Em Thu, 2 May 2024 10:37:42 +0000
Sakari Ailus <sakari.ailus@xxxxxx> escreveu:
> Hi Mauro,
>
> On Thu, May 02, 2024 at 11:25:47AM +0100, Mauro Carvalho Chehab wrote:
> > Em Thu, 2 May 2024 08:35:57 +0000
> > Sakari Ailus <sakari.ailus@xxxxxx> escreveu:
> >
> > > Hi Hans, Mauro,
> > >
> > > Here are a few fixes for the IPU6 and Unicam drivers, for issues found
> > > since merging them. Most are are due to Kconfig options.
> > >
> > > Please pull.
> > >
> > >
> > > The following changes since commit a1c6d22421501fc1016b99ac8570a1030714c70f:
> > >
> > > media: staging: max96712: fix copy-paste error (2024-04-30 14:33:08 +0100)
> > >
> > > are available in the Git repository at:
> > >
> > > git://linuxtv.org/sailus/media_tree.git tags/fixes-6.10-1.1-signed
> > >
> > > for you to fetch changes up to f28348ebadf7c7127646672f363e055cb19f3b66:
> > >
> > > media: intel/ipu6: Fix build with !ACPI (2024-05-02 11:34:26 +0300)
> > >
> > > ----------------------------------------------------------------
> > > IPU6 and Unicam fixes
> > >
> > > ----------------------------------------------------------------
> > > Laurent Pinchart (1):
> > > media: bcm2835-unicam: Include v4l2-subdev.h
> > >
> > > Ricardo Ribalda (4):
> > > media: bcm2835-unicam: Fix build with !PM
> > > media: intel/ipu6: Switch to RUNTIME_PM_OPS() and SYSTEM_SLEEP_PM_OPS
> > > media: intel/ipu6: Fix direct dependency Kconfig error
> > > media: intel/ipu6: Fix build with !ACPI
> >
> > The above were applied, thanks!
> >
> > >
> > > Sakari Ailus (1):
> > > media: intel/ipu6: Fix vmalloc memory allocation
> >
> > I didn't applied this one, as it sounds to me that the fix is
> > incomplete. I sent an email with the potential issues I found on it.
>
> This is a compile issue at the moment.
On such case, please include the compile error at the commit message.
> There may be other bugs (like not
> releasing the allocated memory) but it's a different problem. That's why
> I'd merge this one now.
If I understand well, there are then a couple of issues there:
1. the driver doesn't compile if no vmalloc;
2. the driver currently only works with vmalloc, as currently the firmware
load logic seems to depend on it:
if (is_vmalloc_addr(fw->data)) {
*firmware_p = fw;
return 0;
}
3. the driver has memory leaks: it doesn't de-allocate what it is
stored at *firmware_p pointer (nor the isp structure).
I would expect a patch that would at least fix (1) and (2) altogether.
I mean, if the driver can't work without vmalloc support, it has to
either one of the possible fixes below:
- depend on some Kconfig symbol that will prevent it to be compiled
when vmalloc is disabled/unsupported;
- depend on BROKEN;
- have the is_vmalloc_addr() and vmalloc()/vfree() replaced by
kvmalloc()/kvfree() at the same patch.
Regards,
Mauro
