On Wed, Aug 22, 2018 at 11:16 PM Zhenyu Wang wrote:
>
> yeah, that's the clear way to fix this imo. We only depend on guest
> life cycle to access guest memory properly. Here's proposed fix, will
> verify and integrate it later.
Thanks, this looks sane to me,
Linus
On 23/08/2018 08:07, Zhenyu Wang wrote:
> On 2018.08.22 20:20:46 +0200, Paolo Bonzini wrote:
>> On 22/08/2018 18:44, Linus Torvalds wrote:
>>> An example of something that *isn't* right, is the i915 kvm interface,
>>> which does
>>>
>>> use_mm(kvm->mm);
>>>
>>> on an mm that was
On 2018.08.22 20:20:46 +0200, Paolo Bonzini wrote:
> On 22/08/2018 18:44, Linus Torvalds wrote:
> > An example of something that *isn't* right, is the i915 kvm interface,
> > which does
> >
> > use_mm(kvm->mm);
> >
> > on an mm that was initialized in virt/kvm/kvm_main.c using
> >
> >
On Wed, Aug 22, 2018 at 12:44 PM Felix Kuehling wrote:
>
> You're right, but that's a bit fragile and convoluted. I'll fix KFD to
> handle this more robustly. See the attached (untested) patch.
Yes, this patch that makes the whole "has to use current mm" or uses
"get_task_mm()" looks good from a
On Wed, Aug 22, 2018 at 10:58 PM Linus Torvalds
wrote:
>
> On Wed, Aug 22, 2018 at 12:37 PM Oded Gabbay wrote:
> >
> > Having said that, I think we *are* protected by the mmu_notifier
> > release because if the process suddenly dies, we will gracefully clean
> > the process's data in our driver
On Wed, Aug 22, 2018 at 12:37 PM Oded Gabbay wrote:
>
> Having said that, I think we *are* protected by the mmu_notifier
> release because if the process suddenly dies, we will gracefully clean
> the process's data in our driver and on the H/W before returning to
> the mm core code. And before we
On 2018-08-22 02:13 PM, Christian König wrote:
> Adding Felix because the KFD part of amdgpu is actually his
> responsibility.
>
> If I'm not completely mistaken the release callback of the
> mmu_notifier should take care of that for amdgpu.
You're right, but that's a bit fragile and convoluted.
On Wed, Aug 22, 2018 at 7:44 PM Linus Torvalds
wrote:
> One of the complex ones is the amdgpu driver. It does a
> "use_mm(mmptr)" deep deep in the guts of a macro that ends up being
> used in fa few places, and it's very hard to tell if it's right.
>
> It looks almost certainly buggy (there is no
Hi Linus:
Thanks for letting us know that. We would fix this ASAP. The kvmgt.c
module is a part of GVT-g code. It's our fault that we didn't find this
mis-uses, not i915 or KVM guys. Wish they would feel better after seeing
this message.
Thanks,
Zhi.
On 08/23/18 00:44, Linus Torvalds
On Wed, Aug 22, 2018 at 11:33 AM Linus Torvalds
wrote:
>
> On Wed, Aug 22, 2018 at 11:21 AM Paolo Bonzini wrote:
> >
> > Yes, KVM is correct but the i915 bits are at least fishy. It's probably
> > as simple as adding a mmget/mmput pair respectively in kvmgt_guest_init
> > and kvmgt_guest_exit,
On Wed, Aug 22, 2018 at 11:21 AM Paolo Bonzini wrote:
>
> Yes, KVM is correct but the i915 bits are at least fishy. It's probably
> as simple as adding a mmget/mmput pair respectively in kvmgt_guest_init
> and kvmgt_guest_exit, or maybe mmget_not_zero.
Definitely mmget_not_zero(). If it was
On 22/08/2018 18:44, Linus Torvalds wrote:
> An example of something that *isn't* right, is the i915 kvm interface,
> which does
>
> use_mm(kvm->mm);
>
> on an mm that was initialized in virt/kvm/kvm_main.c using
>
> mmgrab(current->mm);
> kvm->mm = current->mm;
>
>
Adding Felix because the KFD part of amdgpu is actually his responsibility.
If I'm not completely mistaken the release callback of the mmu_notifier
should take care of that for amdgpu.
Regards,
Christian.
Am 22.08.2018 um 18:44 schrieb Linus Torvalds:
Guys and gals,
this is a *very*
Guys and gals,
this is a *very* random list of people on the recipients list, but we
had a subtle TLB shootdown issue in the VM, and that brought up some
issues when people then went through the code more carefully.
I think we have a handle on the TLB shootdown bug itself. But when
people were
14 matches
Mail list logo
