From: Thierry Reding <treding at nvidia.com>
Add a new nouveau_pushbuf_kick_fence() function that takes and emits a
sync fence FD. The fence FD can be waited on, or merged with other fence
FDs, or passed back to the kernel as a prerequisite for a subsequent HW
operation.
Based heavily on work by Lauri Peltonen <lpeltonen at nvidia.com>
Signed-off-by: Thierry Reding <treding at nvidia.com>
---
For the kernel patches that add the new IOCTL, see the series at:
https://patchwork.freedesktop.org/series/36361/
include/drm/nouveau_drm.h | 23 ++++++++
nouveau/nouveau.h | 2 +
nouveau/pushbuf.c | 130 ++++++++++++++++++++++++++++++++++------------
3 files changed, 121 insertions(+), 34 deletions(-)
diff --git a/include/drm/nouveau_drm.h b/include/drm/nouveau_drm.h
index cb077821c43f..e7f1a2aefd22 100644
--- a/include/drm/nouveau_drm.h
+++ b/include/drm/nouveau_drm.h
@@ -178,6 +178,28 @@ struct drm_nouveau_gem_pushbuf {
__u64 gart_available;
};
+#define NOUVEAU_GEM_PUSHBUF_FENCE_WAIT (1 << 0)
+#define NOUVEAU_GEM_PUSHBUF_FENCE_EMIT (1 << 1)
+#define NOUVEAU_GEM_PUSHBUF_FLAGS (NOUVEAU_GEM_PUSHBUF_FENCE_WAIT | \
+ NOUVEAU_GEM_PUSHBUF_FENCE_EMIT)
+
+struct drm_nouveau_gem_pushbuf2 {
+ uint32_t channel;
+ uint32_t nr_buffers;
+ uint64_t buffers;
+ uint32_t nr_relocs;
+ uint32_t nr_push;
+ uint64_t relocs;
+ uint64_t push;
+ uint32_t suffix0;
+ uint32_t suffix1;
+ uint64_t vram_available;
+ uint64_t gart_available;
+ uint32_t flags;
+ int32_t fence;
+ uint64_t reserved;
+};
+
#define NOUVEAU_GEM_CPU_PREP_NOWAIT 0x00000001
#define NOUVEAU_GEM_CPU_PREP_NOBLOCK 0x00000002
#define NOUVEAU_GEM_CPU_PREP_WRITE 0x00000004
@@ -212,6 +234,7 @@ struct drm_nouveau_sarea {
#define DRM_NOUVEAU_GEM_CPU_PREP 0x42
#define DRM_NOUVEAU_GEM_CPU_FINI 0x43
#define DRM_NOUVEAU_GEM_INFO 0x44
+#define DRM_NOUVEAU_GEM_PUSHBUF2 0x45
#if defined(__cplusplus)
}
diff --git a/nouveau/nouveau.h b/nouveau/nouveau.h
index 335ce77dca77..70d680700faf 100644
--- a/nouveau/nouveau.h
+++ b/nouveau/nouveau.h
@@ -226,6 +226,8 @@ void nouveau_pushbuf_reloc(struct nouveau_pushbuf *, struct
nouveau_bo *,
int nouveau_pushbuf_validate(struct nouveau_pushbuf *);
uint32_t nouveau_pushbuf_refd(struct nouveau_pushbuf *, struct nouveau_bo *);
int nouveau_pushbuf_kick(struct nouveau_pushbuf *, struct nouveau_object
*chan);
+int nouveau_pushbuf_kick_fence(struct nouveau_pushbuf *,
+ struct nouveau_object *chan, int *fence);
struct nouveau_bufctx *
nouveau_pushbuf_bufctx(struct nouveau_pushbuf *, struct nouveau_bufctx *);
diff --git a/nouveau/pushbuf.c b/nouveau/pushbuf.c
index 035e3019f2cd..f13804db7534 100644
--- a/nouveau/pushbuf.c
+++ b/nouveau/pushbuf.c
@@ -33,6 +33,7 @@
#include <string.h>
#include <assert.h>
#include <errno.h>
+#include <unistd.h>
#include <xf86drm.h>
#include <xf86atomic.h>
@@ -77,7 +78,7 @@ nouveau_pushbuf(struct nouveau_pushbuf *push)
}
static int pushbuf_validate(struct nouveau_pushbuf *, bool);
-static int pushbuf_flush(struct nouveau_pushbuf *);
+static int pushbuf_flush(struct nouveau_pushbuf *, int *);
static bool
pushbuf_kref_fits(struct nouveau_pushbuf *push, struct nouveau_bo *bo,
@@ -172,7 +173,7 @@ pushbuf_kref(struct nouveau_pushbuf *push, struct nouveau_bo
