Re: [PATCH 11/34] x86/entry/32: Handle Entry from Kernel-Mode on Entry-Stack

[Brian Gerst]

On Mon, Mar 5, 2018 at 1:25 PM, Joerg Roedel <joro@xxxxxxxxxx> wrote:
> Hi Brian,
>
> thanks for your review and helpful input.
>
> On Mon, Mar 05, 2018 at 11:41:01AM -0500, Brian Gerst wrote:
>> On Mon, Mar 5, 2018 at 5:25 AM, Joerg Roedel <joro@xxxxxxxxxx> wrote:
>> > +.Lentry_from_kernel_\@:
>> > +
>> > +       /*
>> > +        * This handles the case when we enter the kernel from
>> > +        * kernel-mode and %esp points to the entry-stack. When this
>> > +        * happens we need to switch to the task-stack to run C code,
>> > +        * but switch back to the entry-stack again when we approach
>> > +        * iret and return to the interrupted code-path. This usually
>> > +        * happens when we hit an exception while restoring user-space
>> > +        * segment registers on the way back to user-space.
>> > +        *
>> > +        * When we switch to the task-stack here, we can't trust the
>> > +        * contents of the entry-stack anymore, as the exception handler
>> > +        * might be scheduled out or moved to another CPU. Therefore we
>> > +        * copy the complete entry-stack to the task-stack and set a
>> > +        * marker in the iret-frame (bit 31 of the CS dword) to detect
>> > +        * what we've done on the iret path.
>>
>> We don't need to worry about preemption changing the entry stack.  The
>> faults that IRET or segment loads can generate just run the exception
>> fixup handler and return.  Interrupts were disabled when the fault
>> occurred, so the kernel cannot be preempted.  The other case to watch
>> is #DB on SYSENTER, but that simply returns and doesn't sleep either.
>>
>> We can keep the same process as the existing debug/NMI handlers -
>> leave the current exception pt_regs on the entry stack and just switch
>> to the task stack for the call to the handler.  Then switch back to
>> the entry stack and continue.  No copying needed.
>
> Okay, I'll look into that. Will it even be true for fully preemptible
> and RT kernels that there can't be any preemption of these handlers?

See resume_kernel in the 32-bit entry for how preemption is handled on
return to kernel mode.  Looking at the RT patches, they still respect
disabling interrupts also disabling preemption.

--
Brian Gerst