The exclusive mode has real-time limitation in reality, such like being
done in 300ms. It's easy observed if running many VF/VMs in single host
with heavy CPU workload.
If we find the init fails due to exclusive mode timeout, try it again.
v2:
- rewrite the condition for readable value.
v3:
- fix typo, add comments for sleep
Acked-by: Alex Deucher <alexander.deucher at amd.com>
Signed-off-by: pding <Pixel.Ding at amd.com>
Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
Signed-off-by: Gary Sun <Gary.Sun at amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 10 ++++++++++
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 15 +++++++++++++--
2 files changed, 23 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 125f77d..385b10e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -2303,6 +2303,15 @@ int amdgpu_device_init(struct amdgpu_device *adev,
r = amdgpu_init(adev);
if (r) {
+ /* failed in exclusive mode due to timeout */
+ if (amdgpu_sriov_vf(adev) &&
+ !amdgpu_sriov_runtime(adev) &&
+ amdgpu_virt_mmio_blocked(adev) &&
+ !amdgpu_virt_wait_reset(adev)) {
+ dev_err(adev->dev, "VF exclusive mode timeout\n");
+ r = -EAGAIN;
+ goto failed;
+ }
dev_err(adev->dev, "amdgpu_init failed\n");
amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0);
amdgpu_fini(adev);
@@ -2390,6 +2399,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
amdgpu_vf_error_trans_all(adev);
if (runtime)
vga_switcheroo_fini_domain_pm_ops(adev->dev);
+
return r;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 720139e..f313eee 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -86,7 +86,7 @@ void amdgpu_driver_unload_kms(struct drm_device *dev)
int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
{
struct amdgpu_device *adev;
- int r, acpi_status;
+ int r, acpi_status, retry = 0;
#ifdef CONFIG_DRM_AMDGPU_SI
if (!amdgpu_si_support) {
@@ -122,6 +122,7 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
}
}
#endif
+retry_init:
adev = kzalloc(sizeof(struct amdgpu_device), GFP_KERNEL);
if (adev == NULL) {
@@ -144,7 +145,17 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
* VRAM allocation
*/
r = amdgpu_device_init(adev, dev, dev->pdev, flags);
- if (r) {
+ if (r == -EAGAIN && ++retry <= 3) {
+ adev->virt.caps &= ~AMDGPU_SRIOV_CAPS_RUNTIME;
+ adev->virt.ops = NULL;
+ amdgpu_device_fini(adev);
+ kfree(adev);
+ dev->dev_private = NULL;
+ /* Don't request EX mode too frequently which is attacking */
+ msleep(5000);
+ dev_err(&dev->pdev->dev, "retry init %d\n", retry);
+ goto retry_init;
+ } else if (r) {
dev_err(&dev->pdev->dev, "Fatal error during GPU init\n");
goto out;
}
--
1.7.1
Regards,
Gary
-----Original Message-----
From: Koenig, Christian
Sent: Tuesday, November 07, 2017 3:48 PM
To: Sun, Gary <Gary.Sun at amd.com>; amd-gfx at lists.freedesktop.org
Subject: Re: [PATCH] drm/amdgpu:remove debugfs file in amdgpu_device_finish
Hi Gary,
not sure what driver re-initialize feature you are talking about, but the last time I tried to re-initialize the driver it deadlocks in the modeset code because of some DC problem.
It's probably a good idea to fix that first, but in general please explain further what are you working on.
Regards,
Christian.
Am 07.11.2017 um 08:23 schrieb Sun, Gary:
> Hi Christian,
>
> The patch is for driver re- initialize feature, not for driver exit or rmmod. When the driver initialize failed at some point, the re- initialize feature will do some little clean and then try to initialize driver again, then it will re-register some registered debugfs , so it will fail.
>
> Regards,
> Gary
>
>
> -----Original Message-----
> From: Koenig, Christian
> Sent: Monday, November 06, 2017 5:26 PM
> To: Sun, Gary <Gary.Sun at amd.com>; amd-gfx at lists.freedesktop.org
> Subject: Re: [PATCH] drm/amdgpu:remove debugfs file in
> amdgpu_device_finish
>
> Am 06.11.2017 um 10:20 schrieb Gary Sun:
>> remove debugfs file in amdgpu_device_finish
> NAK, the debugfs files are removed automatically by drm_debugfs_cleanup().
>
> So that patch is unnecessary.
>
> Regards,
> Christian.
>
>> Signed-off-by: Gary Sun <Gary.Sun at amd.com>
>> ---
>> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 +
>> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 18 ++++++++++++++++++
>> 2 files changed, 19 insertions(+), 0 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> index 4f919d3..6cfcb5f 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> @@ -1250,6 +1250,7 @@ struct amdgpu_debugfs {
>> int amdgpu_debugfs_add_files(struct amdgpu_device *adev,
>> const struct drm_info_list *files,
>> unsigned nfiles);
>> +int amdgpu_debugfs_cleanup_files(struct amdgpu_device *adev);
>> int amdgpu_debugfs_fence_init(struct amdgpu_device *adev);
>>
>> #if defined(CONFIG_DEBUG_FS)
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> index 7b7439f..ee800ab 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> @@ -2520,6 +2520,7 @@ void amdgpu_device_fini(struct amdgpu_device *adev)
>> amdgpu_doorbell_fini(adev);
>> amdgpu_pm_sysfs_fini(adev);
>> amdgpu_debugfs_regs_cleanup(adev);
>> + amdgpu_debugfs_cleanup_files(adev);
>> }
>>
>>
>> @@ -3304,6 +3305,23 @@ int amdgpu_debugfs_add_files(struct amdgpu_device *adev,
>> return 0;
>> }
>>
>> +int amdgpu_debugfs_cleanup_files(struct amdgpu_device *adev) {
>> + unsigned int i;
>> +
>> + for (i = 0; i < adev->debugfs_count; i++) { #if
>> +defined(CONFIG_DEBUG_FS)
>> + drm_debugfs_remove_files(adev->debugfs[i].files,
>> + adev->debugfs[i].num_files,
>> + adev->ddev->primary);
>> +#endif
>> + adev->debugfs[i].files = NUL;
>> + adev->debugfs[i].num_files = 0;
>> + }
>> + adev->debugfs_count = 0;
>> + return 0;
>> +}
>> +
>> #if defined(CONFIG_DEBUG_FS)
>>
>> static ssize_t amdgpu_debugfs_regs_read(struct file *f, char
>> __user *buf,
>