*bo,
*/
fpush = cli_push_get(push->client, bo);
if (fpush && fpush != push)
- pushbuf_flush(fpush);
+ pushbuf_flush(fpush, NULL);
kref = cli_kref_get(push->client, bo);
if (kref) {
@@ -307,7 +308,8 @@ pushbuf_dump(struct nouveau_pushbuf_krec *krec, int krec_id,
int chid)
}
static int
-pushbuf_submit(struct nouveau_pushbuf *push, struct nouveau_object *chan)
+pushbuf_submit(struct nouveau_pushbuf *push, struct nouveau_object *chan,
+ int *fence)
{
struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
struct nouveau_pushbuf_krec *krec = nvpb->list;
@@ -315,9 +317,9 @@ pushbuf_submit(struct nouveau_pushbuf *push, struct
nouveau_object *chan)
struct nouveau_drm *drm = nouveau_drm(&dev->object);
struct drm_nouveau_gem_pushbuf_bo_presumed *info;
struct drm_nouveau_gem_pushbuf_bo *kref;
- struct drm_nouveau_gem_pushbuf req;
struct nouveau_fifo *fifo = chan->data;
struct nouveau_bo *bo;
+ int fence_out = -1;
int krec_id = 0;
int ret = 0, i;
@@ -330,35 +332,81 @@ pushbuf_submit(struct nouveau_pushbuf *push, struct
nouveau_object *chan)
nouveau_pushbuf_data(push, NULL, 0, 0);
while (krec && krec->nr_push) {
- req.channel = fifo->channel;
- req.nr_buffers = krec->nr_buffer;
- req.buffers = (uint64_t)(unsigned long)krec->buffer;
- req.nr_relocs = krec->nr_reloc;
- req.nr_push = krec->nr_push;
- req.relocs = (uint64_t)(unsigned long)krec->reloc;
- req.push = (uint64_t)(unsigned long)krec->push;
- req.suffix0 = nvpb->suffix0;
- req.suffix1 = nvpb->suffix1;
- req.vram_available = 0; /* for valgrind */
- req.gart_available = 0;
-
if (dbg_on(0))
pushbuf_dump(krec, krec_id++, fifo->channel);
+ /* TODO If fence is requested, force kickoff. */
+ if (fence) {
+ struct drm_nouveau_gem_pushbuf2 req;
+
+ memset(&req, 0, sizeof(req));
+ req.channel = fifo->channel;
+ req.nr_buffers = krec->nr_buffer;
+ req.buffers = (uint64_t)(unsigned long)krec->buffer;
+ req.nr_relocs = krec->nr_reloc;
+ req.nr_push = krec->nr_push;
+ req.relocs = (uint64_t)(unsigned long)krec->reloc;
+ req.push = (uint64_t)(unsigned long)krec->push;
+ req.suffix0 = nvpb->suffix0;
+ req.suffix1 = nvpb->suffix1;
+ req.vram_available = 0; /* for valgrind */
+ req.gart_available = 0;
+ req.flags = 0;
+
+ if (*fence >= 0)
+ req.flags |= NOUVEAU_GEM_PUSHBUF_FENCE_WAIT;
+
+ req.flags |= NOUVEAU_GEM_PUSHBUF_FENCE_EMIT;
+
+ req.fence = *fence;
+ req.reserved = 0;
+
#ifndef SIMULATE
- ret = drmCommandWriteRead(drm->fd, DRM_NOUVEAU_GEM_PUSHBUF,
- &req, sizeof(req));
- nvpb->suffix0 = req.suffix0;
- nvpb->suffix1 = req.suffix1;
- dev->vram_limit = (req.vram_available *
- nouveau_device(dev)->vram_limit_percent) / 100;
- dev->gart_limit = (req.gart_available *
- nouveau_device(dev)->gart_limit_percent) / 100;
+ ret = drmCommandWriteRead(drm->fd, DRM_NOUVEAU_GEM_PUSHBUF2,
+ &req, sizeof(req));
+ nvpb->suffix0 = req.suffix0;
+ nvpb->suffix1 = req.suffix1;
+ dev->vram_limit = (req.vram_available *
+ nouveau_device(dev)->vram_limit_percent) / 100;
+ dev->gart_limit = (req.gart_available *
+ nouveau_device(dev)->gart_limit_percent) / 100;
#else
- if (dbg_on(31))
- ret = -EINVAL;
+ if (dbg_on(31))
+ ret = -EINVAL;
#endif
+ if (!ret)
+ fence_out = req.fence;
