github-mirror/linux
1
0
Fork 0
mirror of https://github.com/torvalds/linux.git synced 2026-05-31 18:43:33 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity Graph

gpu: host1x: Wait prefences outside MLOCK

Browse Source 
The current submission opcode sequence first takes the engine MLOCK,
and then switches to HOST1X class to wait prefences. This is fine
while we only use a single channel per engine and there is no
virtualization, since jobs are serialized on that one channel anyway.
However, when that assumption doesn't hold, we are keeping the
engine locked while not running anything on it while waiting for
prefences to complete.

To resolve this, execute wait commands in the beginning of the job
outside the engine MLOCK. We still take the HOST1X MLOCK because
recent hardware requires register opcodes to be executed within some
MLOCK, but the hardware also allows unlimited channels to take the
HOST1X MLOCK at the same time.

Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20250708-host1x-wait-prefences-outside-mlock-v1-1-13e98044e35a@nvidia.com
This commit is contained in:
Mikko Perttunen 2025-07-08 20:25:08 +09:00 committed by Thierry Reding
parent c7d393267c
commit 63d47cc6ee
1 changed files with 69 additions and 43 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

112
drivers/gpu/host1x/hw/channel_hw.c
View File

@ -47,8 +47,36 @@ static void trace_write_gather(struct host1x_cdma *cdma, struct host1x_bo *bo,
}
}
static void submit_wait(struct host1x_job *job, u32 id, u32 threshold,
u32 next_class)
static void submit_wait(struct host1x_job *job, u32 id, u32 threshold)
{
struct host1x_cdma *cdma = &job->channel->cdma;
#if HOST1X_HW >= 2
host1x_cdma_push_wide(cdma,
host1x_opcode_setclass(
HOST1X_CLASS_HOST1X,
HOST1X_UCLASS_LOAD_SYNCPT_PAYLOAD_32,
/* WAIT_SYNCPT_32 is at SYNCPT_PAYLOAD_32+2 */
BIT(0) | BIT(2)
),
threshold,
id,
HOST1X_OPCODE_NOP
);
#else
/* TODO add waitchk or use waitbases or other mitigation */
host1x_cdma_push(cdma,
host1x_opcode_setclass(
HOST1X_CLASS_HOST1X,
host1x_uclass_wait_syncpt_r(),
BIT(0)
),
host1x_class_host_wait_syncpt(id, threshold)
);
#endif
}
static void submit_setclass(struct host1x_job *job, u32 next_class)
{
struct host1x_cdma *cdma = &job->channel->cdma;
@ -66,43 +94,11 @@ static void submit_wait(struct host1x_job *job, u32 id, u32 threshold,
stream_id = job->engine_fallback_streamid;
host1x_cdma_push_wide(cdma,
host1x_opcode_setclass(
HOST1X_CLASS_HOST1X,
HOST1X_UCLASS_LOAD_SYNCPT_PAYLOAD_32,
/* WAIT_SYNCPT_32 is at SYNCPT_PAYLOAD_32+2 */
BIT(0) | BIT(2)
),
threshold,
id,
HOST1X_OPCODE_NOP
);
host1x_cdma_push_wide(&job->channel->cdma,
host1x_opcode_setclass(job->class, 0, 0),
host1x_opcode_setclass(next_class, 0, 0),
host1x_opcode_setpayload(stream_id),
host1x_opcode_setstreamid(job->engine_streamid_offset / 4),
HOST1X_OPCODE_NOP);
#elif HOST1X_HW >= 2
host1x_cdma_push_wide(cdma,
host1x_opcode_setclass(
HOST1X_CLASS_HOST1X,
HOST1X_UCLASS_LOAD_SYNCPT_PAYLOAD_32,
/* WAIT_SYNCPT_32 is at SYNCPT_PAYLOAD_32+2 */
BIT(0) | BIT(2)
),
threshold,
id,
host1x_opcode_setclass(next_class, 0, 0)
);
#else
/* TODO add waitchk or use waitbases or other mitigation */
host1x_cdma_push(cdma,
host1x_opcode_setclass(
HOST1X_CLASS_HOST1X,
host1x_uclass_wait_syncpt_r(),
BIT(0)
),
host1x_class_host_wait_syncpt(id, threshold)
