Re: io_page_fault seen in normal kernel boot log

[Joerg Roedel]

On Mon, Sep 05, 2016 at 06:11:58PM +0800, Baoquan He wrote:
> On 09/05/16 at 11:59am, Joerg Roedel wrote:
> > > This happened on 4.8.0-rc5
> > > 
> > > [  +0.006771] AMD-Vi: Event logged [IO_PAGE_FAULT device=00:14.4 domain=0x0009 address=0x0000000000001000 flags=0x0000]
> > > [  +0.010593] AMD-Vi: Event logged [IO_PAGE_FAULT device=00:14.4 domain=0x0009 address=0x0000000000001040 flags=0x0000]
> > > [  +0.010596] AMD-Vi: Event logged [IO_PAGE_FAULT device=00:14.4 domain=0x0009 address=0x0000000000001080 flags=0x0000]
> > > [  +0.010597] AMD-Vi: Event logged [IO_PAGE_FAULT device=00:14.4 domain=0x0009 address=0x00000000000010c0 flags=0x0000]
> > > [  +0.010592] AMD-Vi: Event logged [IO_PAGE_FAULT device=00:14.4 domain=0x0009 address=0x0000000000001100 flags=0x0000]
> > > [  +0.010592] AMD-Vi: Event logged [IO_PAGE_FAULT device=00:14.4 domain=0x0009 address=0x0000000000001140 flags=0x0000]
> > 
> > This is probably a radeon driver issue, the address looks like its
> > handed out by the IOMMU driver, so probably a use-after-free bug. I also
> > see those GPU related faults from time to time. The system usually boots
> > fine anyway. Does boot complete on that system?
> 
> Yes, except for this message, system is not impacted.

Hmm, on the other hand, the AMD-Vi driver in 4.8 uses the generic IOVA
allocator, so the addresses it hands out should be closer to 4GB.
Anyway, this is probably a bug/problem in the GPU driver.



	Joerg