All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH 1/2] drm/amdgpu: Add show_fdinfo() interface
@ 2021-04-07 13:31 Roy Sun
  2021-04-07 13:31 ` [PATCH 2/2] drm/scheduler: Change scheduled fence track Roy Sun
  2021-04-07 14:20 ` [PATCH 1/2] drm/amdgpu: Add show_fdinfo() interface Christian König
  0 siblings, 2 replies; 6+ messages in thread
From: Roy Sun @ 2021-04-07 13:31 UTC (permalink / raw)
  To: amd-gfx; +Cc: Alexander.Deucher, Roy Sun, David M Nieto

Tracking devices, process info and fence info using
/proc/pid/fdinfo

Signed-off-by: David M Nieto <David.Nieto@amd.com>
Signed-off-by: Roy Sun <Roy.Sun@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/Makefile        |   2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |   1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c    |   5 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c | 247 +++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h |  51 +++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c    |   1 +
 6 files changed, 306 insertions(+), 1 deletion(-)
 create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c
 create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h

diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
index ee85e8aba636..d216b7ecb5d1 100644
--- a/drivers/gpu/drm/amd/amdgpu/Makefile
+++ b/drivers/gpu/drm/amd/amdgpu/Makefile
@@ -58,6 +58,8 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
 	amdgpu_umc.o smu_v11_0_i2c.o amdgpu_fru_eeprom.o amdgpu_rap.o \
 	amdgpu_fw_attestation.o amdgpu_securedisplay.o
 
+amdgpu-$(CONFIG_PROC_FS) += amdgpu_fdinfo.o
+
 amdgpu-$(CONFIG_PERF_EVENTS) += amdgpu_pmu.o
 
 # add asic specific block
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 616c85a01299..c2338a0dd1f0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -107,6 +107,7 @@
 #include "amdgpu_gfxhub.h"
 #include "amdgpu_df.h"
 #include "amdgpu_smuio.h"
+#include "amdgpu_fdinfo.h"
 
 #define MAX_GPU_INSTANCE		16
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index 4bcc03c4c6c5..07aed377dec8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -42,7 +42,7 @@
 #include "amdgpu_irq.h"
 #include "amdgpu_dma_buf.h"
 #include "amdgpu_sched.h"
-
+#include "amdgpu_fdinfo.h"
 #include "amdgpu_amdkfd.h"
 
 #include "amdgpu_ras.h"
@@ -1691,6 +1691,9 @@ static const struct file_operations amdgpu_driver_kms_fops = {
 #ifdef CONFIG_COMPAT
 	.compat_ioctl = amdgpu_kms_compat_ioctl,
 #endif
+#ifdef CONFIG_PROC_FS
+	.show_fdinfo = amdgpu_show_fdinfo
+#endif
 };
 
