<html> <head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> <style type="text/css" style="display:none;"> P {margin-top:0;margin-bottom:0;} </style> </head> <body dir="ltr"> [Public] <div> <div class="elementToProof" style="font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> Hi Jesse,</div> <div class="elementToProof" style="font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> </div> <div class="elementToProof" style="font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> I checked the timeout code in amdgpu_userqueue_submit(), this does not satisfy the requirement for fwm_preemption test case I am adding.</div> <div class="elementToProof" style="font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> </div> <div class="elementToProof" style="font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> In case of fwm_premption test case, unsatisfied fence will be added to FENCE_WAIT_MULTI packet, so amdgpu_cs_syncobj_wait will always timeout initially.</div> <div class="elementToProof" style="font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> This will lead to syncobj wait assert. I will have to skip synobj wait code in amdgpu_userqueue_submit().</div> <div class="elementToProof" style="font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> </div> <div class="elementToProof" style="font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> Thank you,</div> <div class="elementToProof" style="font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> Yogesh</div> <div id="appendonsend"></div> <div class="elementToProof" style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> </div> <hr style="display: inline-block; width: 98%;"> <div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> From: igt-dev <igt-dev-bounces@lists.freedesktop.org> on behalf of Mohan Marimuthu, Yogesh <Yogesh.Mohanmarimuthu@amd.com> Sent: Friday, May 16, 2025 3:39 PM To: Zhang, Jesse(Jie) <Jesse.Zhang@amd.com>; igt-dev@lists.freedesktop.org <igt-dev@lists.freedesktop.org> Cc: Prosyak, Vitaly <Vitaly.Prosyak@amd.com> Subject: Re: [PATCH i-g-t 1/2] tests/amdgpu: userq skip waiting for signal fence </div> <div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);"> </div> [Public] <div style="direction: ltr;"> </div> [Public] <div style="direction: ltr;"> </div> <div style="direction: ltr; font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: black;"> Hi Jesse,</div> <div style="direction: ltr; font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: black;"> </div> <div style="direction: ltr; font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: black;"> Thank you for the review comments.</div> <div style="direction: ltr; font-family: Aptos, Aptos_EmbeddedFont, Aptos_MSFontService, Calibri, Helvetica, sans-serif; font-size: 12pt; color: black;"> I will update the test case and send revised patch.</div> <div id="x_appendonsend"></div> <div style="direction: ltr;"> </div> <div style="direction: ltr; font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: black;"> Thank you,</div> <div style="direction: ltr; font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: black;"> Yogesh</div> <hr style="direction: ltr; display: inline-block; width: 98%;"> <div style="direction: ltr; font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: black;"> From: Zhang, Jesse(Jie) <Jesse.Zhang@amd.com> Sent: Friday, May 16, 2025 2:36 PM To: Mohan Marimuthu, Yogesh <Yogesh.Mohanmarimuthu@amd.com>; igt-dev@lists.freedesktop.org <igt-dev@lists.freedesktop.org> Cc: Prosyak, Vitaly <Vitaly.Prosyak@amd.com> Subject: RE: [PATCH i-g-t 1/2] tests/amdgpu: userq skip waiting for signal fence</div> <div style="direction: ltr; font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: black;"> </div> [Public] <div style="direction: ltr;"> </div> hi Yogesh, Please pull the latest code, we can set a shorter timeout for amdgpu_cs_syncobj_wait to replace the skip flag. Thanks Jesse <div style="padding-top: 3pt; border-top: 1pt solid rgb(225, 225, 225);"> From: igt-dev <igt-dev-bounces@lists.freedesktop.org> On Behalf Of Mohan Marimuthu, Yogesh Sent: Friday, May 16, 2025 4:31 PM To: igt-dev@lists.freedesktop.org Cc: Prosyak, Vitaly <Vitaly.Prosyak@amd.com> Subject: [PATCH i-g-t 1/2] tests/amdgpu: userq skip waiting for signal fence </div> [Public] [Public] For negative test cases where the job will not complete need to skip signal fence. Pass a flag to amdgpu_user_queue_submit() to skip signal fence wait. Cc: Prosyak, Vitaly <<a href="mailto:Vitaly.Prosyak@amd.com" id="OWA5dbe1626-8429-1978-ee57-a5e751cc75a2" class="OWAAutoLink" style="margin-top: 0px; margin-bottom: 0px;">Vitaly.Prosyak@amd.com</a>> Signed-off-by: Yogesh Mohan Marimuthu <<a href="mailto:yogesh.mohanmarimuthu@amd.com" id="OWA6b352236-7559-955e-8011-ffcd967b1e98" class="OWAAutoLink" style="margin-top: 0px; margin-bottom: 0px;">yogesh.mohanmarimuthu@amd.com</a>> --- lib/amdgpu/amd_command_submission.c | 2 +- lib/amdgpu/amd_compute.c | 2 +- lib/amdgpu/amd_userq.c | 36 +++++++++++++++-------------- lib/amdgpu/amd_userq.h | 2 +- tests/amdgpu/amd_basic.c | 4 ++-- tests/amdgpu/amd_cs_nop.c | 2 +- 6 files changed, 25 insertions(+), 23 deletions(-) diff --git a/lib/amdgpu/amd_command_submission.c b/lib/amdgpu/amd_command_submission.c index 80d03a498..74091da5a 100644 --- a/lib/amdgpu/amd_command_submission.c +++ b/lib/amdgpu/amd_command_submission.c @@ -68,7 +68,7 @@ int amdgpu_test_exec_cs_helper(amdgpu_device_handle device, unsigned int ip_type memcpy(ring_ptr, ring_context->pm4, ring_context->pm4_dw * sizeof(*ring_context->pm4)); if (user_queue) - amdgpu_user_queue_submit(device, ring_context, ip_type, ib_result_mc_address); + amdgpu_user_queue_submit(device, ring_context, ip_type, ib_result_mc_address, false); else { ring_context->ib_info.ib_mc_address = ib_result_mc_address; ring_context->ib_info.size = ring_context->pm4_dw; diff --git a/lib/amdgpu/amd_compute.c b/lib/amdgpu/amd_compute.c index 95bfa53aa..008186049 100644 --- a/lib/amdgpu/amd_compute.c +++ b/lib/amdgpu/amd_compute.c @@ -91,7 +91,7 @@ void amdgpu_command_submission_compute_nop(amdgpu_device_handle device, bool use if (user_queue) { amdgpu_user_queue_submit(device, ring_context, AMD_IP_COMPUTE, - ib_result_mc_address); + ib_result_mc_address, false); } else { memset(&ib_info, 0, sizeof(struct amdgpu_cs_ib_info)); ib_info.ib_mc_address = ib_result_mc_address; diff --git a/lib/amdgpu/amd_userq.c b/lib/amdgpu/amd_userq.c index 50d058609..6593af822 100644 --- a/lib/amdgpu/amd_userq.c +++ b/lib/amdgpu/amd_userq.c @@ -127,7 +127,7 @@ int amdgpu_timeline_syncobj_wait(amdgpu_device_handle device_handle, } void amdgpu_user_queue_submit(amdgpu_device_handle device, struct amdgpu_ring_context *ring_context, - unsigned int ip_type, uint64_t mc_address) + unsigned int ip_type, uint64_t mc_address, bool skip_signal) { int r; uint32_t control = ring_context->pm4_dw; @@ -166,22 +166,24 @@ void amdgpu_user_queue_submit(amdgpu_device_handle device, struct amdgpu_ring_co /* Update the door bell */ ring_context->doorbell_cpu[DOORBELL_INDEX] = *ring_context->wptr_cpu; - /* Add