Re: [PATCH] x86/entry/64: Fix context tracking state warning when load_gs_index fails

[Andy Lutomirski]

On Sep 29, 2016 5:43 PM, "Wanpeng Li" <kernellwp@xxxxxxxxx> wrote:
>
> 2016-09-30 5:01 GMT+08:00 Andy Lutomirski <luto@xxxxxxxxxxxxxx>:
> > On Mon, Sep 26, 2016 at 4:49 AM, Wanpeng Li <kernellwp@xxxxxxxxx> wrote:
> >> From: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
> >>
> >>  WARNING: CPU: 0 PID: 3331 at arch/x86/entry/common.c:45 enter_from_user_mode+0x32/0x50
> >>  CPU: 0 PID: 3331 Comm: ldt_gdt_64 Not tainted 4.8.0-rc7+ #13
> >>  Call Trace:
> >>   dump_stack+0x99/0xd0
> >>   __warn+0xd1/0xf0
> >>   warn_slowpath_null+0x1d/0x20
> >>   enter_from_user_mode+0x32/0x50
> >>   error_entry+0x6d/0xc0
> >>   ? general_protection+0x12/0x30
> >>   ? native_load_gs_index+0xd/0x20
> >>   ? do_set_thread_area+0x19c/0x1f0
> >>   SyS_set_thread_area+0x24/0x30
> >>   do_int80_syscall_32+0x7c/0x220
> >>   entry_INT80_compat+0x38/0x50
> >>
> >> This can be reproduced by running the GS testcase of ldt_gdt test unit in
> >> selftests.
> >>
> >> do_int80_syscall_32() will call enter_form_user_mode() to convert context
> >> tracking state from user state to kernel state. The load_gs_index can fail
> >> with user gsbase, gsbase will be fixed up and proceed if this happen.
> >> However, enter_from_user_mode() will be called again in the fixed up path
> >> though it is context tracking kernel state currently.
> >>
> >> This patch fix it by just fixing up gsbase and telling lockdep that IRQs
> >> are off once load_gs_index failed with user gsbase.
> >>
> >> Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> >> Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> >> Cc: "H. Peter Anvin" <hpa@xxxxxxxxx>
> >> Cc: Andy Lutomirski <luto@xxxxxxxxxx>
> >> Cc: Borislav Petkov <bp@xxxxxxxxx>
> >> Signed-off-by: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
> >> ---
> >>  arch/x86/entry/entry_64.S | 4 +++-
> >>  1 file changed, 3 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
> >> index d172c61..dc1ec23 100644
> >> --- a/arch/x86/entry/entry_64.S
> >> +++ b/arch/x86/entry/entry_64.S
> >> @@ -1045,7 +1045,9 @@ ENTRY(error_entry)
> >>          * gsbase and proceed.  We'll fix up the exception and land in
> >>          * .Lgs_change's error handler with kernel gsbase.
> >>          */
> >> -       jmp     .Lerror_entry_from_usermode_swapgs
> >> +       SWAPGS
> >> +       TRACE_IRQS_OFF
> >
> > Let's make this more readable: can you change this to:
> >
> > SWAPGS
> > jmp .Lerror_entry_done
> >
> > and remove the .Lerror_entry_from_usermode_swapgs label as well?
>
> I will do this in v2. Btw, could you point out why the GS testcase in
> tools/testing/selftests/x86/ldt_gdt.c will #GP? SDM said that swapgs
> will #GP if CPL != 0, however, native_load_gs_index() is CPL == 0.

The fault is on MOV, not SWAPGS.

--Andy