 int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c
new file mode 100644
index 000000000000..72e61a89c0ea
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c
@@ -0,0 +1,247 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: David Nieto
+ *          Roy Sun
+ */
+
+#include <linux/debugfs.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/uaccess.h>
+#include <linux/reboot.h>
+#include <linux/syscalls.h>
+
+#include <drm/amdgpu_drm.h>
+#include <drm/drm_debugfs.h>
+
+#include "amdgpu.h"
+#include "amdgpu_fdinfo.h"
+
+
+static const char *amdgpu_ip_name[AMDGPU_HW_IP_NUM] = {
+	[AMDGPU_HW_IP_GFX]	=	"gfx",
+	[AMDGPU_HW_IP_COMPUTE]	=	"compute",
+	[AMDGPU_HW_IP_DMA]	=	"dma",
+	[AMDGPU_HW_IP_UVD]	=	"dec",
+	[AMDGPU_HW_IP_VCE]	=	"enc",
+	[AMDGPU_HW_IP_UVD_ENC]	=	"enc_1",
+	[AMDGPU_HW_IP_VCN_DEC]	=	"dec",
+	[AMDGPU_HW_IP_VCN_ENC]	=	"enc",
+	[AMDGPU_HW_IP_VCN_JPEG]	=	"jpeg",
+};
+
+uint64_t amdgpu_get_proc_mem(struct drm_file *file, struct amdgpu_fpriv *fpriv)
+{
+	int id;
+	struct drm_gem_object *gobj;
+	uint64_t total = 0;
+
+	spin_lock(&file->table_lock);
+	idr_for_each_entry(&file->object_idr, gobj, id) {
+		struct amdgpu_bo *bo = gem_to_amdgpu_bo(gobj);
+		unsigned int domain = amdgpu_mem_type_to_domain(
+			bo->tbo.mem.mem_type);
+
+		if (domain == AMDGPU_GEM_DOMAIN_VRAM)
+			total += amdgpu_bo_size(bo);
+	}
+	spin_unlock(&file->table_lock);
+
+	if (fpriv->vm.process_info) {
+		struct kgd_mem *mem;
+
+		mutex_lock(&fpriv->vm.process_info->lock);
+		list_for_each_entry(mem, &fpriv->vm.process_info->kfd_bo_list,
+			validate_list.head) {
+			struct amdgpu_bo *bo = mem->bo;
+			unsigned int domain = amdgpu_mem_type_to_domain(
+				bo->tbo.mem.mem_type);
+
+			if (domain == AMDGPU_GEM_DOMAIN_VRAM)
+				total += amdgpu_bo_size(bo);
+			}
+
+		list_for_each_entry(mem, &fpriv->vm.process_info->userptr_valid_list,
+			validate_list.head) {
+			struct amdgpu_bo *bo = mem->bo;
+			unsigned int domain = amdgpu_mem_type_to_domain(
+				bo->tbo.mem.mem_type);
+
+			if (domain == AMDGPU_GEM_DOMAIN_VRAM)
+				total += amdgpu_bo_size(bo);
+		}
+
+		mutex_unlock(&fpriv->vm.process_info->lock);
+	}
+
+	return total;
+}
+
+uint64_t amdgpu_get_fence_usage(struct amdgpu_fpriv *fpriv, uint32_t hwip,
+		uint32_t idx, uint64_t *elapsed)
+{
+	struct amdgpu_ctx_entity *centity;
+	struct idr *idp;
+	struct amdgpu_ctx *ctx;
+	uint32_t id, i;
+	uint64_t now, t1, t2;
+	uint64_t total = 0, min = 0;
+
+
+	if (idx >= AMDGPU_MAX_ENTITY_NUM)
+		return 0;
+
+	idp = &fpriv->ctx_mgr.ctx_handles;
+
+	mutex_lock(&fpriv->ctx_mgr.lock);
+	idr_for_each_entry(idp, ctx, id) {
+		if (!ctx->entities[hwip][idx])
+			continue;
+
+		centity = ctx->entities[hwip][idx];
+
+		for (i = 0; i < amdgpu_sched_jobs; i++) {
+			struct dma_fence *fence;
+			struct drm_sched_fence *s_fence;
+
+			spin_lock(&ctx->ring_lock);
+			fence = dma_fence_get(centity->fences[i]);
+			spin_unlock(&ctx->ring_lock);
+			if (!fence)
+				continue;
+			s_fence = to_drm_sched_fence(fence);
+			if (!dma_fence_is_signaled(&s_fence->scheduled))
+				continue;
+			now = ktime_to_ns(ktime_get());
+			t1 = ktime_to_ns(s_fence->scheduled.timestamp);
+			t2 = !dma_fence_is_signaled(&s_fence->finished) ?
+				0 : ktime_to_ns(s_fence->finished.timestamp);
+			dma_fence_put(fence);
+
+			t1 = now - t1;
+			t2 = (t2 == 0) ? 0 : now - t2;
+
+			total += t1 - t2;
+			if (t1 > min)
+				min = t1;
+		}
+
+	}
+
+	mutex_unlock(&fpriv->ctx_mgr.lock);
+
+	if (elapsed)
+		*elapsed = min;
+
+	return total;
+}
+
+uint32_t amdgpu_get_ip_count(struct amdgpu_device *adev, int id)
+{
+	enum amd_ip_block_type type;
+	uint32_t count = 0;
+	int i;
+
+	switch (id) {
+	case AMDGPU_HW_IP_GFX:
+		type = AMD_IP_BLOCK_TYPE_GFX;
+		break;
+	case AMDGPU_HW_IP_COMPUTE:
+		type = AMD_IP_BLOCK_TYPE_GFX;
+		break;
+	case AMDGPU_HW_IP_DMA:
+		type = AMD_IP_BLOCK_TYPE_SDMA;
+		break;
+	case AMDGPU_HW_IP_UVD:
+		type = AMD_IP_BLOCK_TYPE_UVD;
+		break;
+	case AMDGPU_HW_IP_VCE:
+		type = AMD_IP_BLOCK_TYPE_VCE;
+		break;
+	case AMDGPU_HW_IP_UVD_ENC:
+		type = AMD_IP_BLOCK_TYPE_UVD;
+		break;
+	case AMDGPU_HW_IP_VCN_DEC:
+	case AMDGPU_HW_IP_VCN_ENC:
+		type = AMD_IP_BLOCK_TYPE_VCN;
+		break;
+	case AMDGPU_HW_IP_VCN_JPEG:
+		type = (amdgpu_device_ip_get_ip_block(adev,
+			AMD_IP_BLOCK_TYPE_JPEG)) ?
+			AMD_IP_BLOCK_TYPE_JPEG : AMD_IP_BLOCK_TYPE_VCN;
+		break;
+	default:
+		return 0;
+	}
+
+	for (i = 0; i < adev->num_ip_blocks; i++)
+		if (adev->ip_blocks[i].version->type == type &&
+		    adev->ip_blocks[i].status.valid &&
+		    count < AMDGPU_HW_IP_INSTANCE_MAX_COUNT)
+			count++;
+	return count;
+
+}
+
+void amdgpu_show_fdinfo(struct seq_file *m, struct file *f)
+{
+	struct amdgpu_fpriv *fpriv;
+	uint32_t bus, dev, fn, i;
+
+	if (amdgpu_file_to_fpriv(f, &fpriv))
+		return;
+	struct drm_file *file = f->private_data;
+	struct amdgpu_device *adev = drm_to_adev(file->minor->dev);
+
+	bus = adev->pdev->bus->number;
+	dev = PCI_SLOT(adev->pdev->devfn);
+	fn = PCI_FUNC(adev->pdev->devfn);
+	seq_printf(m, "pdev:\t%02x:%02x.%d\npasid:\t%u\n", bus, dev, fn,
+			fpriv->vm.pasid);
+
+	seq_printf(m, "mem:\t%llu kB\n", amdgpu_get_proc_mem(file, fpriv)/1024UL);
+
+	for (i = 0; i < AMDGPU_HW_IP_NUM; i++) {
+		uint32_t enabled = amdgpu_get_ip_count(adev, i);
+		uint32_t count = amdgpu_ctx_num_entities[i];
+		int idx = 0;
+		uint64_t total = 0, min = 0;
+		uint32_t perc, frac;
+
+		if (enabled) {
+			for (idx = 0; idx < count; idx++) {
+				total = amdgpu_get_fence_usage(fpriv,
+					i, idx, &min);
+
+				if ((total == 0) || (min == 0))
+					continue;
+
+				perc = div64_u64(10000 * total, min);
+				frac = perc % 100;
+
+				seq_printf(m, "%s%d:\t%d.%d%%\n",
+						amdgpu_ip_name[i],
+						idx, perc/100, frac);
+			}
+		}
+	}
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h
new file mode 100644
index 000000000000..4d73a820f7e7
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h
@@ -0,0 +1,51 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: David Nieto
+ *          Roy Sun
+ */
+#ifndef __AMDGPU_SMI_H__
+#define __AMDGPU_SMI_H__
+
+#include <linux/idr.h>
+#include <linux/kfifo.h>
+#include <linux/rbtree.h>
+#include <drm/gpu_scheduler.h>
+#include <drm/drm_file.h>
+#include <drm/ttm/ttm_bo_driver.h>
+#include <linux/sched/mm.h>
+
+#include "amdgpu_sync.h"
+#include "amdgpu_ring.h"
+#include "amdgpu_ids.h"
+
+struct amdgpu_proc;
+struct amdgpu_ctx;
+uint32_t amdgpu_get_ip_count(struct amdgpu_device *adev, int id);
+
+uint64_t amdgpu_get_fence_usage(struct amdgpu_fpriv *fpriv, uint32_t hwip,
+		uint32_t idx, uint64_t *elapsed);
+
+uint64_t amdgpu_get_proc_mem(struct drm_file *file, struct amdgpu_fpriv *fpriv);
+
+void amdgpu_show_fdinfo(struct seq_file *m, struct file *f);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 39ee88d29cca..b2e774aeab45 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -41,6 +41,7 @@
 #include "amdgpu_gem.h"
 #include "amdgpu_display.h"
 #include "amdgpu_ras.h"
+#include "amdgpu_fdinfo.h"
 
