Re: [PATCH v2] x86/mm: Refuse W^X violations

[Guenter Roeck]

On 9/21/22 13:59, Dave Hansen wrote:
> On 9/21/22 13:07, Guenter Roeck wrote:
> > [    2.042861] CPA refuse W^X violation: 8000000000000063 -> 0000000000000063 range: 0x00000000c00a0000 - 0x00000000c00a0fff PFN a0
> > ILLOPC: cbc65efa: 0f 0b
> > [    2.043267] WARNING: CPU: 0 PID: 1 at arch/x86/mm/pat/set_memory.c:600 __change_page_attr_set_clr+0xdca/0xdd0
> ...
> > [    2.050307] ---[ end trace 0000000000000000 ]---
> > [    2.050762] PCI: PCI BIOS area is rw and x. Use pci=nobios if you want it NX.
> > [    2.051115] kernel tried to execute NX-protected page - exploit attempt? (uid: 0)
> > [    2.051115] BUG: unable to handle page fault for address: c00fd2bf
>
> This _looks_ like it is working as intended.  The PCI BIOS code tried to
> make a RWX page.  The CPA code refused to do it and presumably returned
> an error, leaving a RW page, non-executable page.  The PCI code didn't
> check the set_memory_x() return code and tried to go execute anyway.
> That resulted in the oops.
>
> I was able to reproduce this pretty easily.  The workaround from dmesg
> is pci=nobios.  That seems to do the trick for me, although that advise
> was sandwiched between a warning and an oops, so not the easiest to find.
>
> I'm a bit torn what to do on this one.  Breaking the boot is bad, but so
> is leaving RWX memory around.
>
> Thoughts?

For my part I'll do what the above suggests, ie run tests with PAE enabled
with pci=nobios command line option. AFAICS that hides the problem in my tests.
I am just not sure if that is really appropriate.

Guenter