Re: [PATCH v4] PPC: use CURRENT_THREAD_INFO instead of open coded assembly

On Fri, 2012-07-13 at 08:45 +1000, Paul Mackerras wrote:

On Thu, Jul 05, 2012 at 09:41:35AM -0500, Stuart Yoder wrote:

quoted

diff --git a/arch/powerpc/kernel/entry_32.S b/arch/powerpc/kernel/entry_32.S
index ba3aeb4..bad42e3 100644
--- a/arch/powerpc/kernel/entry_32.S
+++ b/arch/powerpc/kernel/entry_32.S

@@ -92,7 +92,7 @@ crit_transfer_to_handler:
 	mfspr	r8,SPRN_SPRG_THREAD
 	lwz	r0,KSP_LIMIT(r8)
 	stw	r0,SAVED_KSP_LIMIT(r11)
-	rlwimi	r0,r1,0,0,(31-THREAD_SHIFT)
+	CURRENT_THREAD_INFO(r0, r1)
 	stw	r0,KSP_LIMIT(r8)
 	/* fall through */
 #endif

@@ -112,7 +112,7 @@ crit_transfer_to_handler:
 	mfspr	r8,SPRN_SPRG_THREAD
 	lwz	r0,KSP_LIMIT(r8)
 	stw	r0,saved_ksp_limit@l(0)
-	rlwimi	r0,r1,0,0,(31-THREAD_SHIFT)
+	CURRENT_THREAD_INFO(r0, r1)
 	stw	r0,KSP_LIMIT(r8)
 	/* fall through */
 #endif

Do you really mean to replace a rlwimi with a rlwinm?  If so, is that
because the rlwinm is a bug fix, or is it because you know something
special about KSP_LIMIT(r8) which means that rlwinm and rlwimi are
equivalent here?

Ah that's an interesting one I hadn't spotted when reviewing. Both
variants (rlwimi and rlwinm)  will effectively replace the top bits of
KSP_LIMIT, switching it to the current stack.

The difference is that the original one (rlwimi) will preserve the
bottom bits.

Now, do we want those bottom bits ? It looks like we do if we want to
make the thread_info at the bottom of the stack as "out of bounds".

So the patch breaks that. It will not generally break a working kernel
but the stack overflow detection may not trigger if the overflow is just
enough to override the thread infos.

Stuart, I already applied v4 of the patch to powerpc-next and I'd rather
not rebase it, can you send a fixup patch please, one that will
effecticely revert those two hunks, we can leave that open coded. While
at it, please add a comment explaining what the code does to avoid
similar confusion in the future.

Cheers,
Ben.