+ } else {
+ struct drm_nouveau_gem_pushbuf req;
+
+ req.channel = fifo->channel;
+ req.nr_buffers = krec->nr_buffer;
+ req.buffers = (uint64_t)(unsigned long)krec->buffer;
+ req.nr_relocs = krec->nr_reloc;
+ req.nr_push = krec->nr_push;
+ req.relocs = (uint64_t)(unsigned long)krec->reloc;
+ req.push = (uint64_t)(unsigned long)krec->push;
+ req.suffix0 = nvpb->suffix0;
+ req.suffix1 = nvpb->suffix1;
+ req.vram_available = 0; /* for valgrind */
+ req.gart_available = 0;
+
+#ifndef SIMULATE
+ ret = drmCommandWriteRead(drm->fd, DRM_NOUVEAU_GEM_PUSHBUF,
+ &req, sizeof(req));
+ nvpb->suffix0 = req.suffix0;
+ nvpb->suffix1 = req.suffix1;
+ dev->vram_limit = (req.vram_available *
+ nouveau_device(dev)->vram_limit_percent) / 100;
+ dev->gart_limit = (req.gart_available *
+ nouveau_device(dev)->gart_limit_percent) / 100;
+#else
+ if (dbg_on(31))
+ ret = -EINVAL;
+#endif
+ }
+
if (ret) {
err("kernel rejected pushbuf: %s\n", strerror(-ret));
pushbuf_dump(krec, krec_id++, fifo->channel);
@@ -388,11 +436,18 @@ pushbuf_submit(struct nouveau_pushbuf *push, struct
nouveau_object *chan)
krec = krec->next;
}
+ if (!ret && fence) {
+ if (*fence >= 0)
+ close(*fence);
+
+ *fence = fence_out;
+ }
+
return ret;
}
static int
-pushbuf_flush(struct nouveau_pushbuf *push)
+pushbuf_flush(struct nouveau_pushbuf *push, int *fence)
{
struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
struct nouveau_pushbuf_krec *krec = nvpb->krec;
@@ -402,7 +457,7 @@ pushbuf_flush(struct nouveau_pushbuf *push)
int ret = 0, i;
if (push->channel) {
- ret = pushbuf_submit(push, push->channel);
+ ret = pushbuf_submit(push, push->channel, fence);
} else {
nouveau_pushbuf_data(push, NULL, 0, 0);
krec->next = malloc(sizeof(*krec));
@@ -472,7 +527,7 @@ pushbuf_refn(struct nouveau_pushbuf *push, bool retry,
if (ret) {
pushbuf_refn_fail(push, sref, krec->nr_reloc);
if (retry) {
- pushbuf_flush(push);
+ pushbuf_flush(push, NULL);
nouveau_pushbuf_space(push, 0, 0, 0);
return pushbuf_refn(push, false, refs, nr);
}
@@ -524,7 +579,7 @@ pushbuf_validate(struct nouveau_pushbuf *push, bool retry)
if (ret) {
pushbuf_refn_fail(push, sref, srel);
if (retry) {
- pushbuf_flush(push);
+ pushbuf_flush(push, NULL);
return pushbuf_validate(push, false);
}
}
@@ -676,7 +731,7 @@ nouveau_pushbuf_space(struct nouveau_pushbuf *push,
krec->nr_reloc + relocs >= NOUVEAU_GEM_MAX_RELOCS ||
krec->nr_push + pushes >= NOUVEAU_GEM_MAX_PUSH) {
if (nvpb->bo && krec->nr_buffer)
- pushbuf_flush(push);
+ pushbuf_flush(push, NULL);
flushed = true;
}
@@ -772,10 +827,17 @@ nouveau_pushbuf_refd(struct nouveau_pushbuf *push, struct
nouveau_bo *bo)
}
int
-nouveau_pushbuf_kick(struct nouveau_pushbuf *push, struct nouveau_object *chan)
+nouveau_pushbuf_kick_fence(struct nouveau_pushbuf *push,
+ struct nouveau_object *chan, int *fence)
{
if (!push->channel)
- return pushbuf_submit(push, chan);
- pushbuf_flush(push);
+ return pushbuf_submit(push, chan, fence);
+ pushbuf_flush(push, fence);
return pushbuf_validate(push, false);
}
+
+int
+nouveau_pushbuf_kick(struct nouveau_pushbuf *pushbuf, struct nouveau_object
*chan)
+{
+ return nouveau_pushbuf_kick_fence(pushbuf, chan, NULL);
+}
--
2.15.1
