Hi Hawking,
I don't want to divert from this important discussion, but would like clarification...
In the rocm-smi library, I previously had this interpretation of the values (I think someone had confirmed this):
none, //!< No current errors
disabled //!< ECC is disabled
parity //!< ECC errors present, but type unknown
single_correctable, //!< Single correctable error
multi_correctable //!< Multiple uncorrectable errors
poison //!< Firmware detected error and isolated
//!< page. Treat as uncorrectable.
It seems now we are just saying it's enabled/disabled, but no indication of whether there are errors or not. Is that right, or am I misunderstanding something?
Chris
-----Original Message-----
From: Koenig, Christian <Christian.Koenig@xxxxxxx>
Sent: Thursday, August 8, 2019 10:24 AM
To: Zhang, Hawking <Hawking.Zhang@xxxxxxx>; Russell, Kent <Kent.Russell@xxxxxxx>; Zhou1, Tao <Tao.Zhou1@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx; Pan, Xinhui <Xinhui.Pan@xxxxxxx>; Freehill, Chris <Chris.Freehill@xxxxxxx>
Subject: Re: [PATCH] drm/amdgpu: update ras sysfs feature info
Hi Hawking,
yeah, but this isn't sufficient to comply to the upstream rules.
See what we need to do is to remove the extra text and the per IP information. In other words something like this:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 1a4412e47810..c8c7f9655ffc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -819,22 +819,7 @@ static ssize_t
amdgpu_ras_sysfs_features_read(struct device *dev,
ssize_t s;
struct ras_manager *obj;
- s = scnprintf(buf, PAGE_SIZE, "feature mask: 0x%x\n",
con->features);
-
- for (i = 0; i < ras_block_count; i++) {
- head.block = i;
-
- if (amdgpu_ras_is_feature_enabled(adev, &head)) {
- obj = amdgpu_ras_find_obj(adev, &head);
- s += scnprintf(&buf[s], PAGE_SIZE - s,
- "%s: %s\n",
- ras_block_str(i),
- ras_err_str(obj->head.type));
- } else
- s += scnprintf(&buf[s], PAGE_SIZE - s,
- "%s: disabled\n",
- ras_block_str(i));
- }
+ s = scnprintf(buf, PAGE_SIZE, "0x%x\n", con->features);
return s;
}
And I'm not talking about some rule we could bend. This is an absolutely
*MUST* have for upstreaming.
Regards,
Christian.
Am 08.08.19 um 17:17 schrieb Zhang, Hawking:
> Hi Chris,
>
> I thought I've stated as "The feature mask is already good enough for this node". It is just a uint32 value. We don't expect ROCm SMI to read any other information from feature node except for the feature mask.
>
> Regards,
> Hawking
>
> -----Original Message-----
> From: Koenig, Christian <Christian.Koenig@xxxxxxx>
> Sent: 2019年8月8日 23:11
> To: Zhang, Hawking <Hawking.Zhang@xxxxxxx>; Russell, Kent
> <Kent.Russell@xxxxxxx>; Zhou1, Tao <Tao.Zhou1@xxxxxxx>;
> amd-gfx@xxxxxxxxxxxxxxxxxxxxx; Pan, Xinhui <Xinhui.Pan@xxxxxxx>;
> Freehill, Chris <Chris.Freehill@xxxxxxx>
> Subject: Re: [PATCH] drm/amdgpu: update ras sysfs feature info
>
> Hi Hawking,
>
> a multi line value is not the problem, but here you have multiple values.
>
> E.g. in the case of the pp tables that is one big array of power profiles and we actually had a discussion if exposing them like this is ok or not.
>
> But in the case of the ras features you got multiple different things in the same file. And that in turn is a clear violation of the sysfs rules.
>
> I don't think we can't upstream the interface like this. See here for
> a good summary of the rules as well: https://lwn.net/Articles/378884/
>
> Regards,
> Christian.
>
> Am 08.08.19 um 17:00 schrieb Zhang, Hawking:
>> Hi Chris,
>>
>> I'm not aware of how ROCM SMI using feature nodes. but not all the sysfs are intended to be used by upper level apps/libs.
>>
>> There are bunches of sysfs entries that have multiple line value. The most complicate one would be pp_power_profile_mode, which looks like.
>>
>> 0 BOOTUP_DEFAULT*:
>> 0( GFXCLK) 0 0 1 0 4 800 4587520 -65536 0
>> 1( SOCCLK) 0 0 1 0 4 800 327680 -6553 0
>> 2( UCLK) 0 0 1 0 4 800 327680 -65536 0
>> .......
>> 1 3D_FULL_SCREEN :
>> 0( GFXCLK) 0 1 1 0 4 800 4587520 -65536 0
>> 1( SOCCLK) 0 1 4 850 4 800 327680 -65536 0
>>
>> Regards,
>> Hawking
>> -----Original Message-----
>> From: Christian König <ckoenig.leichtzumerken@xxxxxxxxx>
>> Sent: 2019年8月8日 22:25
>> To: Zhang, Hawking <Hawking.Zhang@xxxxxxx>; Russell, Kent
>> <Kent.Russell@xxxxxxx>; Zhou1, Tao <Tao.Zhou1@xxxxxxx>;
>> amd-gfx@xxxxxxxxxxxxxxxxxxxxx; Pan, Xinhui <Xinhui.Pan@xxxxxxx>;
>> Freehill, Chris <Chris.Freehill@xxxxxxx>
>> Subject: Re: [PATCH] drm/amdgpu: update ras sysfs feature info
>>
>> Hi Hawking,
>>
>> looks like you skipped my response.
