Re: [v3][PATCH 2/3] ppc/kprobe: complete kprobe and migrate exception frame

On 09/11/2012 01:51 PM, Benjamin Herrenschmidt wrote:

On Tue, 2012-09-11 at 10:20 +0800, Tiejun Chen wrote:

quoted

We can't emulate stwu since that may corrupt current exception stack.
So we will have to do real store operation in the exception return code.

Firstly we'll allocate a trampoline exception frame below the kprobed
function stack and copy the current exception frame to the trampoline.
Then we can do this real store operation to implement 'stwu', and reroute
the trampoline frame to r1 to complete this exception migration.

Ok, so not quite there yet :-)

See below:

quoted

Signed-off-by: Tiejun Chen <redacted>
---
  arch/powerpc/kernel/entry_32.S |   45 ++++++++++++++++++++++++++++++++++------
  arch/powerpc/kernel/entry_64.S |   32 ++++++++++++++++++++++++++++
  2 files changed, 71 insertions(+), 6 deletions(-)

diff --git a/arch/powerpc/kernel/entry_32.S b/arch/powerpc/kernel/entry_32.S
index ead5016..6cfe12f 100644
--- a/arch/powerpc/kernel/entry_32.S
+++ b/arch/powerpc/kernel/entry_32.S

@@ -831,19 +831,54 @@ restore_user:
  	bnel-	load_dbcr0
  #endif

-#ifdef CONFIG_PREEMPT
  	b	restore

  /* N.B. the only way to get here is from the beq following ret_from_except. */
  resume_kernel:
+	/* check current_thread_info, _TIF_EMULATE_STACK_STORE */
+	CURRENT_THREAD_INFO(r9, r1)
+	lwz	r0,TI_FLAGS(r9)
+	andis.	r0,r0,_TIF_EMULATE_STACK_STORE@h
+	beq+	1f

So you used r0 to load the TI_FLAGS and immediately clobbered it in
andis. forcing you to re-load them later down. Instead, put them in r8

	lwz	r8,TI_FLAGS(r9)
	andis.	r0,r8,_TIF_*
	beq+	*

quoted

+	addi	r8,r1,INT_FRAME_SIZE	/* Get the kprobed function entry */

Then you put your entry in r8 ....

I'll update this for 32b and 64b sections.

quoted

+	lwz	r3,GPR1(r1)
+	subi	r3,r3,INT_FRAME_SIZE	/* dst: Allocate a trampoline exception frame */
+	mr	r4,r1			/* src:  current exception frame */
+	li	r5,INT_FRAME_SIZE	/* size: INT_FRAME_SIZE */
+	mr	r1,r3			/* Reroute the trampoline frame to r1 */
+	bl	memcpy			/* Copy from the original to the trampoline */

Which you just clobbered... oops :-)

So you need to store that old r1 somewhere fist then retrieve it
after the memcpy call. That or open-code the memcpy to avoid all
the clobbering problems.

Maybe we can use copy_and_flush() since looks copy_and_flush() only clobber r0, 
r6 and LR explicitly.

I'll resync these comments for v4.

Tiejun