On 20/02/2024 16.09, Arnd Bergmann wrote:
On Tue, Feb 20, 2024, at 15:13, Greg Ungerer wrote:
On 20/2/24 02:01, Thomas Huth wrote:
We should not use any CONFIG switches in uapi headers since these
only work during kernel compilation; they are not defined for
userspace. Fix it by moving the struct pt_regs to the kernel-internal
header instead - struct pt_regs does not seem to be required for
the userspace headers on m68k at all.
Suggested-by: Greg Ungerer <gerg@xxxxxxxxxxxxxx>
Signed-off-by: Thomas Huth <thuth@xxxxxxxxxx>
---
v2: Move the struct instead of changing the #ifdef
See previous discussion here:
https://lore.kernel.org/lkml/6e3f2a2e-2430-4b4f-9ead-d9a4d5e42713@xxxxxxxxxxxxxx/
I am fine with this. FWIW the following architectures do
not define pt_regs in their uapi/ptrace.h header either:
arc, arm64, loongarch, nios2, openrisc, riscv, s390, xtensa
Though quite a few of them have a user_pt_regs instead.
So for me:
Acked-by: Greg Ungerer <gerg@xxxxxxxxxxxxxx>
Geert, Arnd, do you have any thoughts on this?
It clearly doesn't change the ABI, so that part is fine.
If asm/ptrace.h is included by some userspace tool to
get the definition, it might cause a compile-time error
that needs a trivial source change.
This could be needed for ptrace (gdb, strace) or signal
handling and setjmp (libc), though it's more likely that these
already have their own copies.
If we still feel unsure, we should maybe rather go with v1:
https://lore.kernel.org/lkml/20231110103120.387517-1-thuth@xxxxxxxxxx/
?
Thomas
