Re: [PATCH v8 2/2] lib: checksum: Use aligned accesses for ip_fast_csum and csum_ipv6_magic tests

[Guenter Roeck]

On 2/15/24 02:27, David Laight wrote:
> ...
> > It would be worthwhile tracking this down since there are
> > lots of unaligned data accesses (8-byte accesses on 4-byte aligned addresses)
> > when running the kernel in 64-bit mode.
>
> Hmmm....
> For performance reasons you really don't want any of them.
> The misaligned 64bit fields need an __attribute((aligned(4)) marker.
>
> If the checksum code can do them it really needs to detect
> and handle the misalignment.
>
> The misaligned trap handler probably ought to contain a
> warn_on_once() to dump stack on the first such error.
> They can then be fixed one at a time.

Unaligned LDD at unwind_once+0x4a8/0x5e0

Decoded:

Unaligned LDD at unwind_once (arch/parisc/kernel/unwind.c:212 arch/parisc/kernel/unwind.c:243 arch/parisc/kernel/unwind.c:371 arch/parisc/kernel/unwind.c:445)

Source:

static bool pc_is_kernel_fn(unsigned long pc, void *fn)
{
        return (unsigned long)dereference_kernel_function_descriptor(fn) == pc;
}

Disassembled:

 c38:   50 3c 00 00     ldd 0(r1),ret0
 c3c:   08 1b 02 44     copy dp,r4
 c40:   0f 80 10 da     ldd 0(ret0),r26    <--
 c44:   37 dd 3f a1     ldo -30(sp),ret1
 c48:   51 02 00 20     ldd 10(r8),rp
 c4c:   e8 40 f0 00     bve,l (rp),rp
 c50:   51 1b 00 30     ldd 18(r8),dp

Hmm, interesting. See below for a possible fix. Should I submit a patch ?

Thanks,
Guenter

---
diff --git a/arch/parisc/kernel/entry.S b/arch/parisc/kernel/entry.S
index ab23e61a6f01..1d2aab619466 100644
--- a/arch/parisc/kernel/entry.S
+++ b/arch/parisc/kernel/entry.S
@@ -772,6 +772,7 @@ ENTRY_CFI(_switch_to)
        bv      %r0(%r2)
        mtctl   %r25,%cr30

+       .align  8
 ENTRY(_switch_to_ret)
        mtctl   %r0, %cr0               /* Needed for single stepping */
        callee_rest
@@ -1702,6 +1703,7 @@ ENTRY_CFI(sys_rt_sigreturn_wrapper)
        LDREG   PT_GR28(%r1),%r28  /* reload original r28 for syscall_exit */
 ENDPROC_CFI(sys_rt_sigreturn_wrapper)

+       .align 8
 ENTRY(syscall_exit)
        /* NOTE: Not all syscalls exit this way.  rt_sigreturn will exit
         * via syscall_exit_rfi if the signal was received while the process