Skip to contentSkip to conversation

drm/amdgpu: reduce amdgpu_device_resume() time

PATCH In review S, Shirish 2 participants 3 messages May 10, 2017 – May 11, 2017 4 messages not in the archive

Subsystems (2)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): 1494407620-1217-1-git-send-email-shirish.s12-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org

S, Shirish

author
#2 lore raw

PATCH 1/1 drm/amdgpu: reduce amdgpu_device_resume() time

From: Shirish S <[email protected]>

amdgpu_device_resume() has a high time consuming call of amdgpu_late_init() which sets the clock_gating state of all IP blocks and is blocking.
This patch defers only this setting of clock gating state operation to post resume of amdgpu driver but ideally before the UI comes up or in some cases post ui as well.

With this change the resume time of amdgpu_device comes down from 1.299s to 0.199s which further helps in reducing the overall system resume time.

TEST:(For ChromiumOS on STONEY only)
* UI comes up
* S3 works multiple times
* Resume time is consistent and lower
* amdgpu_late_init() call gets called consistently and no errors reported.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  3 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 57 +++++++++++++++++++++++-------
 2 files changed, 47 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +3 -0
index 1be8aed..addb204 100644
@@ -1615,6 +1615,9 @@ struct amdgpu_device {
 	/* amdkfd interface */
 	struct kfd_dev          *kfd;
 
+	/* delayed work_func for deferring clockgating during resume */
+	struct delayed_work     late_init_work;
+
 	struct amdgpu_virt	virt;
 
 	/* link all shadow bo */
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +44 -13

113 lines not on this page · raw message

Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): BN6PR12MB18266F505C3E22384395DAD3F2EC0-/b2+HYfkarSCndJ3FSsNdQdYzm3356FpvxpqHgZTriW3zl9H0oFU5g@public.gmane.org

#4 lore raw
Quoted 13 lines · Shirish

From: Shirish S <[email protected]>

amdgpu_device_resume() has a high time consuming call of amdgpu_late_init() which sets the clock_gating state of all IP blocks and is blocking.
This patch defers only this setting of clock gating state operation to post resume of amdgpu driver but ideally before the UI comes up or in some cases post ui as well.

With this change the resume time of amdgpu_device comes down from 1.299s to 0.199s which further helps in reducing the overall system resume time.

Quoted 8 lines

TEST:(For ChromiumOS on STONEY only)
* UI comes up
* S3 works multiple times
* Resume time is consistent and lower
* amdgpu_late_init() call gets called consistently and no errors reported.

Please make sure this doesn't cause problems with S4. A few comments below.

Quoted 108 lines
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  3 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 57 +++++++++++++++++++++++-------
 2 files changed, 47 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +3 -0
index 1be8aed..addb204 100644
@@ -1615,6 +1615,9 @@ struct amdgpu_device {
        /* amdkfd interface */
        struct kfd_dev          *kfd;
 
+       /* delayed work_func for deferring clockgating during resume */
+       struct delayed_work     late_init_work;
+
        struct amdgpu_virt      virt;
 
        /* link all shadow bo */
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +33 -12
index cfc650c..a6850cf 100644
@@ -57,6 +57,8 @@
 
 MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin");
 
+#define AMDGPU_RESUME_MS               2000
+
 static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev);
 static void amdgpu_debugfs_regs_cleanup(struct amdgpu_device *adev);
 
@@ -1655,22 +1657,11 @@ static int amdgpu_init(struct amdgpu_device *adev)
        return 0;
 }
 
-static int amdgpu_late_init(struct amdgpu_device *adev)
+static int amdgpu_late_set_cg_state(struct amdgpu_device *adev)
 {
        int i = 0, r;
 
        for (i = 0; i < adev->num_ip_blocks; i++) {
-               if (!adev->ip_blocks[i].status.valid)
-                       continue;
-               if (adev->ip_blocks[i].version->funcs->late_init) {
-                       r = adev->ip_blocks[i].version->funcs->late_init((void *)adev);
-                       if (r) {
-                               DRM_ERROR("late_init of IP block <%s> failed %d\n",
-                                         adev->ip_blocks[i].version->funcs->name, r);
-                               return r;
-                       }
-                       adev->ip_blocks[i].status.late_initialized = true;
-               }
                /* skip CG for VCE/UVD, it's handled specially */
                if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
                    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE) {
@@ -1688,6 +1679,27 @@ static int amdgpu_late_init(struct amdgpu_device *adev)
        return 0;
 }
 
