From: Jesse Zhang <Jesse.Zhang@amd.com>
To: <amd-gfx@lists.freedesktop.org>
Cc: <Alexander.Deucher@amd.com>,
Christian Koenig <christian.koenig@amd.com>,
Jesse.zhang <Jesse.zhang@amd.com>,
Jesse Zhang <Jesse.Zhang@amd.com>
Subject: [PATCH v4 01/10] drm/amdgpu/mes: add NOTIFY_WORK_ON_UNMAPPED_QUEUE op + ADD_QUEUE fields
Date: Fri, 1 May 2026 00:03:29 +0800 [thread overview]
Message-ID: <20260430161146.2851078-1-Jesse.Zhang@amd.com> (raw)
From: "Jesse.zhang" <Jesse.zhang@amd.com>
Kernel-side abstraction work for the SDMA usermode-queue plumbing
landed in subsequent per-engine patches:
- mes_add_queue_input gains is_user_mode_submission and
unmap_flag_addr. Without is_user_mode_submission MES treats SDMA
queues as kernel-managed and uses the end-of-MQD slot for the unmap
flag, so PROTECTED_FENCE at the tail of every SDMA IB looks like a
"queue done" signal and MES gangs the queue out forever.
- mes_misc_opcode gains MES_MISC_OP_NOTIFY_WORK_ON_UNMAPPED_QUEUE
with a notify_work.priority_level payload. This wakes a gangs-out
SDMA UMQ so subsequent IBs get re-mapped (SDMA has no
CP_UNMAPPED_DOORBELL HW intercept).
Also surface the matching firmware bits in mes_v12_api_def.h:
is_user_mode_submission / enable_perf_profiling /
exclude_process_limit / is_video_blit_queue bitfields in
MESAPI__ADD_QUEUE, and the unmap_flag_addr packet field.
Signed-off-by: Jesse Zhang <Jesse.Zhang@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h | 7 +++++++
drivers/gpu/drm/amd/include/mes_v12_api_def.h | 12 +++++++++++-
2 files changed, 18 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
index cafc5caae822..705056de94b0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
@@ -265,6 +265,8 @@ struct mes_add_queue_input {
uint32_t exclusively_scheduled;
uint32_t sh_mem_config_data;
uint32_t vm_cntx_cntl;
+ uint32_t is_user_mode_submission;
+ uint64_t unmap_flag_addr;
};
struct mes_remove_queue_input {
@@ -343,6 +345,7 @@ enum mes_misc_opcode {
MES_MISC_OP_WRM_REG_WR_WAIT,
MES_MISC_OP_SET_SHADER_DEBUGGER,
MES_MISC_OP_CHANGE_CONFIG,
+ MES_MISC_OP_NOTIFY_WORK_ON_UNMAPPED_QUEUE,
};
struct mes_misc_op_input {
@@ -397,6 +400,10 @@ struct mes_misc_op_input {
uint32_t tdr_delay;
} tdr_config;
} change_config;
+
+ struct {
+ uint32_t priority_level;
+ } notify_work;
};
};
diff --git a/drivers/gpu/drm/amd/include/mes_v12_api_def.h b/drivers/gpu/drm/amd/include/mes_v12_api_def.h
index e541a43714a1..cd6e60184a06 100644
--- a/drivers/gpu/drm/amd/include/mes_v12_api_def.h
+++ b/drivers/gpu/drm/amd/include/mes_v12_api_def.h
@@ -381,7 +381,11 @@ union MESAPI__ADD_QUEUE {
uint32_t exclusively_scheduled : 1;
uint32_t is_long_running : 1;
uint32_t is_dwm_queue : 1;
- uint32_t reserved : 15;
+ uint32_t is_video_blit_queue : 1;
+ uint32_t is_user_mode_submission : 1;
+ uint32_t enable_perf_profiling : 1;
+ uint32_t exclude_process_limit : 1;
+ uint32_t reserved : 11;
};
struct MES_API_STATUS api_status;
uint64_t tma_addr;
@@ -393,6 +397,12 @@ union MESAPI__ADD_QUEUE {
uint32_t queue_id;
uint32_t alignment_mode_setting;
uint32_t full_sh_mem_config_data;
+ /*
+ * MC addr where MES writes 1 when it unmaps the queue. Used
+ * by user-mode SDMA UMQs so the kernel/userspace can detect
+ * the unmapped state and re-arm work via NOTIFY_WORK_ON_UNMAPPED_QUEUE.
+ */
+ uint64_t unmap_flag_addr;
};
uint32_t max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS];
--
2.49.0
next reply other threads:[~2026-04-30 16:12 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-04-30 16:03 Jesse Zhang [this message]
2026-04-30 16:03 ` [PATCH v4 02/10] drm/amdgpu/mes11: plumb unmap_flag_addr + NOTIFY_WORK_ON_UNMAPPED_QUEUE Jesse Zhang
2026-04-30 16:03 ` [PATCH v4 03/10] drm/amdgpu/mes12: plumb is_user_mode_submission, unmap_flag_addr, NOTIFY Jesse Zhang
2026-04-30 16:03 ` [PATCH v4 04/10] drm/amdgpu/mes_userqueue: mark SDMA UMQs as user-mode submission Jesse Zhang
2026-04-30 16:03 ` [PATCH v4 05/10] amdgpu: add global aggregated doorbell bo Jesse Zhang
2026-04-30 16:03 ` [PATCH v4 06/10] drm/amdgpu: add AMDGPU_INFO_DOORBELL Jesse Zhang
2026-05-01 13:37 ` Alex Deucher
2026-04-30 16:03 ` [PATCH v4 07/10] drm/amdgpu: Add AMDGPU_GEM_OP_OPEN_GLOBAL Jesse Zhang
2026-04-30 16:03 ` [PATCH v4 08/10] drm/amdgpu/mes: route NORMAL aggregated doorbell through global agdb_bo Jesse Zhang
2026-04-30 16:03 ` [PATCH v4 09/10] drm/amdgpu/userq: report SDMA UMQ doorbell info via AMDGPU_INFO_DOORBELL Jesse Zhang
2026-05-01 13:35 ` Alex Deucher
2026-04-30 16:03 ` [PATCH v4 10/10] drm/amdgpu/userq_fence: NOTIFY MES on SDMA UMQ submit Jesse Zhang
2026-05-01 13:30 ` Alex Deucher
2026-05-04 9:01 ` Christian König
2026-05-06 6:08 ` Zhang, Jesse(Jie)
2026-05-06 7:21 ` Christian König
2026-05-01 13:26 ` [PATCH v4 01/10] drm/amdgpu/mes: add NOTIFY_WORK_ON_UNMAPPED_QUEUE op + ADD_QUEUE fields Alex Deucher
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260430161146.2851078-1-Jesse.Zhang@amd.com \
--to=jesse.zhang@amd.com \
--cc=Alexander.Deucher@amd.com \
--cc=amd-gfx@lists.freedesktop.org \
--cc=christian.koenig@amd.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox