[PATCH] gpu: host1x: Wait prefences outside MLOCK
Mikko Perttunen
cyndis at kapsi.fi
Tue Jul 8 11:25:08 UTC 2025
From: Mikko Perttunen <mperttunen at nvidia.com>
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 at nvidia.com>
---
drivers/gpu/host1x/hw/channel_hw.c | 106 +++++++++++++++++++++++--------------
1 file changed, 66 insertions(+), 40 deletions(-)
diff --git a/drivers/gpu/host1x/hw/channel_hw.c b/drivers/gpu/host1x/hw/channel_hw.c
index d44b8de890be05b697a7c4b5be697b708cf41cbb..2df6a16d484e046c9aed6bdf6b59f22de573c9d7 100644
--- a/drivers/gpu/host1x/hw/channel_hw.c
+++ b/drivers/gpu/host1x/hw/channel_hw.c
@@ -47,24 +47,11 @@ 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 >= 6
- u32 stream_id;
-
- /*
- * If a memory context has been set, use it. Otherwise
- * (if context isolation is disabled) use the engine's
- * firmware stream ID.
- */
- if (job->memory_context)
- stream_id = job->memory_context->stream_id;
- else
- stream_id = job->engine_fallback_streamid;
-
+#if HOST1X_HW >= 2
host1x_cdma_push_wide(cdma,
host1x_opcode_setclass(
HOST1X_CLASS_HOST1X,
@@ -76,23 +63,6 @@ static void submit_wait(struct host1x_job *job, u32 id, u32 threshold,
id,
HOST1X_OPCODE_NOP
);
- host1x_cdma_push_wide(&job->channel->cdma,
- host1x_opcode_setclass(job->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,
@@ -103,6 +73,32 @@ static void submit_wait(struct host1x_job *job, u32 id, u32 threshold,
),
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;
+
+#if HOST1X_HW >= 6
+ u32 stream_id;
+
+ /*
+ * If a memory context has been set, use it. Otherwise
+ * (if context isolation is disabled) use the engine's
+ * firmware stream ID.
+ */
+ if (job->memory_context)
+ stream_id = job->memory_context->stream_id;
+ else
+ stream_id = job->engine_fallback_streamid;
+
+ host1x_cdma_push_wide(cdma,
+ host1x_opcode_setclass(next_class, 0, 0),
+ host1x_opcode_setpayload(stream_id),
+ host1x_opcode_setstreamid(job->engine_streamid_offset / 4),
+ HOST1X_OPCODE_NOP);
+#else
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
}
---
base-commit: 2aeda9592360c200085898a258c4754bfe879921
change-id: 20250708-host1x-wait-prefences-outside-mlock-bfd6022b9d92
More information about the dri-devel
mailing list