 void amdgpu_unregister_gpu_instance(struct amdgpu_device *adev)
 {
-- 
2.31.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply related	[flat|nested] 6+ messages in thread

* [PATCH 2/2] drm/scheduler: Change scheduled fence track
  2021-04-07 13:31 [PATCH 1/2] drm/amdgpu: Add show_fdinfo() interface Roy Sun
@ 2021-04-07 13:31 ` Roy Sun
  2021-04-07 14:04   ` Christian König
  2021-04-07 15:44     ` Alex Deucher
  2021-04-07 14:20 ` [PATCH 1/2] drm/amdgpu: Add show_fdinfo() interface Christian König
  1 sibling, 2 replies; 6+ messages in thread
From: Roy Sun @ 2021-04-07 13:31 UTC (permalink / raw)
  To: amd-gfx; +Cc: Alexander.Deucher, Roy Sun, David M Nieto

Update the timestamp of scheduled fence on HW
completion of the previous fences

This allow more accurate tracking of the fence
execution in HW

Signed-off-by: David M Nieto <david.nieto@amd.com>
Signed-off-by: Roy Sun <Roy.Sun@amd.com>
---
 drivers/gpu/drm/scheduler/sched_main.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c
index 92d8de24d0a1..4e5d8d4af010 100644
--- a/drivers/gpu/drm/scheduler/sched_main.c
+++ b/drivers/gpu/drm/scheduler/sched_main.c
@@ -515,7 +515,7 @@ void drm_sched_resubmit_jobs(struct drm_gpu_scheduler *sched)
 EXPORT_SYMBOL(drm_sched_resubmit_jobs);
 
 /**
- * drm_sched_resubmit_jobs_ext - helper to relunch certain number of jobs from mirror ring list
+ * drm_sched_resubmit_jobs_ext - helper to relunch certain number of jobs from pending list
  *
  * @sched: scheduler instance
  * @max: job numbers to relaunch
@@ -671,7 +671,7 @@ drm_sched_select_entity(struct drm_gpu_scheduler *sched)
 static struct drm_sched_job *
 drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched)
 {
-	struct drm_sched_job *job;
+	struct drm_sched_job *job, *next;
 
