From d54f375796045fa79a0d287494f331076ed43a98 Mon Sep 17 00:00:00 2001 From: Felix Kuehling Date: Fri, 12 Aug 2016 19:28:41 -0400 Subject: [PATCH 1180/4131] drm/amdkfd: Don't synchronize DIQ after a wave control IB Issuing a release_mem packet on the DIQ can hang the CP if a compute queue and the DIQ are running on the same pipe. Change-Id: Iea60140426bb98e84f15ee7b2010ea4fce5b53be Signed-off-by: Felix Kuehling --- drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c | 18 ++++++++++++++---- 1 file changed, 14 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c index 4f2311e..74109d0 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c @@ -47,14 +47,16 @@ static void dbgdev_address_watch_disable_nodiq(struct kfd_dev *dev) static int dbgdev_diq_submit_ib(struct kfd_dbgdev *dbgdev, unsigned int pasid, uint64_t vmid0_address, - uint32_t *packet_buff, size_t size_in_bytes) + uint32_t *packet_buff, size_t size_in_bytes, + bool sync) { int status = 0; unsigned int *ib_packet_buff = NULL; struct pm4__release_mem *rm_packet; struct pm4__indirect_buffer_pasid *ib_packet; struct kernel_queue *kq = dbgdev->kq; - size_t pq_packets_size_in_bytes = sizeof(struct pm4__release_mem) + sizeof(struct pm4__indirect_buffer_pasid); + size_t pq_packets_size_in_bytes = + sizeof(struct pm4__indirect_buffer_pasid); struct kfd_mem_obj *mem_obj; uint64_t *rm_state = NULL; @@ -74,6 +76,9 @@ static int dbgdev_diq_submit_ib(struct kfd_dbgdev *dbgdev, * The receive packet buff will be sitting on the Indirect Buffer * and in the PQ we put the IB packet + sync packet(s). */ + if (sync) + pq_packets_size_in_bytes += + sizeof(struct pm4__release_mem); status = kq->ops.acquire_packet_buffer(kq, pq_packets_size_in_bytes / sizeof(uint32_t), &ib_packet_buff); if (status != 0) { pr_debug("Error! kfd: In func %s >> acquire_packet_buffer failed\n", __func__); @@ -98,6 +103,11 @@ static int dbgdev_diq_submit_ib(struct kfd_dbgdev *dbgdev, ib_packet->bitfields5.pasid = pasid; + if (!sync) { + kq->ops.submit_packet(kq); + break; + } + /* * for now we use release mem for GPU-CPU synchronization * Consider WaitRegMem + WriteData as a better alternative @@ -491,7 +501,7 @@ static int dbgdev_address_watch_diq(struct kfd_dbgdev *dbgdev, adw_info->process->pasid, mem_obj->gpu_addr, packet_buff_uint, - ib_size); + ib_size, true); if (status != 0) { pr_debug("Error! kfd: In func %s >> failed to submit DIQ packet\n", __func__); @@ -707,7 +717,7 @@ static int dbgdev_wave_control_diq(struct kfd_dbgdev *dbgdev, wac_info->process->pasid, mem_obj->gpu_addr, packet_buff_uint, - ib_size); + ib_size, false); if (status != 0) pr_debug("%s\n", " Critical Error ! Submit diq packet failed "); -- 2.7.4