a fence packet for signal */ - syncarray[0] = ring_context->timeline_syncobj_handle; - signal_data.queue_id = ring_context->queue_id; - signal_data.syncobj_handles = (uintptr_t)syncarray; - signal_data.num_syncobj_handles = 1; - signal_data.bo_read_handles = 0; - signal_data.bo_write_handles = 0; - signal_data.num_bo_read_handles = 0; - signal_data.num_bo_write_handles = 0; - - r = amdgpu_userq_signal(device, &signal_data); - igt_assert_eq(r, 0); + if (!skip_signal) { + /* Add a fence packet for signal */ + syncarray[0] = ring_context->timeline_syncobj_handle; + signal_data.queue_id = ring_context->queue_id; + signal_data.syncobj_handles = (uintptr_t)syncarray; + signal_data.num_syncobj_handles = 1; + signal_data.bo_read_handles = 0; + signal_data.bo_write_handles = 0; + signal_data.num_bo_read_handles = 0; + signal_data.num_bo_write_handles = 0; + + r = amdgpu_userq_signal(device, &signal_data); + igt_assert_eq(r, 0); - r = amdgpu_cs_syncobj_wait(device, &ring_context->timeline_syncobj_handle, 1, INT64_MAX, - DRM_SYNCOBJ_WAIT_FLAGS_WAIT_ALL, NULL); - igt_assert_eq(r, 0); + r = amdgpu_cs_syncobj_wait(device, &ring_context->timeline_syncobj_handle, 1, + INT64_MAX, DRM_SYNCOBJ_WAIT_FLAGS_WAIT_ALL, NULL); + igt_assert_eq(r, 0); + } } void amdgpu_user_queue_destroy(amdgpu_device_handle device_handle, struct amdgpu_ring_context *ctxt, @@ -456,7 +458,7 @@ int amdgpu_timeline_syncobj_wait(amdgpu_device_handle device_handle, } void amdgpu_user_queue_submit(amdgpu_device_handle device, struct amdgpu_ring_context *ring_context, - unsigned int ip_type, uint64_t mc_address) + unsigned int ip_type, uint64_t mc_address, bool skip_signal) { } diff --git a/lib/amdgpu/amd_userq.h b/lib/amdgpu/amd_userq.h index b29e97ccf..dc39c1ca4 100644 --- a/lib/amdgpu/amd_userq.h +++ b/lib/amdgpu/amd_userq.h @@ -50,6 +50,6 @@ void amdgpu_user_queue_destroy(amdgpu_device_handle device_handle, struct amdgpu unsigned int ip_type); void amdgpu_user_queue_submit(amdgpu_device_handle device, struct amdgpu_ring_context *ring_context, - unsigned int ip_type, uint64_t mc_address); + unsigned int ip_type, uint64_t mc_address, bool skip_signal); #endif diff --git a/tests/amdgpu/amd_basic.c b/tests/amdgpu/amd_basic.c index 97a08a9a3..914d27909 100644 --- a/tests/amdgpu/amd_basic.c +++ b/tests/amdgpu/amd_basic.c @@ -607,7 +607,7 @@ amdgpu_sync_dependency_test(amdgpu_device_handle device_handle, bool user_queue) if (user_queue) { ring_context->pm4_dw = ib_info.size; amdgpu_user_queue_submit(device_handle, ring_context, ip_block->type, - ib_result_mc_address); + ib_result_mc_address, false); } else { r = amdgpu_cs_submit(context_handle[1], 0, &ibs_request, 1); } @@ -647,7 +647,7 @@ amdgpu_sync_dependency_test(amdgpu_device_handle device_handle, bool user_queue) if (user_queue) { ring_context->pm4_dw = ib_info.size; amdgpu_user_queue_submit(device_handle, ring_context, ip_block->type, - ib_info.ib_mc_address); + ib_info.ib_mc_address, false); } else { r = amdgpu_cs_submit(context_handle[0], 0, &ibs_request, 1); igt_assert_eq(r, 0); diff --git a/tests/amdgpu/amd_cs_nop.c b/tests/amdgpu/amd_cs_nop.c index 268bc9201..658c8d050 100644 --- a/tests/amdgpu/amd_cs_nop.c +++ b/tests/amdgpu/amd_cs_nop.c @@ -108,7 +108,7 @@ static void nop_cs(amdgpu_device_handle device, if (user_queue) { ring_context->pm4_dw = ib_info.size; amdgpu_user_queue_submit(device, ring_context, ip_type, - ib_info.ib_mc_address); + ib_info.ib_mc_address, false); igt_assert_eq(r, 0); } else { r = amdgpu_cs_submit(context, 0, &ibs_request, 1); -- 2.43.0 </div> </body> </html>