 	/*
 	 * Don't destroy jobs while the timeout worker is running  OR thread
@@ -690,6 +690,13 @@ drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched)
 	if (job && dma_fence_is_signaled(&job->s_fence->finished)) {
 		/* remove job from pending_list */
 		list_del_init(&job->list);
+		/* account for the next fence in the queue */
+		next = list_first_entry_or_null(&sched->pending_list,
+				struct drm_sched_job, list);
+		if (next) {
+			next->s_fence->scheduled.timestamp =
+				job->s_fence->finished.timestamp;
+		}
 	} else {
 		job = NULL;
 		/* queue timeout for next job */
-- 
2.31.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply related	[flat|nested] 6+ messages in thread

* Re: [PATCH 2/2] drm/scheduler: Change scheduled fence track
  2021-04-07 13:31 ` [PATCH 2/2] drm/scheduler: Change scheduled fence track Roy Sun
@ 2021-04-07 14:04   ` Christian König
  2021-04-07 15:44     ` Alex Deucher
  1 sibling, 0 replies; 6+ messages in thread
From: Christian König @ 2021-04-07 14:04 UTC (permalink / raw)
  To: Roy Sun, amd-gfx; +Cc: Alexander.Deucher, David M Nieto



Am 07.04.21 um 15:31 schrieb Roy Sun:
> Update the timestamp of scheduled fence on HW
> completion of the previous fences
>
> This allow more accurate tracking of the fence
> execution in HW
>
> Signed-off-by: David M Nieto <david.nieto@amd.com>
> Signed-off-by: Roy Sun <Roy.Sun@amd.com>
> ---
>   drivers/gpu/drm/scheduler/sched_main.c | 11 +++++++++--
>   1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c
> index 92d8de24d0a1..4e5d8d4af010 100644
> --- a/drivers/gpu/drm/scheduler/sched_main.c
> +++ b/drivers/gpu/drm/scheduler/sched_main.c
> @@ -515,7 +515,7 @@ void drm_sched_resubmit_jobs(struct drm_gpu_scheduler *sched)
>   EXPORT_SYMBOL(drm_sched_resubmit_jobs);
>   
>   /**
> - * drm_sched_resubmit_jobs_ext - helper to relunch certain number of jobs from mirror ring list
> + * drm_sched_resubmit_jobs_ext - helper to relunch certain number of jobs from pending list

s/relunch/relaunch/

With that nit pick fixed the patch is Reviewed-by: Christian König 
<christian.koenig@amd.com>

>    *
>    * @sched: scheduler instance
>    * @max: job numbers to relaunch
> @@ -671,7 +671,7 @@ drm_sched_select_entity(struct drm_gpu_scheduler *sched)
>   static struct drm_sched_job *
>   drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched)
>   {
> -	struct drm_sched_job *job;
> +	struct drm_sched_job *job, *next;
>   
>   	/*
>   	 * Don't destroy jobs while the timeout worker is running  OR thread
> @@ -690,6 +690,13 @@ drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched)
>   	if (job && dma_fence_is_signaled(&job->s_fence->finished)) {
>   		/* remove job from pending_list */
>   		list_del_init(&job->list);
> +		/* account for the next fence in the queue */
> +		next = list_first_entry_or_null(&sched->pending_list,
> +				struct drm_sched_job, list);
> +		if (next) {
> +			next->s_fence->scheduled.timestamp =
> +				job->s_fence->finished.timestamp;
> +		}
>   	} else {
>   		job = NULL;
>   		/* queue timeout for next job */

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH 1/2] drm/amdgpu: Add show_fdinfo() interface
  2021-04-07 13:31 [PATCH 1/2] drm/amdgpu: Add show_fdinfo() interface Roy Sun
  2021-04-07 13:31 ` [PATCH 2/2] drm/scheduler: Change scheduled fence track Roy Sun
@ 2021-04-07 14:20 ` Christian König
  1 sibling, 0 replies; 6+ messages in thread
From: Christian König @ 2021-04-07 14:20 UTC (permalink / raw)
  To: Roy Sun, amd-gfx; +Cc: Alexander.Deucher, David M Nieto

