Exynos IOMMU driver doesn't work?

Exynos IOMMU driver doesn't work?

[Hongbo Zhang]

Hi Cho KyongHo, Joerg et al,
I found the latest Exynos IOMMU driver doesn't work, the line 481:
BUG_ON(!has_sysmmu(dev));
in function __exynos_sysmmu_enable() in file exynos-iommu.c triggers
kernel panic.

Then I found the dev->archdata.iommu isn't initialized at all, it
should be the root cause.

Am I right? any comment, solution?

Another problem is this driver is added support of device tree, but
there is no device tree nodes in the dts file, so I had to search from
internet and added those nodes manually.

I've found these links of v12 and v13 patches
https://lkml.org/lkml/2014/4/27/171
https://lkml.org/lkml/2014/5/12/34
patch v13 was merged into mainline kernel, but as a part of v12, it
isn't complete and doesn't work alone, eg dts nodes are missing.
(I didn't research much dev->archdata.iommu initialization error is
introduced by which patch, but it seems in very old codes there is no
such problem)

Thanks.

Re: Exynos IOMMU driver doesn't work?

[Hongbo Zhang]

Add linux-samsung-soc-u79uwXL29TY76Z2rM5mHXA@public.gmane.org mailing list.

On 7 January 2015 at 18:31, Hongbo Zhang <hongbo.zhang-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org> wrote:
> Hi Cho KyongHo, Joerg et al,
> I found the latest Exynos IOMMU driver doesn't work, the line 481:
> BUG_ON(!has_sysmmu(dev));
> in function __exynos_sysmmu_enable() in file exynos-iommu.c triggers
> kernel panic.
>
> Then I found the dev->archdata.iommu isn't initialized at all, it
> should be the root cause.
>
> Am I right? any comment, solution?
>
> Another problem is this driver is added support of device tree, but
> there is no device tree nodes in the dts file, so I had to search from
> internet and added those nodes manually.
>
> I've found these links of v12 and v13 patches
> https://lkml.org/lkml/2014/4/27/171
> https://lkml.org/lkml/2014/5/12/34
> patch v13 was merged into mainline kernel, but as a part of v12, it
> isn't complete and doesn't work alone, eg dts nodes are missing.
> (I didn't research much dev->archdata.iommu initialization error is
> introduced by which patch, but it seems in very old codes there is no
> such problem)
>
> Thanks.

Re: Exynos IOMMU driver doesn't work?

[Javier Martinez Canillas]

[adding Marek, Sjoerd and Joonyoung that were discussing about iommu
support in another thread]

Hello Hongbo,

On Fri, Jan 9, 2015 at 8:31 AM, Hongbo Zhang <hongbo.zhang-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org> wrote:
> Add linux-samsung-soc-u79uwXL29TY76Z2rM5mHXA@public.gmane.org mailing list.
>
> On 7 January 2015 at 18:31, Hongbo Zhang <hongbo.zhang-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org> wrote:
>> Hi Cho KyongHo, Joerg et al,
>> I found the latest Exynos IOMMU driver doesn't work, the line 481:
>> BUG_ON(!has_sysmmu(dev));
>> in function __exynos_sysmmu_enable() in file exynos-iommu.c triggers
>> kernel panic.
>>
>> Then I found the dev->archdata.iommu isn't initialized at all, it
>> should be the root cause.
>>

That's correct, I found the same the other day since and thought about
posting a patch to return -ENODEV if !has_sysmmu(dev) instead to avoid
the driver to panic the kernel. But then I realized this is already
fixed in Marek's "[PATCH v3 00/19] Exynos SYSMMU (IOMMU) integration
with DT and DMA-mapping subsystem" series [0].

>> Another problem is this driver is added support of device tree, but
>> there is no device tree nodes in the dts file, so I had to search from
>> internet and added those nodes manually.
>>
>> I've found these links of v12 and v13 patches
>> https://lkml.org/lkml/2014/4/27/171
>> https://lkml.org/lkml/2014/5/12/34
>> patch v13 was merged into mainline kernel, but as a part of v12, it
>> isn't complete and doesn't work alone, eg dts nodes are missing.
>> (I didn't research much dev->archdata.iommu initialization error is
>> introduced by which patch, but it seems in very old codes there is no
>> such problem)
>>

Yes, please take a look to Marek series [0]. Keep in mind that the
series does not support all sysmmu revisions so IOMMU is not supported
for some SoCs (e.g: Exynos5). Support for that is planned once that
series land into mainline though [1].

May I ask why are you interested in IOMMU support on Exynos? I'm
asking because the reason why I tried to enable IOMMU support (and hit
the same issue) was to try using the Exynos DRM HDMI driver with IOMMU
since I found that HDMI is working on the downstream Samsung kernel
[2] that has IOMMU support, but is not working on mainline.

At the end the HDMI problem seems to not be IOMMU related but
something with the power domains and clocking but in case you are
facing the same issue, you may be interested in that discussion [3].

Best regards,
Javier

[0]: http://www.spinics.net/lists/linux-samsung-soc/msg39168.html
[1]: http://www.spinics.net/lists/linux-samsung-soc/msg39980.html
[2]: git-9UaJU3cA/F/QT0dZR+AlfA@public.gmane.org:exynos-reference/kernel.git
[3]: http://www.spinics.net/lists/linux-samsung-soc/msg40828.html

Re: Exynos IOMMU driver doesn't work?

