nv-tegra-mirror/linux-nvgpu
1
0
Fork 0
mirror of git://nv-tegra.nvidia.com/linux-nvgpu.git synced 2025-12-23 09:57:08 +03:00
Code Issues Packages Projects Releases Wiki Activity

gpu: nvgpu: support skipping buffer refcounting in submit

Browse Source 
In job submission path, we always take refcount on all
the mapped buffers to safeguard against case where user
space releases the buffer early

But in case user space itself is doing proper buffer
management, kernel need not take refcounts on all the
buffers - which is also a overhead in submit path

Hence, provide a new submit flag
NVGPU_SUBMIT_GPFIFO_FLAGS_SKIP_BUFFER_REFCOUNTING to
optionally skip taking refcounts on all the buffers

Also, if we do not take refcounts, then no need to drop
any refcounts in gk20a_channel_update() as well

Bug 1698667
Bug 200141116

Change-Id: I81bb7a03240300b691c70bcec04ea1badd5934f4
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/824718
(cherry picked from commit 8c8978fa303ec4e6db0233becdbdcbad4a248173)
Reviewed-on: http://git-master/r/835801
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>
This commit is contained in:
Deepak Nibade
2015-10-29 15:20:50 +05:30
committed by Terje Bergstrom
parent 67fe5f6d73
commit f50d0ffb15
2 changed files with 18 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
drivers/gpu/nvgpu/gk20a/channel_gk20a.c
View File

@@ -1726,21 +1726,25 @@ fail_unlock:
static int gk20a_channel_add_job(struct channel_gk20a *c, static int gk20a_channel_add_job(struct channel_gk20a *c,
struct gk20a_fence *pre_fence, struct gk20a_fence *pre_fence,
struct gk20a_fence *post_fence) struct gk20a_fence *post_fence,
bool skip_buffer_refcounting)
{ {
struct vm_gk20a *vm = c->vm; struct vm_gk20a *vm = c->vm;
struct channel_gk20a_job *job = NULL; struct channel_gk20a_job *job = NULL;
struct mapped_buffer_node **mapped_buffers = NULL; struct mapped_buffer_node **mapped_buffers = NULL;
int err = 0, num_mapped_buffers; int err = 0, num_mapped_buffers = 0;
/* job needs reference to this vm (released in channel_update) */ /* job needs reference to this vm (released in channel_update) */
gk20a_vm_get(vm); gk20a_vm_get(vm);
err = gk20a_vm_get_buffers(vm, &mapped_buffers, &num_mapped_buffers); if (!skip_buffer_refcounting) {
err = gk20a_vm_get_buffers(vm, &mapped_buffers,
&num_mapped_buffers);
if (err) { if (err) {
gk20a_vm_put(vm); gk20a_vm_put(vm);
return err; return err;
} }
}
job = kzalloc(sizeof(*job), GFP_KERNEL); job = kzalloc(sizeof(*job), GFP_KERNEL);
if (!job) { if (!job) {
@@ -1795,6 +1799,7 @@ void gk20a_channel_update(struct channel_gk20a *c, int nr_completed)
if (c->sync) if (c->sync)
c->sync->signal_timeline(c->sync); c->sync->signal_timeline(c->sync);
if (job->num_mapped_buffers)
gk20a_vm_put_buffers(vm, job->mapped_buffers, gk20a_vm_put_buffers(vm, job->mapped_buffers,
job->num_mapped_buffers); job->num_mapped_buffers);
@@ -1858,6 +1863,8 @@ int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
const int extra_entries = 2; const int extra_entries = 2;
bool need_wfi = !(flags & NVGPU_SUBMIT_GPFIFO_FLAGS_SUPPRESS_WFI); bool need_wfi = !(flags & NVGPU_SUBMIT_GPFIFO_FLAGS_SUPPRESS_WFI);
struct nvgpu_gpfifo *gpfifo_mem = c->gpfifo.mem.cpu_va; struct nvgpu_gpfifo *gpfifo_mem = c->gpfifo.mem.cpu_va;
bool skip_buffer_refcounting = (flags &
NVGPU_SUBMIT_GPFIFO_FLAGS_SKIP_BUFFER_REFCOUNTING);
if (c->has_timedout) if (c->has_timedout)
return -ETIMEDOUT; return -ETIMEDOUT;
@@ -2106,7 +2113,8 @@ int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
*fence_out = gk20a_fence_get(post_fence); *fence_out = gk20a_fence_get(post_fence);
/* TODO! Check for errors... */ /* TODO! Check for errors... */
gk20a_channel_add_job(c, pre_fence, post_fence); gk20a_channel_add_job(c, pre_fence, post_fence,
skip_buffer_refcounting);
c->cmds_pending = true; c->cmds_pending = true;
gk20a_bar1_writel(g, gk20a_bar1_writel(g,

2
include/uapi/linux/nvgpu.h
View File

@@ -669,6 +669,8 @@ struct nvgpu_fence {
#define NVGPU_SUBMIT_GPFIFO_FLAGS_SYNC_FENCE (1 << 3) #define NVGPU_SUBMIT_GPFIFO_FLAGS_SYNC_FENCE (1 << 3)
/* suppress WFI before fence trigger */ /* suppress WFI before fence trigger */
#define NVGPU_SUBMIT_GPFIFO_FLAGS_SUPPRESS_WFI (1 << 4) #define NVGPU_SUBMIT_GPFIFO_FLAGS_SUPPRESS_WFI (1 << 4)
/* skip buffer refcounting during submit */
#define NVGPU_SUBMIT_GPFIFO_FLAGS_SKIP_BUFFER_REFCOUNTING (1 << 5)
struct nvgpu_submit_gpfifo_args { struct nvgpu_submit_gpfifo_args {
__u64 gpfifo; __u64 gpfifo;