powerpc/lib: Avoid array bounds warnings in vec ops
authorMichael Ellerman <mpe@ellerman.id.au>
Mon, 20 Nov 2023 23:54:36 +0000 (10:54 +1100)
committerMichael Ellerman <mpe@ellerman.id.au>
Mon, 27 Nov 2023 11:05:14 +0000 (22:05 +1100)
Building with GCC with -Warray-bounds enabled there are several warnings
in sstep.c along the lines of:

  In function ‘do_byte_reverse’,
      inlined from ‘do_vec_load’ at arch/powerpc/lib/sstep.c:691:3,
      inlined from ‘emulate_loadstore’ at arch/powerpc/lib/sstep.c:3439:9:
  arch/powerpc/lib/sstep.c:289:23: error: array subscript 2 is outside array bounds of ‘u8[16]’ {aka ‘unsigned char[16]’} [-Werror=array-bounds=]
    289 |                 up[2] = byterev_8(up[1]);
        |                 ~~~~~~^~~~~~~~~~~~~~~~~~
  arch/powerpc/lib/sstep.c: In function ‘emulate_loadstore’:
  arch/powerpc/lib/sstep.c:681:11: note: at offset 16 into object ‘u’ of size 16
    681 |         } u = {};
        |           ^

do_byte_reverse() supports a size up to 32 bytes, but in these cases the
caller is only passing a 16 byte buffer. In practice there is no bug,
do_vec_load() is only called from the LOAD_VMX case in emulate_loadstore().
That in turn is only reached when analyse_instr() recognises VMX ops,
and in all cases the size is no greater than 16:

  $ git grep -w LOAD_VMX arch/powerpc/lib/sstep.c
  arch/powerpc/lib/sstep.c:                        op->type = MKOP(LOAD_VMX, 0, 1);
  arch/powerpc/lib/sstep.c:                        op->type = MKOP(LOAD_VMX, 0, 2);
  arch/powerpc/lib/sstep.c:                        op->type = MKOP(LOAD_VMX, 0, 4);
  arch/powerpc/lib/sstep.c:                        op->type = MKOP(LOAD_VMX, 0, 16);

Similarly for do_vec_store().

Although the warning is incorrect, the code would be safer if it clamped
the size from the caller to the known size of the buffer. Do that using
min_t().

Reported-by: Bagas Sanjaya <bagasdotme@gmail.com>
Closes: https://lore.kernel.org/linuxppc-dev/YpbUcPrm61RLIiZF@debian.me/
Reported-by: Jan-Benedict Glaw <jbglaw@lug-owl.de>
Closes: https://lore.kernel.org/linuxppc-dev/20221212215117.aa7255t7qd6yefk4@lug-owl.de/
Reported-by: "Gustavo A. R. Silva" <gustavo@embeddedor.com>
Closes: https://lore.kernel.org/linuxppc-dev/6a8bf78c-aedb-4d5a-b0aa-82a51a17b884@embeddedor.com/
Reviewed-by: "Gustavo A. R. Silva" <gustavoars@kernel.org>
Build-tested-by: "Gustavo A. R. Silva" <gustavoars@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://msgid.link/20231120235436.1569255-1-mpe@ellerman.id.au
arch/powerpc/lib/sstep.c

index a4ab862..a13f05c 100644 (file)
@@ -688,7 +688,7 @@ static nokprobe_inline int do_vec_load(int rn, unsigned long ea,
        if (err)
                return err;
        if (unlikely(cross_endian))
-               do_byte_reverse(&u.b[ea & 0xf], size);
+               do_byte_reverse(&u.b[ea & 0xf], min_t(size_t, size, sizeof(u)));
        preempt_disable();
        if (regs->msr & MSR_VEC)
                put_vr(rn, &u.v);
@@ -719,7 +719,7 @@ static nokprobe_inline int do_vec_store(int rn, unsigned long ea,
                u.v = current->thread.vr_state.vr[rn];
        preempt_enable();
        if (unlikely(cross_endian))
-               do_byte_reverse(&u.b[ea & 0xf], size);
+               do_byte_reverse(&u.b[ea & 0xf], min_t(size_t, size, sizeof(u)));
        return copy_mem_out(&u.b[ea & 0xf], ea, size, regs);
 }
 #endif /* CONFIG_ALTIVEC */