[Hongbo Zhang]

On 9 January 2015 at 23:34, Javier Martinez Canillas
<javier-0uQlZySMnqxg9hUCZPvPmw@public.gmane.org> wrote:
> [adding Marek, Sjoerd and Joonyoung that were discussing about iommu
> support in another thread]

Thank you Javier.

>
> Hello Hongbo,
>
> On Fri, Jan 9, 2015 at 8:31 AM, Hongbo Zhang <hongbo.zhang-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org> wrote:
>> Add linux-samsung-soc-u79uwXL29TY76Z2rM5mHXA@public.gmane.org mailing list.
>>
>> On 7 January 2015 at 18:31, Hongbo Zhang <hongbo.zhang-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org> wrote:
>>> Hi Cho KyongHo, Joerg et al,
>>> I found the latest Exynos IOMMU driver doesn't work, the line 481:
>>> BUG_ON(!has_sysmmu(dev));
>>> in function __exynos_sysmmu_enable() in file exynos-iommu.c triggers
>>> kernel panic.
>>>
>>> Then I found the dev->archdata.iommu isn't initialized at all, it
>>> should be the root cause.
>>>
>
> That's correct, I found the same the other day since and thought about
> posting a patch to return -ENODEV if !has_sysmmu(dev) instead to avoid
> the driver to panic the kernel. But then I realized this is already
> fixed in Marek's "[PATCH v3 00/19] Exynos SYSMMU (IOMMU) integration
> with DT and DMA-mapping subsystem" series [0].
>
>>> Another problem is this driver is added support of device tree, but
>>> there is no device tree nodes in the dts file, so I had to search from
>>> internet and added those nodes manually.
>>>
>>> I've found these links of v12 and v13 patches
>>> https://lkml.org/lkml/2014/4/27/171
>>> https://lkml.org/lkml/2014/5/12/34
>>> patch v13 was merged into mainline kernel, but as a part of v12, it
>>> isn't complete and doesn't work alone, eg dts nodes are missing.
>>> (I didn't research much dev->archdata.iommu initialization error is
>>> introduced by which patch, but it seems in very old codes there is no
>>> such problem)
>>>
>
> Yes, please take a look to Marek series [0]. Keep in mind that the
> series does not support all sysmmu revisions so IOMMU is not supported
> for some SoCs (e.g: Exynos5). Support for that is planned once that
> series land into mainline though [1].
>
> May I ask why are you interested in IOMMU support on Exynos? I'm
> asking because the reason why I tried to enable IOMMU support (and hit
> the same issue) was to try using the Exynos DRM HDMI driver with IOMMU
> since I found that HDMI is working on the downstream Samsung kernel
> [2] that has IOMMU support, but is not working on mainline.

Because I am testing vfio-platform patches, IOMMU is used in this case.
http://www.spinics.net/lists/kvm-arm/msg12445.html

And I am glad to find a working kernel as you pointed out, then I
found these two commits in this tree may solve my problem:
841a7fe TEMP/TO POST: iommu: exynos: Add "mmu-masters" support
bd7e4c7 TEMP/TO POST: ARM: dts: add System MMU nodes of Exynos SoCs

>
> At the end the HDMI problem seems to not be IOMMU related but
> something with the power domains and clocking but in case you are
> facing the same issue, you may be interested in that discussion [3].
>
> Best regards,
> Javier
>
> [0]: http://www.spinics.net/lists/linux-samsung-soc/msg39168.html
> [1]: http://www.spinics.net/lists/linux-samsung-soc/msg39980.html
> [2]: git-9UaJU3cA/F/QT0dZR+AlfA@public.gmane.org:exynos-reference/kernel.git
> [3]: http://www.spinics.net/lists/linux-samsung-soc/msg40828.html

Re: Exynos IOMMU driver doesn't work?

[Javier Martinez Canillas]

Hello Hongbo,

On Mon, Jan 12, 2015 at 11:51 AM, Hongbo Zhang <hongbo.zhang-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org> wrote:
> On 9 January 2015 at 23:34, Javier Martinez Canillas
>>
>> Yes, please take a look to Marek series [0]. Keep in mind that the
>> series does not support all sysmmu revisions so IOMMU is not supported
>> for some SoCs (e.g: Exynos5). Support for that is planned once that
>> series land into mainline though [1].
>>
>> May I ask why are you interested in IOMMU support on Exynos? I'm
>> asking because the reason why I tried to enable IOMMU support (and hit
>> the same issue) was to try using the Exynos DRM HDMI driver with IOMMU
>> since I found that HDMI is working on the downstream Samsung kernel
>> [2] that has IOMMU support, but is not working on mainline.
>
> Because I am testing vfio-platform patches, IOMMU is used in this case.
> http://www.spinics.net/lists/kvm-arm/msg12445.html
>

Ok, different use case then.

> And I am glad to find a working kernel as you pointed out, then I

Glad that you found the information useful.

> found these two commits in this tree may solve my problem:
> 841a7fe TEMP/TO POST: iommu: exynos: Add "mmu-masters" support
> bd7e4c7 TEMP/TO POST: ARM: dts: add System MMU nodes of Exynos SoCs
>
>>

Yes, that's what we cherry-picked as well to test HDMI since enabling
IOMMU has a side effect of turning on the right power domains and
enabling the needed clocks.

Best regards,
Javier