Messages in this thread | | | Date | Sun, 27 Aug 2006 09:36:46 -0700 | From | Jeremy Fitzhardinge <> | Subject | Re: [PATCH RFC 3/6] Use %gs as the PDA base-segment in the kernel. |
| |
Andi Kleen wrote: >> >> - /* Clear %fs and %gs. */ >> - asm volatile ("movl %0, %%fs; movl %0, %%gs" : : "r" (0)); >> + /* Clear %fs. */ >> + asm volatile ("mov %0, %%fs" : : "r" (0)); >> + >> + /* Set %gs for this CPU's PDA */ >> + asm volatile ("mov %0, %%gs" : : "r" (__KERNEL_PDA)); >> > > I would add memory clobbers here to make sure the dependency on read/write pda > is right. >
Yep. And the "m" args in the pda asm isn't quite right for rmw PDA ops (not that there are any at the moment).
>> +1: movw GS(%esp), %gs >> > > movl is recommended in 32bit mode >
OK. I thought the assembler objected to me about it.
>> --- a/arch/i386/kernel/signal.c >> +++ b/arch/i386/kernel/signal.c >> @@ -128,7 +128,7 @@ restore_sigcontext(struct pt_regs *regs, >> X86_EFLAGS_TF | X86_EFLAGS_SF | X86_EFLAGS_ZF | \ >> X86_EFLAGS_AF | X86_EFLAGS_PF | X86_EFLAGS_CF) >> >> - GET_SEG(gs); >> + COPY_SEG(gs); >> GET_SEG(fs); >> COPY_SEG(es); >> COPY_SEG(ds); >> @@ -244,9 +244,7 @@ setup_sigcontext(struct sigcontext __use >> { >> int tmp, err = 0; >> >> - tmp = 0; >> - savesegment(gs, tmp); >> - err |= __put_user(tmp, (unsigned int __user *)&sc->gs); >> + err |= __put_user(regs->xgs, (unsigned int __user *)&sc->gs); >> savesegment(fs, tmp); >> err |= __put_user(tmp, (unsigned int __user *)&sc->fs); >> > > Hmm, changing it for the sc looks a bit bogus. If everything > is right nothing should change for user space, but this changes something. >
The sigcontext contains the userspace register state at the time of the signal. Since userspace %gs is stored in the on-stack pt_regs, that should be where it fetches it from to fill out the sigcontext, rather than the kernel's internal value of %gs - in other words, it should be the same as ds and es. Or am I missing something?
>> @@ -306,7 +306,7 @@ static void do_sys_vm86(struct kernel_vm >> tsk->thread.screen_bitmap = info->screen_bitmap; >> if (info->flags & VM86_SCREEN_BITMAP) >> mark_screen_rdonly(tsk->mm); >> - __asm__ __volatile__("xorl %eax,%eax; movl %eax,%fs; movl %eax,%gs\n\t"); >> + __asm__ __volatile__("movl %0,%%fs\n\t" : : "r" (0)); >> > > This is actually a useful bug fix on its own. >
Yep. But there seems to be some other very dubious code in there as well (the asm("mov %%eax,%0" : "=r" (eax)) sequence). I was wondering about what it all does...
J - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |