gem5-dev@gem5.org

The gem5 Developer List

View all threads

[S] Change in gem5/gem5[develop]: dev-amdgpu: Add a few MQD attributes to GPUFS checkpoint

VR
VISHNU RAMADAS (Gerrit)
Wed, Apr 26, 2023 7:42 PM

VISHNU RAMADAS has uploaded this change for review. (
https://gem5-review.googlesource.com/c/public/gem5/+/70077?usp=email )

Change subject: dev-amdgpu: Add a few MQD attributes to GPUFS checkpoint
......................................................................

dev-amdgpu: Add a few MQD attributes to GPUFS checkpoint

During GPUFS checkpoint restore, doorbells callbacks are created based
on certain MQD attributes. These callbacks are required to create new
SDMA doorbells. If these attributes are not present in the checkpoint,
the restore hangs indefinitely waiting for ioctl calls that access these
doorbells to finish execution. This commit adds the attributes required
for checkpoint restore to proceed.

Change-Id: Id3d1b7a2627d4c50133d923096495957a233f675

M src/dev/amdgpu/pm4_packet_processor.cc
1 file changed, 30 insertions(+), 0 deletions(-)

diff --git a/src/dev/amdgpu/pm4_packet_processor.cc
b/src/dev/amdgpu/pm4_packet_processor.cc
index 071fe8b..3690113 100644
--- a/src/dev/amdgpu/pm4_packet_processor.cc
+++ b/src/dev/amdgpu/pm4_packet_processor.cc
@@ -1025,6 +1025,11 @@
uint32_t pipe[num_queues];
uint32_t queue[num_queues];
bool privileged[num_queues];

  • uint32_t hqd_active[num_queues];

  • uint32_t hqd_vmid[num_queues];

  • Addr aql_rptr[num_queues];

  • uint32_t doorbell[num_queues];

  • uint32_t hqd_pq_control[num_queues];

    int i = 0;
    for (auto iter : queues) {
    @@ -1048,6 +1053,11 @@
    pipe[i] = q->pipe();
    queue[i] = q->queue();
    privileged[i] = q->privileged();

  •    hqd_active[i] = q->getMQD()->hqd_active;
    
  •    hqd_vmid[i] = q->getMQD()->hqd_vmid;
    
  •    aql_rptr[i] = q->getMQD()->aqlRptr;
    
  •    doorbell[i] = q->getMQD()->doorbell;
    
  •    hqd_pq_control[i] = q->getMQD()->hqd_pq_control;
        i++;
    }
    

@@ -1067,6 +1077,11 @@
SERIALIZE_ARRAY(pipe, num_queues);
SERIALIZE_ARRAY(queue, num_queues);
SERIALIZE_ARRAY(privileged, num_queues);

  • SERIALIZE_ARRAY(hqd_active, num_queues);
  • SERIALIZE_ARRAY(hqd_vmid, num_queues);
  • SERIALIZE_ARRAY(aql_rptr, num_queues);
  • SERIALIZE_ARRAY(doorbell, num_queues);
  • SERIALIZE_ARRAY(hqd_pq_control, num_queues);
    }

void
@@ -1093,6 +1108,11 @@
uint32_t pipe[num_queues];
uint32_t queue[num_queues];
bool privileged[num_queues];

  • uint32_t hqd_active[num_queues];

  • uint32_t hqd_vmid[num_queues];

  • Addr aql_rptr[num_queues];

  • uint32_t doorbell[num_queues];

  • uint32_t hqd_pq_control[num_queues];

    UNSERIALIZE_ARRAY(id, num_queues);
    UNSERIALIZE_ARRAY(mqd_base, num_queues);
    @@ -1109,6 +1129,11 @@
    UNSERIALIZE_ARRAY(pipe, num_queues);
    UNSERIALIZE_ARRAY(queue, num_queues);
    UNSERIALIZE_ARRAY(privileged, num_queues);

  • UNSERIALIZE_ARRAY(hqd_active, num_queues);

  • UNSERIALIZE_ARRAY(hqd_vmid, num_queues);

  • UNSERIALIZE_ARRAY(aql_rptr, num_queues);

  • UNSERIALIZE_ARRAY(doorbell, num_queues);

  • UNSERIALIZE_ARRAY(hqd_pq_control, num_queues);

    for (int i = 0; i < num_queues; i++) {
    QueueDesc *mqd = new QueueDesc();
    @@ -1132,6 +1157,11 @@
    queues[id[i]]->processing(processing[i]);
    queues[id[i]]->ib(ib[i]);
    queues[id[i]]->setPkt(me[i], pipe[i], queue[i], privileged[i]);

  •    queues[id[i]]->getMQD()->hqd_active = hqd_active[i];
    
  •    queues[id[i]]->getMQD()->hqd_vmid = hqd_vmid[i];
    
  •    queues[id[i]]->getMQD()->aqlRptr = aql_rptr[i];
    
  •    queues[id[i]]->getMQD()->doorbell = doorbell[i];
    
  •    queues[id[i]]->getMQD()->hqd_pq_control = hqd_pq_control[i];
    
        DPRINTF(PM4PacketProcessor, "PM4 queue %d, rptr: %p wptr: %p\n",
                queues[id[i]]->id(), queues[id[i]]->rptr(),
    

--
To view, visit
https://gem5-review.googlesource.com/c/public/gem5/+/70077?usp=email
To unsubscribe, or for help writing mail filters, visit
https://gem5-review.googlesource.com/settings

Gerrit-MessageType: newchange
Gerrit-Project: public/gem5
Gerrit-Branch: develop
Gerrit-Change-Id: Id3d1b7a2627d4c50133d923096495957a233f675
Gerrit-Change-Number: 70077
Gerrit-PatchSet: 1
Gerrit-Owner: VISHNU RAMADAS vramadas@wisc.edu

VISHNU RAMADAS has uploaded this change for review. ( https://gem5-review.googlesource.com/c/public/gem5/+/70077?usp=email ) Change subject: dev-amdgpu: Add a few MQD attributes to GPUFS checkpoint ...................................................................... dev-amdgpu: Add a few MQD attributes to GPUFS checkpoint During GPUFS checkpoint restore, doorbells callbacks are created based on certain MQD attributes. These callbacks are required to create new SDMA doorbells. If these attributes are not present in the checkpoint, the restore hangs indefinitely waiting for ioctl calls that access these doorbells to finish execution. This commit adds the attributes required for checkpoint restore to proceed. Change-Id: Id3d1b7a2627d4c50133d923096495957a233f675 --- M src/dev/amdgpu/pm4_packet_processor.cc 1 file changed, 30 insertions(+), 0 deletions(-) diff --git a/src/dev/amdgpu/pm4_packet_processor.cc b/src/dev/amdgpu/pm4_packet_processor.cc index 071fe8b..3690113 100644 --- a/src/dev/amdgpu/pm4_packet_processor.cc +++ b/src/dev/amdgpu/pm4_packet_processor.cc @@ -1025,6 +1025,11 @@ uint32_t pipe[num_queues]; uint32_t queue[num_queues]; bool privileged[num_queues]; + uint32_t hqd_active[num_queues]; + uint32_t hqd_vmid[num_queues]; + Addr aql_rptr[num_queues]; + uint32_t doorbell[num_queues]; + uint32_t hqd_pq_control[num_queues]; int i = 0; for (auto iter : queues) { @@ -1048,6 +1053,11 @@ pipe[i] = q->pipe(); queue[i] = q->queue(); privileged[i] = q->privileged(); + hqd_active[i] = q->getMQD()->hqd_active; + hqd_vmid[i] = q->getMQD()->hqd_vmid; + aql_rptr[i] = q->getMQD()->aqlRptr; + doorbell[i] = q->getMQD()->doorbell; + hqd_pq_control[i] = q->getMQD()->hqd_pq_control; i++; } @@ -1067,6 +1077,11 @@ SERIALIZE_ARRAY(pipe, num_queues); SERIALIZE_ARRAY(queue, num_queues); SERIALIZE_ARRAY(privileged, num_queues); + SERIALIZE_ARRAY(hqd_active, num_queues); + SERIALIZE_ARRAY(hqd_vmid, num_queues); + SERIALIZE_ARRAY(aql_rptr, num_queues); + SERIALIZE_ARRAY(doorbell, num_queues); + SERIALIZE_ARRAY(hqd_pq_control, num_queues); } void @@ -1093,6 +1108,11 @@ uint32_t pipe[num_queues]; uint32_t queue[num_queues]; bool privileged[num_queues]; + uint32_t hqd_active[num_queues]; + uint32_t hqd_vmid[num_queues]; + Addr aql_rptr[num_queues]; + uint32_t doorbell[num_queues]; + uint32_t hqd_pq_control[num_queues]; UNSERIALIZE_ARRAY(id, num_queues); UNSERIALIZE_ARRAY(mqd_base, num_queues); @@ -1109,6 +1129,11 @@ UNSERIALIZE_ARRAY(pipe, num_queues); UNSERIALIZE_ARRAY(queue, num_queues); UNSERIALIZE_ARRAY(privileged, num_queues); + UNSERIALIZE_ARRAY(hqd_active, num_queues); + UNSERIALIZE_ARRAY(hqd_vmid, num_queues); + UNSERIALIZE_ARRAY(aql_rptr, num_queues); + UNSERIALIZE_ARRAY(doorbell, num_queues); + UNSERIALIZE_ARRAY(hqd_pq_control, num_queues); for (int i = 0; i < num_queues; i++) { QueueDesc *mqd = new QueueDesc(); @@ -1132,6 +1157,11 @@ queues[id[i]]->processing(processing[i]); queues[id[i]]->ib(ib[i]); queues[id[i]]->setPkt(me[i], pipe[i], queue[i], privileged[i]); + queues[id[i]]->getMQD()->hqd_active = hqd_active[i]; + queues[id[i]]->getMQD()->hqd_vmid = hqd_vmid[i]; + queues[id[i]]->getMQD()->aqlRptr = aql_rptr[i]; + queues[id[i]]->getMQD()->doorbell = doorbell[i]; + queues[id[i]]->getMQD()->hqd_pq_control = hqd_pq_control[i]; DPRINTF(PM4PacketProcessor, "PM4 queue %d, rptr: %p wptr: %p\n", queues[id[i]]->id(), queues[id[i]]->rptr(), -- To view, visit https://gem5-review.googlesource.com/c/public/gem5/+/70077?usp=email To unsubscribe, or for help writing mail filters, visit https://gem5-review.googlesource.com/settings Gerrit-MessageType: newchange Gerrit-Project: public/gem5 Gerrit-Branch: develop Gerrit-Change-Id: Id3d1b7a2627d4c50133d923096495957a233f675 Gerrit-Change-Number: 70077 Gerrit-PatchSet: 1 Gerrit-Owner: VISHNU RAMADAS <vramadas@wisc.edu>