Re: [PATCH] io_thread/x86: don't reset 'cs', 'ss', 'ds' and 'es' registers for io_threads

[Stefan Metzmacher]

Am 04.05.21 um 01:16 schrieb Linus Torvalds:
> On Mon, May 3, 2021 at 3:56 PM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>>
>> It's all fine that we have lots of blurb about GDB, but there is no
>> reasoning why this does not affect regular kernel threads which take the
>> same code path.
> 
> Actual kernel threads don't get attached to by ptrace.
> 
>> This is a half setup user space thread which is assumed to behave like a
>> regular kernel thread, but is this assumption actually true?
> 
> No, no.
> 
> It's a *fully set up USER thread*.
> 
> Those IO threads used to be kernel threads. That didn't work out for
> the reasons already mentioned earlier.
> 
> These days they really are fully regular user threads, they just don't
> return to user space because they continue to do the IO work that they
> were created for.
> 
> Maybe instead of Stefan's patch, we could do something like this:
> 
>    diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
>    index 43cbfc84153a..890f3992e781 100644
>    --- a/arch/x86/kernel/process.c
>    +++ b/arch/x86/kernel/process.c
>    @@ -156,7 +156,7 @@ int copy_thread(unsigned long clone_flags,
> unsigned long sp, unsigned long arg,
>     #endif
> 
>         /* Kernel thread ? */
>    -    if (unlikely(p->flags & (PF_KTHREAD | PF_IO_WORKER))) {
>    +    if (unlikely(p->flags & PF_KTHREAD)) {
>                 memset(childregs, 0, sizeof(struct pt_regs));
>                 kthread_frame_init(frame, sp, arg);
>                 return 0;
>    @@ -168,6 +168,17 @@ int copy_thread(unsigned long clone_flags,
> unsigned long sp, unsigned long arg,
>         if (sp)
>                 childregs->sp = sp;
> 
>    +    /*
>    +     * An IO thread is a user space thread, but it doesn't
>    +     * return to ret_after_fork(), it does the same kernel
>    +     * frame setup to return to a kernel function that
>    +     * a kernel thread does.
>    +     */
>    +    if (unlikely(p->flags & PF_IO_WORKER)) {
>    +            kthread_frame_init(frame, sp, arg);
>    +            return 0;
>    +    }
>    +
>     #ifdef CONFIG_X86_32
>         task_user_gs(p) = get_user_gs(current_pt_regs());
>     #endif
> 
> does that clarify things and make people happier?
> 
> Maybe the compiler might even notice that the
> 
>                 kthread_frame_init(frame, sp, arg);
>                 return 0;
> 
> part is common code and then it will result in less generated code too.
> 
> NOTE! The above is - as usual - COMPLETELY UNTESTED. It looks obvious
> enough, and it builds cleanly. But that's all I'm going to guarantee.

I think I also tested something similar, see:

https://git.samba.org/?p=metze/linux/wip.git;a=commitdiff;h=82fcee2774add04fbc0e4755c405e6c0b7467e3a

If I remember correctly gdb showed bogus addresses for the backtraces of the io_threads,
as some regs where not cleared.

The patch I posted shows this instead:

  Thread 2 (LWP 8744):
  #0  0x0000000000000000 in ?? ()
  Backtrace stopped: Cannot access memory at address 0x0

I think that's a saner behavior.

However splitting the if statements might be a good idea to make things
more clear.

Thanks discussing this again!
metze