Am 07.04.21 um 15:31 schrieb Roy Sun:
> Tracking devices, process info and fence info using
> /proc/pid/fdinfo
>
> Signed-off-by: David M Nieto <David.Nieto@amd.com>
> Signed-off-by: Roy Sun <Roy.Sun@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/Makefile        |   2 +
>   drivers/gpu/drm/amd/amdgpu/amdgpu.h        |   1 +
>   drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c    |   5 +-
>   drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c | 247 +++++++++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h |  51 +++++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c    |   1 +
>   6 files changed, 306 insertions(+), 1 deletion(-)
>   create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c
>   create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
> index ee85e8aba636..d216b7ecb5d1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/Makefile
> +++ b/drivers/gpu/drm/amd/amdgpu/Makefile
> @@ -58,6 +58,8 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
>   	amdgpu_umc.o smu_v11_0_i2c.o amdgpu_fru_eeprom.o amdgpu_rap.o \
>   	amdgpu_fw_attestation.o amdgpu_securedisplay.o
>   
> +amdgpu-$(CONFIG_PROC_FS) += amdgpu_fdinfo.o
> +
>   amdgpu-$(CONFIG_PERF_EVENTS) += amdgpu_pmu.o
>   
>   # add asic specific block
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> index 616c85a01299..c2338a0dd1f0 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> @@ -107,6 +107,7 @@
>   #include "amdgpu_gfxhub.h"
>   #include "amdgpu_df.h"
>   #include "amdgpu_smuio.h"
> +#include "amdgpu_fdinfo.h"
>   
>   #define MAX_GPU_INSTANCE		16
>   
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> index 4bcc03c4c6c5..07aed377dec8 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> @@ -42,7 +42,7 @@
>   #include "amdgpu_irq.h"
>   #include "amdgpu_dma_buf.h"
>   #include "amdgpu_sched.h"
> -
> +#include "amdgpu_fdinfo.h"
>   #include "amdgpu_amdkfd.h"
>   
>   #include "amdgpu_ras.h"
> @@ -1691,6 +1691,9 @@ static const struct file_operations amdgpu_driver_kms_fops = {
>   #ifdef CONFIG_COMPAT
>   	.compat_ioctl = amdgpu_kms_compat_ioctl,
>   #endif
> +#ifdef CONFIG_PROC_FS
> +	.show_fdinfo = amdgpu_show_fdinfo
> +#endif
>   };
>   
>   int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv)
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c
> new file mode 100644
> index 000000000000..72e61a89c0ea
> --- /dev/null
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c
> @@ -0,0 +1,247 @@

Still no SPDX license tag.

> +/*
> + * Copyright 2021 Advanced Micro Devices, Inc.
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the "Software"),
> + * to deal in the Software without restriction, including without limitation
> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> + * and/or sell copies of the Software, and to permit persons to whom the
> + * Software is furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice shall be included in
> + * all copies or substantial portions of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
> + * OTHER DEALINGS IN THE SOFTWARE.
> + *
> + * Authors: David Nieto
> + *          Roy Sun
> + */
> +
> +#include <linux/debugfs.h>
> +#include <linux/list.h>
> +#include <linux/module.h>
> +#include <linux/uaccess.h>
> +#include <linux/reboot.h>
> +#include <linux/syscalls.h>
> +
> +#include <drm/amdgpu_drm.h>
> +#include <drm/drm_debugfs.h>
> +
> +#include "amdgpu.h"
> +#include "amdgpu_fdinfo.h"
> +
> +
> +static const char *amdgpu_ip_name[AMDGPU_HW_IP_NUM] = {
> +	[AMDGPU_HW_IP_GFX]	=	"gfx",
> +	[AMDGPU_HW_IP_COMPUTE]	=	"compute",
> +	[AMDGPU_HW_IP_DMA]	=	"dma",
> +	[AMDGPU_HW_IP_UVD]	=	"dec",
> +	[AMDGPU_HW_IP_VCE]	=	"enc",
> +	[AMDGPU_HW_IP_UVD_ENC]	=	"enc_1",
> +	[AMDGPU_HW_IP_VCN_DEC]	=	"dec",
> +	[AMDGPU_HW_IP_VCN_ENC]	=	"enc",
> +	[AMDGPU_HW_IP_VCN_JPEG]	=	"jpeg",
> +};
> +
> +uint64_t amdgpu_get_proc_mem(struct drm_file *file, struct amdgpu_fpriv *fpriv)
> +{
> +	int id;
> +	struct drm_gem_object *gobj;
> +	uint64_t total = 0;
> +
> +	spin_lock(&file->table_lock);
> +	idr_for_each_entry(&file->object_idr, gobj, id) {
> +		struct amdgpu_bo *bo = gem_to_amdgpu_bo(gobj);
> +		unsigned int domain = amdgpu_mem_type_to_domain(
> +			bo->tbo.mem.mem_type);
> +
> +		if (domain == AMDGPU_GEM_DOMAIN_VRAM)
> +			total += amdgpu_bo_size(bo);

Why only VRAM memory?

> +	}
> +	spin_unlock(&file->table_lock);

This should probably be as helper in the amdgpu_gem.c file.

> +
> +	if (fpriv->vm.process_info) {
> +		struct kgd_mem *mem;
> +
> +		mutex_lock(&fpriv->vm.process_info->lock);
> +		list_for_each_entry(mem, &fpriv->vm.process_info->kfd_bo_list,
> +			validate_list.head) {
> +			struct amdgpu_bo *bo = mem->bo;
> +			unsigned int domain = amdgpu_mem_type_to_domain(
> +				bo->tbo.mem.mem_type);
> +
> +			if (domain == AMDGPU_GEM_DOMAIN_VRAM)
> +				total += amdgpu_bo_size(bo);
> +			}
> +
> +		list_for_each_entry(mem, &fpriv->vm.process_info->userptr_valid_list,
> +			validate_list.head) {
> +			struct amdgpu_bo *bo = mem->bo;
> +			unsigned int domain = amdgpu_mem_type_to_domain(
> +				bo->tbo.mem.mem_type);
> +
> +			if (domain == AMDGPU_GEM_DOMAIN_VRAM)
> +				total += amdgpu_bo_size(bo);
> +		}
> +
> +		mutex_unlock(&fpriv->vm.process_info->lock);

This needs to be moved into the VM or even better the 
amdkfd_process_info code.

