[PATCH] i915/selftest/igt_mmap: let mmap tests run in kthread

[Mikolaj Wasiak]

Hi Andi,

> > This makes it impossible to use current->mm in selftests which currently
> > creates null pointer exception.
> > This patch allows selftest to use current->mm by using active_mm.
> 
> What is the failure you are getting? I'm not sure this might be
> the right solution, but it might make sense, though... if there
> is a valid reason.

current->mm is both used by the test and by functions it uses. It is used
as argument of vm_lookup() and inside of vm_mmap(), hence we cannot use
any other pointer.

> I would first ask myself why are we getting a NULL pointer
> dereference?

It's an expected behaviour when the system has multiple NUMA nodes.
The probe is called via work_on_cpu (pci_call_probe()) which makes it
run in separate workqueue on another cpu.
The workqueue is run in kthread and therefore we get NULL pointer.

> Why do we need to use current->mm? Which memory's
> should current->mm point to? Is it a kernel thread or a user
> thread? Why are we peaking inside current->mm?

This test is supposed to simulate user controling the graphics card -
get and set values in mapped /dev/dri/card0 file as a user.
It needs a mapping that the simulated user has access to. I don't know
how to let the user access the memory other way than creating a mapping
via vm_mmap() which uses current->mm.

> If we are hiding a bug with another bug then this patch is not
> OK, even if the second bug is less serious. So, please, don't do
> it.

Ok, I'll work on another solution.

Mikołaj