>>
>> Even the current way how sysfs is used in the ras code is a clear NO-GO and should be fixed before this is pushed upstream.
>>
>> A sysfs entry should seriously NOT return a multi line value which needs to be extensively parsed by the application.
>>
>> Regards,
>> Christian.
>>
>> Am 08.08.19 um 15:50 schrieb Zhang, Hawking:
>>> Understood and agree we should keep stable interfaces.
>>>
>>> But the information in feature node is not correct and makes people confusing. Basically, each IP blocks can support all the four error types, not just multi-uncorrectable. As a result, any upper level apps/libs that read from this file will just get confusing information as well. The feature mask is already good enough for this node.
>>>
>>> Regards,
>>> Hawking
>>> -----Original Message-----
>>> From: Russell, Kent <Kent.Russell@xxxxxxx>
>>> Sent: 2019年8月8日 20:51
>>> To: Zhang, Hawking <Hawking.Zhang@xxxxxxx>; Zhou1, Tao
>>> <Tao.Zhou1@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx; Pan, Xinhui
>>> <Xinhui.Pan@xxxxxxx>; Freehill, Chris <Chris.Freehill@xxxxxxx>
>>> Cc: Zhou1, Tao <Tao.Zhou1@xxxxxxx>
>>> Subject: RE: [PATCH] drm/amdgpu: update ras sysfs feature info
>>>
>>> +Chris Freehill
>>>
>>> While I can understand this change, this broke our SMI interface, which was expecting a specific string format for the ras/features file. This has happened a few times now, where changes to the RAS sysfs files has broke the SMI CLI and/or SMI LIB. Can we please get a stable interface and sysfs format set up before publishing patches? This is creating a lot of extra work for developers with the SMI to constantly keep up with the changes being made to sysfs files. Thank you.
>>>
>>> Kent
>>>
>>> -----Original Message-----
>>> From: amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> On Behalf Of
>>> Zhang, Hawking
>>> Sent: Monday, August 5, 2019 4:15 AM
>>> To: Zhou1, Tao <Tao.Zhou1@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx;
>>> Pan, Xinhui <Xinhui.Pan@xxxxxxx>
>>> Cc: Zhou1, Tao <Tao.Zhou1@xxxxxxx>
>>> Subject: RE: [PATCH] drm/amdgpu: update ras sysfs feature info
>>>
>>> Reviewed-by: Hawking Zhang <Hawking.Zhang@xxxxxxx>
>>>
>>> Regards,
>>> Hawking
>>> -----Original Message-----
>>> From: amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> On Behalf Of
>>> Tao Zhou
>>> Sent: 2019年8月5日 16:04
>>> To: amd-gfx@xxxxxxxxxxxxxxxxxxxxx; Pan, Xinhui <Xinhui.Pan@xxxxxxx>;
>>> Zhang, Hawking <Hawking.Zhang@xxxxxxx>
>>> Cc: Zhou1, Tao <Tao.Zhou1@xxxxxxx>
>>> Subject: [PATCH] drm/amdgpu: update ras sysfs feature info
>>>
>>> remove confused ras error type info
>>>
>>> Signed-off-by: Tao Zhou <tao.zhou1@xxxxxxx>
>>> ---
>>> drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 17 +++++------------
>>> 1 file changed, 5 insertions(+), 12 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
>>> index d2e8a85f6e38..369651247b23 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
>>> @@ -787,25 +787,18 @@ static ssize_t amdgpu_ras_sysfs_features_read(struct device *dev,
>>> struct amdgpu_device *adev = ddev->dev_private;
>>> struct ras_common_if head;
>>> int ras_block_count = AMDGPU_RAS_BLOCK_COUNT;
>>> - int i;
>>> + int i, enabled;
>>> ssize_t s;
>>> - struct ras_manager *obj;
>>>
>>> s = scnprintf(buf, PAGE_SIZE, "feature mask: 0x%x\n",
>>> con->features);
>>>
>>> for (i = 0; i < ras_block_count; i++) {
>>> head.block = i;
>>> + enabled = amdgpu_ras_is_feature_enabled(adev, &head);
>>>
>>> - if (amdgpu_ras_is_feature_enabled(adev, &head)) {
>>> - obj = amdgpu_ras_find_obj(adev, &head);
>>> - s += scnprintf(&buf[s], PAGE_SIZE - s,
>>> - "%s: %s\n",
>>> - ras_block_str(i),
>>> - ras_err_str(obj->head.type));
>>> - } else
>>> - s += scnprintf(&buf[s], PAGE_SIZE - s,
>>> - "%s: disabled\n",
>>> - ras_block_str(i));
>>> + s += scnprintf(&buf[s], PAGE_SIZE - s,
>>> + "%s ras feature mask: %s\n",
>>> + ras_block_str(i), enabled?"on":"off");
>>> }
>>>
>>> return s;
_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