> +	}
> +
> +	return total;
> +}
> +
> +uint64_t amdgpu_get_fence_usage(struct amdgpu_fpriv *fpriv, uint32_t hwip,
> +		uint32_t idx, uint64_t *elapsed)
> +{
> +	struct amdgpu_ctx_entity *centity;
> +	struct idr *idp;
> +	struct amdgpu_ctx *ctx;
> +	uint32_t id, i;
> +	uint64_t now, t1, t2;
> +	uint64_t total = 0, min = 0;
> +
> +
> +	if (idx >= AMDGPU_MAX_ENTITY_NUM)
> +		return 0;
> +
> +	idp = &fpriv->ctx_mgr.ctx_handles;
> +
> +	mutex_lock(&fpriv->ctx_mgr.lock);
> +	idr_for_each_entry(idp, ctx, id) {
> +		if (!ctx->entities[hwip][idx])
> +			continue;
> +
> +		centity = ctx->entities[hwip][idx];
> +
> +		for (i = 0; i < amdgpu_sched_jobs; i++) {
> +			struct dma_fence *fence;
> +			struct drm_sched_fence *s_fence;
> +
> +			spin_lock(&ctx->ring_lock);
> +			fence = dma_fence_get(centity->fences[i]);
> +			spin_unlock(&ctx->ring_lock);

That is quite racy.

> +			if (!fence)
> +				continue;
> +			s_fence = to_drm_sched_fence(fence);
> +			if (!dma_fence_is_signaled(&s_fence->scheduled))
> +				continue;
> +			now = ktime_to_ns(ktime_get());
> +			t1 = ktime_to_ns(s_fence->scheduled.timestamp);
> +			t2 = !dma_fence_is_signaled(&s_fence->finished) ?
> +				0 : ktime_to_ns(s_fence->finished.timestamp);
> +			dma_fence_put(fence);
> +
> +			t1 = now - t1;
> +			t2 = (t2 == 0) ? 0 : now - t2;
> +
> +			total += t1 - t2;
> +			if (t1 > min)
> +				min = t1;

Please use ktime for the calculation here.

> +		}
> +
> +	}
> +
> +	mutex_unlock(&fpriv->ctx_mgr.lock);

Again that needs to be in the ctx handling code in general.

> +
> +	if (elapsed)
> +		*elapsed = min;
> +
> +	return total;
> +}
> +
> +uint32_t amdgpu_get_ip_count(struct amdgpu_device *adev, int id)
> +{
> +	enum amd_ip_block_type type;
> +	uint32_t count = 0;
> +	int i;
> +
> +	switch (id) {
> +	case AMDGPU_HW_IP_GFX:
> +		type = AMD_IP_BLOCK_TYPE_GFX;
> +		break;
> +	case AMDGPU_HW_IP_COMPUTE:
> +		type = AMD_IP_BLOCK_TYPE_GFX;
> +		break;
> +	case AMDGPU_HW_IP_DMA:
> +		type = AMD_IP_BLOCK_TYPE_SDMA;
> +		break;
> +	case AMDGPU_HW_IP_UVD:
> +		type = AMD_IP_BLOCK_TYPE_UVD;
> +		break;
> +	case AMDGPU_HW_IP_VCE:
> +		type = AMD_IP_BLOCK_TYPE_VCE;
> +		break;
> +	case AMDGPU_HW_IP_UVD_ENC:
> +		type = AMD_IP_BLOCK_TYPE_UVD;
> +		break;
> +	case AMDGPU_HW_IP_VCN_DEC:
> +	case AMDGPU_HW_IP_VCN_ENC:
> +		type = AMD_IP_BLOCK_TYPE_VCN;
> +		break;
> +	case AMDGPU_HW_IP_VCN_JPEG:
> +		type = (amdgpu_device_ip_get_ip_block(adev,
> +			AMD_IP_BLOCK_TYPE_JPEG)) ?
> +			AMD_IP_BLOCK_TYPE_JPEG : AMD_IP_BLOCK_TYPE_VCN;
> +		break;
> +	default:
> +		return 0;
> +	}
> +
> +	for (i = 0; i < adev->num_ip_blocks; i++)
> +		if (adev->ip_blocks[i].version->type == type &&
> +		    adev->ip_blocks[i].status.valid &&
> +		    count < AMDGPU_HW_IP_INSTANCE_MAX_COUNT)
> +			count++;
> +	return count;
> +
> +}

Please just completely drop that and use AMDGPU_HW_IP_* directly.

We certainly don't want decode and encode show up as one value.

