Re: [PATCH] powerpc/ptrace: Add offsetof check for exit_flags in pt_regs_check()

Le 03/07/2026 à 10:55, Amit Machhiwal a écrit :

On 2026/07/03 01:19 PM, Mukesh Kumar Chaurasiya (IBM) wrote:

quoted

pt_regs_check() verifies that fields shared between struct pt_regs and
struct user_pt_regs sit at the same offset, to catch any accidental
layout divergence between the kernel and uapi structures.

Add the missing check for exit_flags, following the same pattern as the
existing checks for result, dsisr, dar and the other shared fields.

Fixes: d7a6797e0bc1 ("powerpc: add exit_flags field in pt_regs")

This commit was almost reverted by commit 263e5159e00a ("powerpc: Fix 
exit_flags field placement in pt_regs for ptrace")

Dropping this patch.

Christophe

You may want to add below in order to get this patch backported:

   Cc: stable@vger.kernel.org

quoted

Signed-off-by: Mukesh Kumar Chaurasiya (IBM) <redacted>
---
  arch/powerpc/kernel/ptrace/ptrace.c | 2 ++
  1 file changed, 2 insertions(+)

diff --git a/arch/powerpc/kernel/ptrace/ptrace.c b/arch/powerpc/kernel/ptrace/ptrace.c
index 316d4f5ead8e..fb3fec73f66a 100644
--- a/arch/powerpc/kernel/ptrace/ptrace.c
+++ b/arch/powerpc/kernel/ptrace/ptrace.c

@@ -235,6 +235,8 @@ void __init pt_regs_check(void)
  		     offsetof(struct user_pt_regs, dsisr));
  	BUILD_BUG_ON(offsetof(struct pt_regs, result) !=
  		     offsetof(struct user_pt_regs, result));
+	BUILD_BUG_ON(offsetof(struct pt_regs, exit_flags) !=
+		     offsetof(struct user_pt_regs, exit_flags));

At first, it appeared to be redundant as we already check

	CHECK_REG(PT_EXIT_FLAGS, exit_flags);

which expands to:

	#define CHECK_REG(_pt, _reg) \
		BUILD_BUG_ON(_pt != (offsetof(struct user_pt_regs, _reg) / \
				     sizeof(unsigned long)));

But looking closely, it turns out to be non-redundant. CHECK_REG only
touches struct user_pt_regs — it validates the PT_EXIT_FLAGS constant
against the uapi layout. It says nothing about struct pt_regs.

struct pt_regs (asm/ptrace.h) embeds struct user_pt_regs in a union
alongside an anonymous struct that re-declares all the same fields with
kernel-internal aliases (dear/dar, esr/dsisr). exit_flags lives in that
anonymous struct. A field accidentally inserted there between result and
exit_flags would silently break the union aliasing without CHECK_REG
catching it. The new BUILD_BUG_ON closes that gap, consistent with how
result, dar and dsisr are already protected in the same block.

Looks good to me.

Reviewed-by: Amit Machhiwal <redacted>

Thanks,
Amit

quoted

  
  	BUILD_BUG_ON(sizeof(struct user_pt_regs) > sizeof(struct pt_regs));
  
-- 
2.55.0