[PATCH 06/10] drm/amdgpu: add condition to enable baco for xgmi/ras case

[Ma, Le]

Agree with your thoughts that we drop amdgpu_ras_enable=2 condition. The only concern in my side is that besides fatal_error, another result may happen that atombios_init timeout on xgmi by baco (not sure psp mode1 reset causes this as well).



Assuming no amdgpu_ras_enable=2 check, if PMFW > 40.52,  the use cases as my understanding includes:

  1.  sGPU without RAS:
     *   new: baco
     *   old: baco
  2.  sGPU with RAS:

  *   new: baco
  *   old: psp mode1 chain reset and legacy fatal_error handling

  1.  XGMI with RAS: baco
     *   new: baco
     *   old: psp mode1 chain reset and legacy fatal_error handling
  2.  XGMI without RAS: baco
     *   new: baco
     *   old: psp mode1 chain reset



That is to say, all uses cases go on baco road when PMFW > 40.52.



Regards,

Ma Le



-----Original Message-----
From: Zhang, Hawking <Hawking.Zhang at amd.com>
Sent: Wednesday, November 27, 2019 7:28 PM
To: Ma, Le <Le.Ma at amd.com>; amd-gfx at lists.freedesktop.org
Cc: Chen, Guchun <Guchun.Chen at amd.com>; Zhou1, Tao <Tao.Zhou1 at amd.com>; Li, Dennis <Dennis.Li at amd.com>; Deucher, Alexander <Alexander.Deucher at amd.com>; Ma, Le <Le.Ma at amd.com>
Subject: RE: [PATCH 06/10] drm/amdgpu: add condition to enable baco for xgmi/ras case



[AMD Public Use]



After thinking it a bit, I think we can just rely on PMFW version to decide to go RAS recovery or legacy fatal_error handling for the platforms that support RAS. Leveraging amdgpu_ras_enable as a temporary solution seems not necessary? Even baco ras recovery not stable, it is the same result as legacy fatal_error handling that user has to reboot the node manually.



So the new soc reset use cases are:

XGMI (without RAS): use PSP mode1 based chain reset, RAS enabled (with PMFW 40.52 and onwards): use BACO based RAS recovery, RAS enabled (with PMFW prior to 40.52): use legacy fatal_error handling.

Anything else?



Regards,

Hawking

-----Original Message-----

From: Le Ma <le.ma at amd.com<mailto:le.ma at amd.com>>

Sent: 2019年11月27日 17:15

To: amd-gfx at lists.freedesktop.org<mailto:amd-gfx at lists.freedesktop.org>

Cc: Zhang, Hawking <Hawking.Zhang at amd.com<mailto:Hawking.Zhang at amd.com>>; Chen, Guchun <Guchun.Chen at amd.com<mailto:Guchun.Chen at amd.com>>; Zhou1, Tao <Tao.Zhou1 at amd.com<mailto:Tao.Zhou1 at amd.com>>; Li, Dennis <Dennis.Li at amd.com<mailto:Dennis.Li at amd.com>>; Deucher, Alexander <Alexander.Deucher at amd.com<mailto:Alexander.Deucher at amd.com>>; Ma, Le <Le.Ma at amd.com<mailto:Le.Ma at amd.com>>

Subject: [PATCH 06/10] drm/amdgpu: add condition to enable baco for xgmi/ras case



Avoid to change default reset behavior for production card by checking amdgpu_ras_enable equal to 2. And only new enough smu ucode can support baco for xgmi/ras case.



Change-Id: I07c3e6862be03e068745c73db8ea71f428ecba6b

Signed-off-by: Le Ma <le.ma at amd.com<mailto:le.ma at amd.com>>

---

drivers/gpu/drm/amd/amdgpu/soc15.c | 4 +++-

1 file changed, 3 insertions(+), 1 deletion(-)



diff --git a/drivers/gpu/drm/amd/amdgpu/soc15.c b/drivers/gpu/drm/amd/amdgpu/soc15.c

index 951327f..6202333 100644

--- a/drivers/gpu/drm/amd/amdgpu/soc15.c

+++ b/drivers/gpu/drm/amd/amdgpu/soc15.c

@@ -577,7 +577,9 @@ soc15_asic_reset_method(struct amdgpu_device *adev)

                                   struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev, 0);

                                   struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);

-                                   if (hive || (ras && ras->supported))

+                                  if ((hive || (ras && ras->supported)) &&

+                                      (amdgpu_ras_enable != 2 ||

+                                      adev->pm.fw_version <= 0x283400))

                                               baco_reset = false;

                       }

                       break;

--

2.7.4
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20191127/a0a46d80/attachment.html>