+static int amdgpu_late_init(struct amdgpu_device *adev)
+{
+       int i = 0, r;
+
+       for (i = 0; i < adev->num_ip_blocks; i++) {
+               if (!adev->ip_blocks[i].status.valid)
+                       continue;
+               if (adev->ip_blocks[i].version->funcs->late_init) {
+                       r = adev->ip_blocks[i].version->funcs->late_init((void *)adev);
+                       if (r) {
+                               DRM_ERROR("late_init of IP block <%s> failed %d\n",
+                                         adev->ip_blocks[i].version->funcs->name, r);
+                               return r;
+                       }
+                       adev->ip_blocks[i].status.late_initialized = true;
+               }
+       }
+
+       return 0;
+}
+
 static int amdgpu_fini(struct amdgpu_device *adev)
 {
        int i, r;
@@ -1775,6 +1787,13 @@ static int amdgpu_fini(struct amdgpu_device *adev)
        return 0;
 }
 
+static void amdgpu_late_init_func_handler(struct work_struct *work)
+{
+       struct amdgpu_device *adev =
+               container_of(work, struct amdgpu_device, late_init_work.work);
+       amdgpu_late_set_cg_state(adev);
+}
+
 int amdgpu_suspend(struct amdgpu_device *adev)
 {
        int i, r;
@@ -2074,6 +2093,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
        INIT_LIST_HEAD(&adev->gtt_list);
        spin_lock_init(&adev->gtt_list_lock);
 
+       INIT_DELAYED_WORK(&adev->late_init_work, amdgpu_late_init_func_handler);
+
        if (adev->asic_type >= CHIP_BONAIRE) {
                adev->rmmio_base = pci_resource_start(adev->pdev, 5);
                adev->rmmio_size = pci_resource_len(adev->pdev, 5);

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c int amdgpu_device_init(struct amdgpu_device *adev, · line 2267

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +5 -0
@@ -2239,6 +2260,12 @@ int amdgpu_device_init(struct amdgpu_device *adev,
                goto failed;
        }
 
+       r = amdgpu_late_set_cg_state(adev);
+       if (r) {
+               dev_err(adev->dev, "amdgpu_late_set_cg_state failed\n");
+               goto failed;
+       }

I think we can probably use the delayed work thread here as well to speed up initial driver load.

Quoted 31 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +6 -2
+
        return 0;
 
 failed:
@@ -2270,6 +2297,7 @@ void amdgpu_device_fini(struct amdgpu_device *adev)
        amdgpu_fbdev_fini(adev);
        r = amdgpu_fini(adev);
        adev->accel_working = false;
+       cancel_delayed_work_sync(&adev->late_init_work);
        /* free i2c buses */
        if (!amdgpu_device_has_dc_support(adev))
                amdgpu_i2c_fini(adev);
@@ -2449,11 +2477,14 @@ int amdgpu_device_resume(struct drm_device *dev, bool resume, bool fbcon)
                if (r)
                        DRM_ERROR("ib ring test failed (%d).\n", r);
        }
-
+       /* defer the cg state setting to post resume */
        r = amdgpu_late_init(adev);
        if (r)
                goto unlock;
 
+       mod_delayed_work(system_wq, &adev->late_init_work,
+                       msecs_to_jiffies(AMDGPU_RESUME_MS));
+
        /* pin cursors */
        list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
                struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
--

2.7.4

Quoted 4 lines

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): CADnq5_MsTXR5JvUSQ8ZxUe4hZvGb7Nm8OzUzpPJSt3iv6svFmg-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org

author
#6 lore raw

Re-spun patch with the corresponding changes.

-----Original Message-----
From: Alex Deucher [mailto:[email protected]]
Sent: Thursday, May 11, 2017 1:24 AM
To: S, Shirish <[email protected]>
Cc: [email protected]; Deucher, Alexander <[email protected]>; Huang, Ray <[email protected]>; Yu, Hui <[email protected]>
Subject: Re: FW: [PATCH] drm/amdgpu: reduce amdgpu_device_resume() time

Quoted 13 lines · Shirish

From: Shirish S <[email protected]>

amdgpu_device_resume() has a high time consuming call of amdgpu_late_init() which sets the clock_gating state of all IP blocks and is blocking.
This patch defers only this setting of clock gating state operation to post resume of amdgpu driver but ideally before the UI comes up or in some cases post ui as well.

With this change the resume time of amdgpu_device comes down from 1.299s to 0.199s which further helps in reducing the overall system resume time.

Quoted 8 lines

TEST:(For ChromiumOS on STONEY only)
* UI comes up
* S3 works multiple times
* Resume time is consistent and lower
* amdgpu_late_init() call gets called consistently and no errors reported.

Please make sure this doesn't cause problems with S4. A few comments below.

