Re: [PATCH] staging: media: atomisp: Fix premature setting of HMM_BO_DEVICE_INITED flag

[Abdelrahman Fekry]

Hi Hans,

On Tue, Jul 1, 2025 at 4:54 PM Hans de Goede <hansg@xxxxxxxxxx> wrote:
>
> Hi,
>
> On 1-Jul-25 2:45 PM, Andy Shevchenko wrote:
> > On Tue, Jul 01, 2025 at 02:58:43PM +0300, Abdelrahman Fekry wrote:
> >> Hello Andy,
> >> On Sat, Jun 28, 2025 at 10:52 PM Andy Shevchenko
> >> <andy.shevchenko@xxxxxxxxx> wrote:
> >>> On Sat, Jun 28, 2025 at 8:26 AM Abdelrahman Fekry
> >>> <abdelrahmanfekry375@xxxxxxxxx> wrote:
> >
> >>>> The HMM_BO_DEVICE_INITED flag was being set in hmm_bo_device_init()
> >>>> before key initialization steps like kmem_cache_create(),
> >>>> kmem_cache_alloc(), and __bo_init().
> >>>>
> >>>> This means that if any of these steps fail, the flag remains set,
> >>>> misleading other parts of the driver (e.g. hmm_bo_alloc())
> >>>> into thinking the device is initialized. This could lead
> >>>> to undefined behavior or invalid memory use.
> >>>
> >>> Nice. Can you make some fault injection (temporary by modifying the
> >>> code to always fail, for example) and actually prove this in practice?
> >>> If so, the few (important) lines from the given Oops would be nice to
> >>> have here.
> >
> >> I have been trying to test it without having any intel atomisp
> >> hardware and failed continuously, do you have any tips or maybe some
> >> resources on how i can test this driver.
> >
> > So, the easiest way as I can see it is to ask people who possess the HW to
> > test, but you need to provide a testing patch (which can be applied on top
> > of this one, for example).
>
> I don't think it is worth testing the error path here,
> the old code is obviously wrong.
>
> What is interesting here is to see if the extra call to hmm_init()
> in __hmm_alloc() is necessary at all.
>
i think its redundant because in all scenarios hmm_init() should
be called before hmm_alloc()

> And obviously hmm_init() needs to propagate the error return
> from hmm_bo_init() right away instead of continuing if nothing
> was wrong and then only returning the error at the end.
>
> So it seems that there plenty of cleanup to do around this
> without needing error injection to test all paths.
>
> Actually I'm pretty sure that there will be quite a few
> error-handling paths with bugs in the atomisp code given
> its overall quality. But lets clean things up first, that
> should make addressing any such cases easier.
>
I totally agree with this , i have submitted a patch that cleans the
custom sysfs atrributes
as you suggested as a beginning , the patch got reviewed by andy and dan
here is the link
https://lore.kernel.org/all/20250627100604.29061-1-abdelrahmanfekry375@xxxxxxxxx/

What do you think I should work on next after these two patches, do
you have any suggestions?

> Regards,
>
> Hans
>
Best Regards,
Abdelrahman Fekry