> +
> +void amdgpu_show_fdinfo(struct seq_file *m, struct file *f)
> +{
> +	struct amdgpu_fpriv *fpriv;
> +	uint32_t bus, dev, fn, i;
> +
> +	if (amdgpu_file_to_fpriv(f, &fpriv))
> +		return;
> +	struct drm_file *file = f->private_data;
> +	struct amdgpu_device *adev = drm_to_adev(file->minor->dev);
> +
> +	bus = adev->pdev->bus->number;
> +	dev = PCI_SLOT(adev->pdev->devfn);
> +	fn = PCI_FUNC(adev->pdev->devfn);
> +	seq_printf(m, "pdev:\t%02x:%02x.%d\npasid:\t%u\n", bus, dev, fn,
> +			fpriv->vm.pasid);
> +
> +	seq_printf(m, "mem:\t%llu kB\n", amdgpu_get_proc_mem(file, fpriv)/1024UL);
> +
> +	for (i = 0; i < AMDGPU_HW_IP_NUM; i++) {
> +		uint32_t enabled = amdgpu_get_ip_count(adev, i);
> +		uint32_t count = amdgpu_ctx_num_entities[i];
> +		int idx = 0;
> +		uint64_t total = 0, min = 0;
> +		uint32_t perc, frac;
> +
> +		if (enabled) {

Please rather use "if (!enabled) continue style here.

Regards,
Christian.

> +			for (idx = 0; idx < count; idx++) {
> +				total = amdgpu_get_fence_usage(fpriv,
> +					i, idx, &min);
> +
> +				if ((total == 0) || (min == 0))
> +					continue;
> +
> +				perc = div64_u64(10000 * total, min);
> +				frac = perc % 100;
> +
> +				seq_printf(m, "%s%d:\t%d.%d%%\n",
> +						amdgpu_ip_name[i],
> +						idx, perc/100, frac);
> +			}
> +		}
> +	}
> +}
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h
> new file mode 100644
> index 000000000000..4d73a820f7e7
> --- /dev/null
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h
> @@ -0,0 +1,51 @@
> +/*
> + * Copyright 2021 Advanced Micro Devices, Inc.
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the "Software"),
> + * to deal in the Software without restriction, including without limitation
> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> + * and/or sell copies of the Software, and to permit persons to whom the
> + * Software is furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice shall be included in
> + * all copies or substantial portions of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
> + * OTHER DEALINGS IN THE SOFTWARE.
> + *
> + * Authors: David Nieto
> + *          Roy Sun
> + */
> +#ifndef __AMDGPU_SMI_H__
> +#define __AMDGPU_SMI_H__
> +
> +#include <linux/idr.h>
> +#include <linux/kfifo.h>
> +#include <linux/rbtree.h>
> +#include <drm/gpu_scheduler.h>
> +#include <drm/drm_file.h>
> +#include <drm/ttm/ttm_bo_driver.h>
> +#include <linux/sched/mm.h>
> +
> +#include "amdgpu_sync.h"
> +#include "amdgpu_ring.h"
> +#include "amdgpu_ids.h"
> +
> +struct amdgpu_proc;
> +struct amdgpu_ctx;
> +uint32_t amdgpu_get_ip_count(struct amdgpu_device *adev, int id);
> +
> +uint64_t amdgpu_get_fence_usage(struct amdgpu_fpriv *fpriv, uint32_t hwip,
> +		uint32_t idx, uint64_t *elapsed);
> +
> +uint64_t amdgpu_get_proc_mem(struct drm_file *file, struct amdgpu_fpriv *fpriv);
> +
> +void amdgpu_show_fdinfo(struct seq_file *m, struct file *f);
> +
> +#endif
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> index 39ee88d29cca..b2e774aeab45 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> @@ -41,6 +41,7 @@
>   #include "amdgpu_gem.h"
>   #include "amdgpu_display.h"
>   #include "amdgpu_ras.h"
> +#include "amdgpu_fdinfo.h"
>   
>   void amdgpu_unregister_gpu_instance(struct amdgpu_device *adev)
>   {

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH 2/2] drm/scheduler: Change scheduled fence track
  2021-04-07 13:31 ` [PATCH 2/2] drm/scheduler: Change scheduled fence track Roy Sun
@ 2021-04-07 15:44     ` Alex Deucher
  2021-04-07 15:44     ` Alex Deucher
  1 sibling, 0 replies; 6+ messages in thread
From: Alex Deucher @ 2021-04-07 15:44 UTC (permalink / raw)
  To: Roy Sun, Maling list - DRI developers
  Cc: Deucher, Alexander, amd-gfx list, David M Nieto

+ dri-devel since GPU scheduler is a shared component.