Quoted 105 lines
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  3 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 57 
+++++++++++++++++++++++-------
 2 files changed, 47 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +3 -0
index 1be8aed..addb204 100644
@@ -1615,6 +1615,9 @@ struct amdgpu_device {
        /* amdkfd interface */
        struct kfd_dev          *kfd;
 
+       /* delayed work_func for deferring clockgating during resume */
+       struct delayed_work     late_init_work;
+
        struct amdgpu_virt      virt;
 
        /* link all shadow bo */
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +3 -12
index cfc650c..a6850cf 100644
@@ -57,6 +57,8 @@
 
 MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin");
 
+#define AMDGPU_RESUME_MS               2000
+
 static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev);  
 static void amdgpu_debugfs_regs_cleanup(struct amdgpu_device *adev);
 
@@ -1655,22 +1657,11 @@ static int amdgpu_init(struct amdgpu_device *adev)
        return 0;
 }
 
-static int amdgpu_late_init(struct amdgpu_device *adev)
+static int amdgpu_late_set_cg_state(struct amdgpu_device *adev)
 {
        int i = 0, r;
 
        for (i = 0; i < adev->num_ip_blocks; i++) {
-               if (!adev->ip_blocks[i].status.valid)
-                       continue;
-               if (adev->ip_blocks[i].version->funcs->late_init) {
-                       r = adev->ip_blocks[i].version->funcs->late_init((void *)adev);
-                       if (r) {
-                               DRM_ERROR("late_init of IP block <%s> failed %d\n",
-                                         adev->ip_blocks[i].version->funcs->name, r);
-                               return r;
-                       }
-                       adev->ip_blocks[i].status.late_initialized = true;
-               }
                /* skip CG for VCE/UVD, it's handled specially */
                if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
                    adev->ip_blocks[i].version->type != 

AMD_IP_BLOCK_TYPE_VCE) { @@ -1688,6 +1679,27 @@ static int amdgpu_late_init(struct amdgpu_device *adev)

        return 0;

}

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +28 -0
+static int amdgpu_late_init(struct amdgpu_device *adev) {
+       int i = 0, r;
+
+       for (i = 0; i < adev->num_ip_blocks; i++) {
+               if (!adev->ip_blocks[i].status.valid)
+                       continue;
+               if (adev->ip_blocks[i].version->funcs->late_init) {
+                       r = adev->ip_blocks[i].version->funcs->late_init((void *)adev);
+                       if (r) {
+                               DRM_ERROR("late_init of IP block <%s> failed %d\n",
+                                         adev->ip_blocks[i].version->funcs->name, r);
+                               return r;
+                       }
+                       adev->ip_blocks[i].status.late_initialized = true;
+               }
+       }
+
+       return 0;
+}
+
 static int amdgpu_fini(struct amdgpu_device *adev)  {
        int i, r;
@@ -1775,6 +1787,13 @@ static int amdgpu_fini(struct amdgpu_device *adev)
        return 0;
 }
 
+static void amdgpu_late_init_func_handler(struct work_struct *work) {
+       struct amdgpu_device *adev =
+               container_of(work, struct amdgpu_device, late_init_work.work);
+       amdgpu_late_set_cg_state(adev); }
+
 int amdgpu_suspend(struct amdgpu_device *adev)  {
        int i, r;
@@ -2074,6 +2093,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
        INIT_LIST_HEAD(&adev->gtt_list);
        spin_lock_init(&adev->gtt_list_lock);
 
+       INIT_DELAYED_WORK(&adev->late_init_work, 
+ amdgpu_late_init_func_handler);
+
        if (adev->asic_type >= CHIP_BONAIRE) {
                adev->rmmio_base = pci_resource_start(adev->pdev, 5);

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c int amdgpu_device_init(struct amdgpu_device *adev,

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +5 -1
                adev->rmmio_size = pci_resource_len(adev->pdev, 5); @@ 
-2239,6 +2260,12 @@ int amdgpu_device_init(struct amdgpu_device *adev,
                goto failed;
        }
 
+       r = amdgpu_late_set_cg_state(adev);
+       if (r) {
+               dev_err(adev->dev, "amdgpu_late_set_cg_state failed\n");
+               goto failed;
+       }

I think we can probably use the delayed work thread here as well to speed up initial driver load.

Quoted 32 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -0
+
        return 0;
 
 failed:
@@ -2270,6 +2297,7 @@ void amdgpu_device_fini(struct amdgpu_device *adev)
        amdgpu_fbdev_fini(adev);
        r = amdgpu_fini(adev);
        adev->accel_working = false;
+       cancel_delayed_work_sync(&adev->late_init_work);
        /* free i2c buses */
        if (!amdgpu_device_has_dc_support(adev))
                amdgpu_i2c_fini(adev); @@ -2449,11 +2477,14 @@ int 

amdgpu_device_resume(struct drm_device *dev, bool resume, bool fbcon)

                if (r)
                        DRM_ERROR("ib ring test failed (%d).\n", r);
        }
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +4 -1
-
+       /* defer the cg state setting to post resume */
        r = amdgpu_late_init(adev);
        if (r)
                goto unlock;
 
+       mod_delayed_work(system_wq, &adev->late_init_work,
+                       msecs_to_jiffies(AMDGPU_RESUME_MS));
+
        /* pin cursors */
        list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
                struct amdgpu_crtc *amdgpu_crtc = 

to_amdgpu_crtc(crtc);
--
2.7.4

Quoted 4 lines

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx