13 @@ 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)
@@ -733,7 +740,19 @@ 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 */
diff --git a/include/uapi/drm/drm_fourcc.h b/include/uapi/drm/drm_fourcc.h
index ef20abb8119b..9aaf633788a7 100644
--- a/include/uapi/drm/drm_fourcc.h
+++ b/include/uapi/drm/drm_fourcc.h
@@ -168,6 +168,7 @@ extern "C" {
#define DRM_FORMAT_MOD_VENDOR_SAMSUNG 0x04
#define DRM_FORMAT_MOD_VENDOR_QCOM 0x05
#define DRM_FORMAT_MOD_VENDOR_VIVANTE 0x06
+#define DRM_FORMAT_MOD_VENDOR_BROADCOM 0x07
/* add more to the end as needed */
#define fourcc_mod_code(vendor, val) \
@@ -292,6 +293,27 @@ extern "C" {
*/
#define DRM_FORMAT_MOD_VIVANTE_SPLIT_SUPER_TILED fourcc_mod_code(VIVANTE, 4)
+/*
+ * Broadcom VC4 "T" format
+ *
+ * This is the primary layout that the V3D GPU can texture from (it
+ * can't do linear). The T format has:
+ *
+ * - 64b utiles of pixels in a raster-order grid according to cpp. It's 4x4
+ * pixels at 32 bit depth.
+ *
+ * - 1k subtiles made of a 4x4 raster-order grid of 64b utiles (so usually
+ * 16x16 pixels).
+ *
+ * - 4k tiles made of a 2x2 grid of 1k subtiles (so usually 32x32 pixels). On
+ * even 4k tile rows, they're arranged as (BL, TL, TR, BR), and on odd rows
+ * they're (TR, BR, BL, TL), where bottom left is start of memory.
+ *
+ * - an image made of 4k tiles in rows either left-to-right (even rows of 4k
+ * tiles) or right-to-left (odd rows of 4k tiles).
+ */
+#define DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED fourcc_mod_code(BROADCOM, 1)
+
#if defined(__cplusplus)
}
#endif
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
--
2.13.0