[Accel-config] Re: [PATCH 1/1] accel-config/test: Remove pasid_enabled restriction from test script

--===============1240020318942107841==
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable

Against upstream? I'm surprised. There's no DMA pasid support and kernel =

pasid would get 0.

On 7/7/2022 3:01 PM, Yu, Fenghua wrote:
> Hi, Dave,
>
> 5.19-rc4 doesn't have the issue. On 5.19-rc4, pasid_enabled=3D1 on sm_on =
and 0 on sm_off. So the value is expected. Tony said the issue exists in a =
earlier kernel. But I don't know which patch fixes the issue.
>
> So seems there is no fix here.
>
> Thanks.
>
> -Fenghua
>
>> -----Original Message-----
>> From: Jiang, Dave <dave.jiang(a)intel.com>
>> Sent: Thursday, July 07, 2022 2:57 PM
>> To: Shen, Xiaochen <xiaochen.shen(a)intel.com>; Zhu, Tony
>> <tony.zhu(a)intel.com>; Yu, Fenghua <fenghua.yu(a)intel.com>
>> Cc: accel-config(a)lists.01.org; Thomas, Ramesh <ramesh.thomas(a)intel.c=
om>
>> Subject: Re: [Accel-config] Re: [PATCH 1/1] accel-config/test: Remove
>> pasid_enabled restriction from test script
>>
>> After discussion with Fenghua, I think the conclusion is that we leave p=
asid_en
>> tied to the user pasid enable. Reason being:
>>
>> 1. preserve legacy usage
>>
>> 2. no current kernel user care about the sysfs attribute, so until a cus=
tomer really
>> wants to know about the kernel side, no need to introduce a kernel_pasid=
_en
>> attribute.
>>
>> so I would change the patch to:
>>
>> sysfs_emit(buf, "%u\n", device_user_pasid_enabled(idxd));
>>
>> Otherwise if user pasid is disabled and DMA pasid is enabled, your scrip=
t will still
>> fail.
>>
>> On 6/28/2022 8:00 PM, Shen, Xiaochen wrote:
>>> Hi Tony and Fenghua,
>>>
>>> This issue may be impacted by this 5.19 upstream patch:
>>> 42a1b73852c4a176d233a192422b5e1d0ba67cbf dmaengine: idxd: Separate
>>> user and kernel pasid enabling
>>>
>>> The sysfs interface "pasid_enabled" doesn't reflect the newly added fla=
g:
>>>           IDXD_FLAG_PASID_ENABLED,
>>> +       IDXD_FLAG_USER_PASID_ENABLED,
>>>
>>>
>>> This patch may fix this issue:
>>>
>>> diff --git a/drivers/dma/idxd/sysfs.c b/drivers/dma/idxd/sysfs.c index
>>> dfd549685c46..53e34a1d62d9 100644
>>> --- a/drivers/dma/idxd/sysfs.c
>>> +++ b/drivers/dma/idxd/sysfs.c
>>> @@ -1223,7 +1223,8 @@ static ssize_t pasid_enabled_show(struct device
>> *dev,
>>>    {
>>>           struct idxd_device *idxd =3D confdev_to_idxd(dev);
>>>
>>> -       return sysfs_emit(buf, "%u\n", device_pasid_enabled(idxd));
>>> +       return sysfs_emit(buf, "%u\n",
>>> +                         device_pasid_enabled(idxd) ||
>>> + device_user_pasid_enabled(idxd));
>>>    }
>>>    static DEVICE_ATTR_RO(pasid_enabled);
>>>
>>>
>>> Best regards,
>>> Xiaochen
>>>
>>> -----Original Message-----
>>> From: Zhu, Tony <tony.zhu(a)intel.com>
>>> Sent: Wednesday, June 29, 2022 10:05
>>> To: Yu, Fenghua <fenghua.yu(a)intel.com>
>>> Cc: accel-config(a)lists.01.org; Thomas, Ramesh
>>> <ramesh.thomas(a)intel.com>
>>> Subject: [Accel-config] Re: [PATCH 1/1] accel-config/test: Remove
>>> pasid_enabled restriction from test script
>>>
>>> Fenghua,
>>>
>>>     pasid_enabled is 1 when it is scalable mode for old kernel such as =
spr-bkc.
>> For kernel code 5.18, pasid_enabled is 0. Though pasid_enable is 0, but =
the
>> passid is still assigned.
>>> I didn't know which commit bring this change. But from the test result,=
 I could
>> see pasid table entry in dmesg log.
>>>
>>> Tony(zhu, xinzhan)
>>> Cube:SHZ1-3W-279
>>> iNet:8821-5077
>>>
>>> -----Original Message-----
>>> From: Yu, Fenghua <fenghua.yu(a)intel.com>
>>> Sent: Tuesday, June 28, 2022 11:38 PM
>>> To: Zhu, Tony <tony.zhu(a)intel.com>
>>> Cc: accel-config(a)lists.01.org; Thomas, Ramesh
>>> <ramesh.thomas(a)intel.com>
>>> Subject: Re: [PATCH 1/1] accel-config/test: Remove pasid_enabled
>>> restriction from test script
>>>
>>> Hi, Tony,
>>>
>>> On Tue, Jun 28, 2022 at 02:40:24PM +0800, Tony Zhu wrote:
>>>> Kernel removed the restriction because it broken accel-config.
>>>> Failure
>>>                                              s/broken/breaks/
>>>> will happen during wq enable. Remove the checking from test script too.
>>>>
>>>> Signed-off-by: Tony Zhu <tony.zhu(a)intel.com>
>>>> ---
>>>>    test/dsa_user_test_runner.sh | 5 -----
>>>>    1 file changed, 5 deletions(-)
>>>>
>>>> diff --git a/test/dsa_user_test_runner.sh
>>>> b/test/dsa_user_test_runner.sh index dfaa930..ae9010d 100755
>>>> --- a/test/dsa_user_test_runner.sh
>>>> +++ b/test/dsa_user_test_runner.sh
>>>> @@ -19,11 +19,6 @@ check_min_kver "5.6" || do_skip "kernel does not
>> support idxd"
>>>>    # skip if no pasid support as dsa_test does not support operation w=
/o pasid
>> yet.
>>>>    [ ! -f "/sys/bus/dsa/devices/$DSA/pasid_enabled" ] && echo "No SVM
>> support" && exit "$EXIT_SKIP"
>>>> -pasid_en=3D$(cat /sys/bus/dsa/devices/$DSA/pasid_enabled)
>>>> -if [ "$pasid_en" -ne 1 ]; then
>>>> -exit "$EXIT_SKIP"
>>>> -fi
>>>> -
>>> Without pasid, what sub-tests can run? Seems to me, without PASID enabl=
ed,
>> the test cannot continue.
>>>>    start_dsa()
>>>>    {
>>>>    configurable=3D$(cat /sys/bus/dsa/devices/$DSA/configurable)
>>>> --
>>>> 2.27.0
>>>>
>>> Thanks.
>>>
>>> -Fenghua
>>> _______________________________________________
>>> Accel-config mailing list -- accel-config(a)lists.01.org To unsubscribe
>>> send an email to accel-config-leave(a)lists.01.org
>>> _______________________________________________
>>> Accel-config mailing list -- accel-config(a)lists.01.org To unsubscribe
>>> send an email to accel-config-leave(a)lists.01.org
--===============1240020318942107841==--