vm_area at addr ffffffffc0800000 is not marked as VM_IOREMAP

[Manivannan Sadhasivam]

On Thu, Mar 07, 2024 at 07:49:16PM -0800, Alexei Starovoitov wrote:
> On Thu, Mar 7, 2024 at 9:54 AM Alexei Starovoitov
> <alexei.starovoitov at gmail.com> wrote:
> >
> > On Thu, Mar 7, 2024 at 9:42 AM Miguel Ojeda
> > <miguel.ojeda.sandonis at gmail.com> wrote:
> > >
> > > Hi arm64/bpf/pci,
> > >
> > > In today's next-20240307 with a defconfig LLVM=1 I am seeing [1] under
> > > QEMU virt, i.e. from
> > > https://lore.kernel.org/all/20240305030516.41519-2-alexei.starovoitov@gmail.com/
> > > applied to the bpf-next tree.
> > >
> > > Cheers,
> > > Miguel
> > >
> > > [1]
> > >
> > > [    0.425177] pci-host-generic 4010000000.pcie: host bridge
> > > /pcie at 10000000 ranges:
> > > [    0.425886] pci-host-generic 4010000000.pcie:       IO
> > > 0x003eff0000..0x003effffff -> 0x0000000000
> > > [    0.426534] pci-host-generic 4010000000.pcie:      MEM
> > > 0x0010000000..0x003efeffff -> 0x0010000000
> > > [    0.426764] pci-host-generic 4010000000.pcie:      MEM
> > > 0x8000000000..0xffffffffff -> 0x8000000000
> > > [    0.427324] ------------[ cut here ]------------
> > > [    0.427456] vm_area at addr ffffffffc0800000 is not marked as VM_IOREMAP
> > > [    0.427944] WARNING: CPU: 0 PID: 1 at mm/vmalloc.c:315
> > > ioremap_page_range+0x25c/0x2bc
> >
> > Great. Thanks for flagging.
> > Looks like this check found some misuse of ioremap_page_range.
> >
> > Note that without marking the address range as VM_IOREMAP
> > the vread_iter() will be bulk reading over IO and might
> > cause hard hangs and what not.
> > pci drivers need to mark their range as VM_IOREMAP.
> > That was the reason for the warning.
> >
> > I'll try to figure out which piece of code missed passing
> > VM_IOREMAP into vm_area.
> > I'm not familiar with pci, so help is greatly appreciated.
> 
> Ok. I think I figured it out.
> Please try the attached patch.

I tested the attached patch on Qcom SDM845 based dev board and it fixes the
issue. So,

Tested-by: Manivannan Sadhasivam <manivannan.sadhasivam at linaro.org>

On a side note, PCI is the only driver making use of this API. If I'm making
such an API change, I would've made sure that the callers are unaffected by the
change.

- Mani

-- 
மணிவண்ணன் சதாசிவம்