Re: ptrace single-stepping change breaks Wine

[Jesse Allen]

On Wed, 29 Dec 2004 10:53:54 -0800 (PST), Linus Torvalds
<torvalds@xxxxxxxx> wrote:
> Ok, I don't remember the context from the Wine lists (and it's not clear
> from the older emails I was cc'd on), so the "#3 signal.c" change
> description is a bit too vague. Jesse, willing to just point to the exact
> diff that you need to make Warcraft work for you (and then maybe Thomas
> Sailer can verify whether that part is indeed the one that causes him
> problems).

I have attached the diff attached in this message for the lkml.

> 
> The code in question now does
> 
>         /*
>          * Iff TF was set because the program is being single-stepped by a
>          * debugger, don't save that information on the signal stack.. We
>          * don't want debugging to change state.
>          */
>         eflags = regs->eflags;
>         if (current->ptrace & PT_DTRACE)
>                 eflags &= ~TF_MASK;
>         err |= __put_user(eflags, &sc->eflags);
> 
> and I guess it originally never cleared it. True?

Yes.

> 
> So does removing the conditional TF clear make everything work again?
> 

Yes, as long as TIF_SINGLESTEP is not set in set_singlestep(). 
set_singlestep also sets PT_DTRACE, so as it now is, a call to the
set_singlestep function will make this condition true clearing TF when
run.  So both the conditional TF clear and setting TIF_SINGLESTEP
needs to be removed, like I show in the diff.   Making these changes
returns the code to a 2.6.8-ish resemblence.

For the wine people, I will try to upload the seh debug channel logs
as soon as possible.

Jesse
diff -u linux/arch/i386/kernel/ptrace.c linux-mod/arch/i386/kernel/ptrace.c
--- linux/arch/i386/kernel/ptrace.c	2004-12-09 15:24:07.000000000 -0700
+++ linux-mod/arch/i386/kernel/ptrace.c	2004-12-25 16:09:52.000000000 -0700
@@ -142,7 +142,7 @@
 {
 	long eflags;
 
-	set_tsk_thread_flag(child, TIF_SINGLESTEP);
+//	set_tsk_thread_flag(child, TIF_SINGLESTEP);
 	eflags = get_stack_long(child, EFL_OFFSET);
 	put_stack_long(child, EFL_OFFSET, eflags | TRAP_FLAG);
 	child->ptrace |= PT_DTRACE;
@@ -153,7 +153,7 @@
 	if (child->ptrace & PT_DTRACE) {
 		long eflags;
 
-		clear_tsk_thread_flag(child, TIF_SINGLESTEP);
+//		clear_tsk_thread_flag(child, TIF_SINGLESTEP);
 		eflags = get_stack_long(child, EFL_OFFSET);
 		put_stack_long(child, EFL_OFFSET, eflags & ~TRAP_FLAG);
 		child->ptrace &= ~PT_DTRACE;
diff -u linux/arch/i386/kernel/signal.c linux-mod/arch/i386/kernel/signal.c
--- linux/arch/i386/kernel/signal.c	2004-12-09 15:24:07.000000000 -0700
+++ linux-mod/arch/i386/kernel/signal.c	2004-12-25 16:10:10.000000000 -0700
@@ -299,8 +299,8 @@
 	 * don't want debugging to change state.
 	 */
 	eflags = regs->eflags;
-	if (current->ptrace & PT_DTRACE)
-		eflags &= ~TF_MASK;
+//	if (current->ptrace & PT_DTRACE)
+//		eflags &= ~TF_MASK;
 	err |= __put_user(eflags, &sc->eflags);
 	err |= __put_user(regs->esp, &sc->esp_at_signal);
 	err |= __put_user(regs->xss, (unsigned int __user *)&sc->ss);