On Wed, Apr 7, 2021 at 9:31 AM Roy Sun <Roy.Sun@amd.com> wrote:
>
> Update the timestamp of scheduled fence on HW
> completion of the previous fences
>
> This allow more accurate tracking of the fence
> execution in HW
>
> Signed-off-by: David M Nieto <david.nieto@amd.com>
> Signed-off-by: Roy Sun <Roy.Sun@amd.com>
> ---
>  drivers/gpu/drm/scheduler/sched_main.c | 11 +++++++++--
>  1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c
> index 92d8de24d0a1..4e5d8d4af010 100644
> --- a/drivers/gpu/drm/scheduler/sched_main.c
> +++ b/drivers/gpu/drm/scheduler/sched_main.c
> @@ -515,7 +515,7 @@ void drm_sched_resubmit_jobs(struct drm_gpu_scheduler *sched)
>  EXPORT_SYMBOL(drm_sched_resubmit_jobs);
>
>  /**
> - * drm_sched_resubmit_jobs_ext - helper to relunch certain number of jobs from mirror ring list
> + * drm_sched_resubmit_jobs_ext - helper to relunch certain number of jobs from pending list
>   *
>   * @sched: scheduler instance
>   * @max: job numbers to relaunch
> @@ -671,7 +671,7 @@ drm_sched_select_entity(struct drm_gpu_scheduler *sched)
>  static struct drm_sched_job *
>  drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched)
>  {
> -       struct drm_sched_job *job;
> +       struct drm_sched_job *job, *next;
>
>         /*
>          * Don't destroy jobs while the timeout worker is running  OR thread
> @@ -690,6 +690,13 @@ drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched)
>         if (job && dma_fence_is_signaled(&job->s_fence->finished)) {
>                 /* remove job from pending_list */
>                 list_del_init(&job->list);
> +               /* account for the next fence in the queue */
> +               next = list_first_entry_or_null(&sched->pending_list,
> +                               struct drm_sched_job, list);
> +               if (next) {
> +                       next->s_fence->scheduled.timestamp =
> +                               job->s_fence->finished.timestamp;
> +               }
>         } else {
>                 job = NULL;
>                 /* queue timeout for next job */
> --
> 2.31.1
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH 2/2] drm/scheduler: Change scheduled fence track
@ 2021-04-07 15:44     ` Alex Deucher
  0 siblings, 0 replies; 6+ messages in thread
From: Alex Deucher @ 2021-04-07 15:44 UTC (permalink / raw)
  To: Roy Sun, Maling list - DRI developers
  Cc: Deucher, Alexander, amd-gfx list, David M Nieto

+ dri-devel since GPU scheduler is a shared component.

On Wed, Apr 7, 2021 at 9:31 AM Roy Sun <Roy.Sun@amd.com> wrote:
>
> Update the timestamp of scheduled fence on HW
> completion of the previous fences
>
> This allow more accurate tracking of the fence
> execution in HW
>
> Signed-off-by: David M Nieto <david.nieto@amd.com>
> Signed-off-by: Roy Sun <Roy.Sun@amd.com>
> ---
>  drivers/gpu/drm/scheduler/sched_main.c | 11 +++++++++--
>  1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c
> index 92d8de24d0a1..4e5d8d4af010 100644
> --- a/drivers/gpu/drm/scheduler/sched_main.c
> +++ b/drivers/gpu/drm/scheduler/sched_main.c
> @@ -515,7 +515,7 @@ void drm_sched_resubmit_jobs(struct drm_gpu_scheduler *sched)
>  EXPORT_SYMBOL(drm_sched_resubmit_jobs);
>
>  /**
> - * drm_sched_resubmit_jobs_ext - helper to relunch certain number of jobs from mirror ring list
> + * drm_sched_resubmit_jobs_ext - helper to relunch certain number of jobs from pending list
>   *
>   * @sched: scheduler instance
>   * @max: job numbers to relaunch
> @@ -671,7 +671,7 @@ drm_sched_select_entity(struct drm_gpu_scheduler *sched)
>  static struct drm_sched_job *
>  drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched)
>  {
> -       struct drm_sched_job *job;
> +       struct drm_sched_job *job, *next;
>
>         /*
>          * Don't destroy jobs while the timeout worker is running  OR thread
> @@ -690,6 +690,13 @@ drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched)
>         if (job && dma_fence_is_signaled(&job->s_fence->finished)) {
>                 /* remove job from pending_list */
>                 list_del_init(&job->list);
> +               /* account for the next fence in the queue */
> +               next = list_first_entry_or_null(&sched->pending_list,
> +                               struct drm_sched_job, list);
> +               if (next) {
> +                       next->s_fence->scheduled.timestamp =
> +                               job->s_fence->finished.timestamp;
> +               }
>         } else {
>                 job = NULL;
>                 /* queue timeout for next job */
> --
> 2.31.1
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2021-04-07 15:45 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-04-07 13:31 [PATCH 1/2] drm/amdgpu: Add show_fdinfo() interface Roy Sun
2021-04-07 13:31 ` [PATCH 2/2] drm/scheduler: Change scheduled fence track Roy Sun
2021-04-07 14:04   ` Christian König
2021-04-07 15:44   ` Alex Deucher
2021-04-07 15:44     ` Alex Deucher
2021-04-07 14:20 ` [PATCH 1/2] drm/amdgpu: Add show_fdinfo() interface Christian König

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.