);
host1x_cdma_push(cdma,
host1x_opcode_setclass(next_class, 0, 0),
HOST1X_OPCODE_NOP
@ -110,7 +106,8 @@ static void submit_wait(struct host1x_job *job, u32 id, u32 threshold,
#endif
}
static void submit_gathers(struct host1x_job *job, u32 job_syncpt_base)
static void submit_gathers(struct host1x_job *job, struct host1x_job_cmd *cmds, u32 num_cmds,
u32 job_syncpt_base)
{
struct host1x_cdma *cdma = &job->channel->cdma;
#if HOST1X_HW < 6
@ -119,8 +116,8 @@ static void submit_gathers(struct host1x_job *job, u32 job_syncpt_base)
unsigned int i;
u32 threshold;
for (i = 0; i < job->num_cmds; i++) {
struct host1x_job_cmd *cmd = &job->cmds[i];
for (i = 0; i < num_cmds; i++) {
struct host1x_job_cmd *cmd = &cmds[i];
if (cmd->is_wait) {
if (cmd->wait.relative)
@ -128,7 +125,8 @@ static void submit_gathers(struct host1x_job *job, u32 job_syncpt_base)
else
threshold = cmd->wait.threshold;
submit_wait(job, cmd->wait.id, threshold, cmd->wait.next_class);
submit_wait(job, cmd->wait.id, threshold);
submit_setclass(job, cmd->wait.next_class);
} else {
struct host1x_job_gather *g = &cmd->gather;
@ -216,7 +214,34 @@ static void channel_program_cdma(struct host1x_job *job)
#if HOST1X_HW >= 6
u32 fence;
int i = 0;
if (job->num_cmds == 0)
goto prefences_done;
if (!job->cmds[0].is_wait || job->cmds[0].wait.relative)
goto prefences_done;
/* Enter host1x class with invalid stream ID for prefence waits. */
host1x_cdma_push_wide(cdma,
host1x_opcode_acquire_mlock(1),
host1x_opcode_setclass(1, 0, 0),
host1x_opcode_setpayload(0),
host1x_opcode_setstreamid(0x1fffff));
for (i = 0; i < job->num_cmds; i++) {
struct host1x_job_cmd *cmd = &job->cmds[i];
if (!cmd->is_wait || cmd->wait.relative)
break;
submit_wait(job, cmd->wait.id, cmd->wait.threshold);
}
host1x_cdma_push(cdma,
HOST1X_OPCODE_NOP,
host1x_opcode_release_mlock(1));
prefences_done:
/* Enter engine class with invalid stream ID. */
host1x_cdma_push_wide(cdma,
host1x_opcode_acquire_mlock(job->class),
@ -230,11 +255,12 @@ static void channel_program_cdma(struct host1x_job *job)
host1x_opcode_nonincr(HOST1X_UCLASS_INCR_SYNCPT, 1),
HOST1X_UCLASS_INCR_SYNCPT_INDX_F(job->syncpt->id) |
HOST1X_UCLASS_INCR_SYNCPT_COND_F(4));
submit_wait(job, job->syncpt->id, fence, job->class);
submit_wait(job, job->syncpt->id, fence);
submit_setclass(job, job->class);
/* Submit work. */
job->syncpt_end = host1x_syncpt_incr_max(sp, job->syncpt_incrs);
submit_gathers(job, job->syncpt_end - job->syncpt_incrs);
submit_gathers(job, job->cmds + i, job->num_cmds - i, job->syncpt_end - job->syncpt_incrs);
/* Before releasing MLOCK, ensure engine is idle again. */
fence = host1x_syncpt_incr_max(sp, 1);
@ -242,7 +268,7 @@ static void channel_program_cdma(struct host1x_job *job)
host1x_opcode_nonincr(HOST1X_UCLASS_INCR_SYNCPT, 1),
HOST1X_UCLASS_INCR_SYNCPT_INDX_F(job->syncpt->id) |
HOST1X_UCLASS_INCR_SYNCPT_COND_F(4));
submit_wait(job, job->syncpt->id, fence, job->class);
submit_wait(job, job->syncpt->id, fence);
/* Release MLOCK. */
host1x_cdma_push(cdma,
@ -272,7 +298,7 @@ static void channel_program_cdma(struct host1x_job *job)
job->syncpt_end = host1x_syncpt_incr_max(sp, job->syncpt_incrs);
submit_gathers(job, job->syncpt_end - job->syncpt_incrs);
submit_gathers(job, job->cmds, job->num_cmds, job->syncpt_end - job->syncpt_incrs);
#endif
}