Dear Christophe,
Am 07.05.21 um 10:59 schrieb Christophe Leroy:
Le 07/05/2021 à 10:42, Paul Menzel a écrit :
[+Andrey]
Am 07.05.21 um 10:31 schrieb Christophe Leroy:
Le 06/05/2021 à 21:32, Paul Menzel a écrit :
[corrected subject]
Am 06.05.21 um 21:31 schrieb Paul Menzel:
On the POWER8 system IBM S822LC, Linux 5.13+, built with USSAN,
logs the warning below.
```
[ 0.030091]
================================================================================
[ 0.030295] UBSAN: array-index-out-of-bounds in
arch/powerpc/kernel/legacy_serial.c:359:56
[ 0.030325] index -1 is out of range for type 'legacy_serial_info [8]'
[ 0.030350] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.12.0+ #2
[ 0.030360] Call Trace:
[ 0.030363] [c000000024f1bad0] [c0000000009f4330] dump_stack+0xc4/0x114
(unreliable)
[ 0.030386] [c000000024f1bb20] [c0000000009efed0] ubsan_epilogue+0x18/0x78
[ 0.030400] [c000000024f1bb80] [c0000000009efafc]
__ubsan_handle_out_of_bounds+0xac/0xd0
[ 0.030414] [c000000024f1bc20] [c000000001711588]
ioremap_legacy_serial_console+0x54/0x144
[ 0.030430] [c000000024f1bc70] [c0000000000123c0] do_one_initcall+0x60/0x2c0
[ 0.030444] [c000000024f1bd40] [c000000001704bc4]
kernel_init_freeable+0x19c/0x25c
[ 0.030458] [c000000024f1bda0] [c000000000012a2c] kernel_init+0x2c/0x180
[ 0.030471] [c000000024f1be10] [c00000000000d6ec]
ret_from_kernel_thread+0x5c/0x70
[ 0.030484]
================================================================================
[ 0.030641]
================================================================================
[ 0.030668] UBSAN: array-index-out-of-bounds in
arch/powerpc/kernel/legacy_serial.c:360:58
[ 0.030697] index -1 is out of range for type 'plat_serial8250_port [9]'
[ 0.030721] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.12.0+ #2
[ 0.030730] Call Trace:
[ 0.030733] [c000000024f1bad0] [c0000000009f4330] dump_stack+0xc4/0x114
(unreliable)
[ 0.030749] [c000000024f1bb20] [c0000000009efed0] ubsan_epilogue+0x18/0x78
[ 0.030762] [c000000024f1bb80] [c0000000009efafc]
__ubsan_handle_out_of_bounds+0xac/0xd0
[ 0.030775] [c000000024f1bc20] [c0000000017115a0]
ioremap_legacy_serial_console+0x6c/0x144
[ 0.030790] [c000000024f1bc70] [c0000000000123c0] do_one_initcall+0x60/0x2c0
[ 0.030802] [c000000024f1bd40] [c000000001704bc4]
kernel_init_freeable+0x19c/0x25c
[ 0.030816] [c000000024f1bda0] [c000000000012a2c] kernel_init+0x2c/0x180
[ 0.030829] [c000000024f1be10] [c00000000000d6ec]
ret_from_kernel_thread+0x5c/0x70
[ 0.030842]
================================================================================
```
The function is as follows, so when legacy_serial_console == -1 as in
your situation, the pointers are just not used.
static int __init ioremap_legacy_serial_console(void)
{
struct legacy_serial_info *info =
&legacy_serial_infos[legacy_serial_console];
struct plat_serial8250_port *port =
&legacy_serial_ports[legacy_serial_console];
void __iomem *vaddr;
if (legacy_serial_console < 0)
return 0;
...
}
When I look into the generated code (UBSAN not selected), we see the
verification and the bail-out is done prior to any calculation based
on legacy_serial_console.
00000000 <ioremap_legacy_serial_console>:
0: 94 21 ff e0 stwu r1,-32(r1)
4: 3d 20 00 00 lis r9,0
6: R_PPC_ADDR16_HA .data
8: 7c 08 02 a6 mflr r0
c: bf 81 00 10 stmw r28,16(r1)
10: 3b 80 00 00 li r28,0
14: 83 a9 00 00 lwz r29,0(r9)
16: R_PPC_ADDR16_LO .data
18: 90 01 00 24 stw r0,36(r1)
1c: 2c 1d 00 00 cmpwi r29,0
20: 41 80 00 80 blt a0 <ioremap_legacy_serial_console+0xa0>
So, is it normal that UBSAN reports an error here ?
If it’s useful, I could disassemble the code here. But please tell me
how.
Sorry, I do not know. I just selected the option, and saw the error.
Maybe Andrey has an idea.
No need for you to disassemble, I just wanted to show that without UBSAN
there is no problem with the index as it is used only after boundary
checking. (But if you want to do so, if is just an 'objdump -dr
legacy_serial.o')
Thank you for the hint.
Now, with UBSAN, I see that UBSAN does the verification of the index
earlier than expected. So what to do here, we can modify the code, but
that modification would just be to make UBSAN happy as there is no
problem in itself.
In #g...@irc.freenode.net I was told by zid (they weren’t so happy with
the wording), but maybe you understand it:
It's not legal C to generate pointers to things other than 0,
objects, or 1 past the end of an object, not just dereference them,
so technically that's not legal per the C spec.
In practice it won't matter until it's dereferenced of course unless
you're doing something weird, let's say.. instrumenting the code
Kind regards,
Paul
