ARM Highbank Crashes Realted to GIC

[Alexander Bulekov]

Hello,
Here are some QTest reproducers for crashes on ARM Highbank that all seem to be related to the gic device.

Reproducer 1:
cat << EOF | ./arm-softmmu/qemu-system-arm -machine highbank \
-nographic -monitor none -serial none -qtest stdio
writel 0xfff11f00 0x8405f559
writel 0xfff117fd 0x5c057bd8
EOF

==10595==ERROR: AddressSanitizer: SEGV on unknown address 0x62b000013e01 (pc 0x55b6ab85cc91 bp 0x7fff60bd4d70 sp 0x7fff60bd4ce0 T0)
==10595==The signal is caused by a READ memory access.
    #0 0x55b6ab85cc91 in gic_get_current_cpu /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:60:12
    #1 0x55b6ab85e1bd in gic_dist_writeb /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:1182:11
    #2 0x55b6ab855a97 in gic_dist_write /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:1514:9
    #3 0x55b6aa1650d4 in memory_region_write_with_attrs_accessor /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:503:12
    #4 0x55b6aa163ac6 in access_with_adjusted_size /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:544:18
    #5 0x55b6aa161f35 in memory_region_dispatch_write /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:1473:13
    #6 0x55b6a9313949 in flatview_write_continue /home/alxndr/Development/qemu/general-fuzz/exec.c:3176:23
    #7 0x55b6a92fca11 in flatview_write /home/alxndr/Development/qemu/general-fuzz/exec.c:3216:14
    #8 0x55b6a92fc54e in address_space_write /home/alxndr/Development/qemu/general-fuzz/exec.c:3308:18
=================================================================

Reproducer 2:
cat << EOF | ./arm-softmmu/qemu-system-arm -machine highbank \
-nographic -monitor none -serial none -qtest stdio
writeq 0xfff11f00 0x613a650f0fda6555
EOF

==1375==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x608000001c80 at pc 0x5618928c486e bp 0x7ffe22c4ee10 sp 0x7ffe22c4ee08
READ of size 8 at 0x608000001c80 thread T0
    #0 0x5618928c486d in address_space_translate_iommu /home/alxndr/Development/qemu/general-fuzz/exec.c:451:23
    #1 0x561892850acc in flatview_do_translate /home/alxndr/Development/qemu/general-fuzz/exec.c:524:16
    #2 0x5618928514ad in flatview_translate /home/alxndr/Development/qemu/general-fuzz/exec.c:584:15
    #3 0x5618928b1e14 in flatview_write_continue /home/alxndr/Development/qemu/general-fuzz/exec.c:3199:14
    #4 0x56189289aa11 in flatview_write /home/alxndr/Development/qemu/general-fuzz/exec.c:3216:14
    #5 0x56189289a54e in address_space_write /home/alxndr/Development/qemu/general-fuzz/exec.c:3308:18
    #6 0x5618937a5e13 in qtest_process_command /home/alxndr/Development/qemu/general-fuzz/softmmu/qtest.c:452:13
    #7 0x56189379d89f in qtest_process_inbuf /home/alxndr/Development/qemu/general-fuzz/softmmu/qtest.c:710:9
    #8 0x56189379c680 in qtest_read /home/alxndr/Development/qemu/general-fuzz/softmmu/qtest.c:722:5
=================================================================

Reproducer 3:
cat << EOF | ./arm-softmmu/qemu-system-arm -machine highbank \
-nographic -monitor none -serial none -qtest stdio
writeq 0xfff11000 0x700000b
writeq 0xfff11f00 0x4f4f4fff54a7afaf
writel 0xfff10100 0x600001ff
EOF

==23743==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x62b000006a92 at pc 0x55d690d980e1 bp 0x7ffe606082d0 sp 0x7ffe606082c8
READ of size 1 at 0x62b000006a92 thread T0
    #0 0x55d690d980e0 in gic_get_best_irq /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:94:13
    #1 0x55d690d9485b in gic_update_internal /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:185:13
    #2 0x55d690d90376 in gic_update /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:226:5
    #3 0x55d690dc0879 in gic_cpu_write /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:1758:9
    #4 0x55d690da41c0 in gic_thiscpu_write /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:1777:12
    #5 0x55d68f6b30d4 in memory_region_write_with_attrs_accessor /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:503:12
    #6 0x55d68f6b1ac6 in access_with_adjusted_size /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:544:18
    #7 0x55d68f6aff35 in memory_region_dispatch_write /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:1473:13
    #8 0x55d68e861949 in flatview_write_continue /home/alxndr/Development/qemu/general-fuzz/exec.c:3176:23
    #9 0x55d68e84aa11 in flatview_write /home/alxndr/Development/qemu/general-fuzz/exec.c:3216:14
    #10 0x55d68e84a54e in address_space_write /home/alxndr/Development/qemu/general-fuzz/exec.c:3308:18
    #11 0x55d68f755537 in qtest_process_command /home/alxndr/Development/qemu/general-fuzz/softmmu/qtest.c:447:13
    #12 0x55d68f74d89f in qtest_process_inbuf /home/alxndr/Development/qemu/general-fuzz/softmmu/qtest.c:710:9
    #13 0x55d68f74c680 in qtest_read /home/alxndr/Development/qemu/general-fuzz/softmmu/qtest.c:722:5
    #14 0x55d692dddc36 in qemu_chr_be_write_impl /home/alxndr/Development/qemu/general-fuzz/chardev/char.c:188:9
    #15 0x55d692dddd79 in qemu_chr_be_write /home/alxndr/Development/qemu/general-fuzz/chardev/char.c:200:9
    #16 0x55d692df105e in fd_chr_read /home/alxndr/Development/qemu/general-fuzz/chardev/char-fd.c:68:9
    #17 0x55d692f395df in qio_channel_fd_source_dispatch /home/alxndr/Development/qemu/general-fuzz/io/channel-watch.c:84:12
    #18 0x7f69a1b50897 in g_main_context_dispatch (/usr/lib/x86_64-linux-gnu/libglib-2.0.so.0+0x4e897)
    #19 0x55d6932f5c83 in glib_pollfds_poll /home/alxndr/Development/qemu/general-fuzz/util/main-loop.c:217:9
    #20 0x55d6932f35b6 in os_host_main_loop_wait /home/alxndr/Development/qemu/general-fuzz/util/main-loop.c:240:5
    #21 0x55d6932f2f97 in main_loop_wait /home/alxndr/Development/qemu/general-fuzz/util/main-loop.c:516:11
    #22 0x55d68f76c62d in qemu_main_loop /home/alxndr/Development/qemu/general-fuzz/softmmu/vl.c:1676:9
    #23 0x55d692f6f20c in main /home/alxndr/Development/qemu/general-fuzz/softmmu/main.c:49:5
    #24 0x7f69a06d6e0a in __libc_start_main /build/glibc-GwnBeO/glibc-2.30/csu/../csu/libc-start.c:308:16
    #25 0x55d68e753459 in _start (/home/alxndr/Development/qemu/general-fuzz/build/arm-softmmu/qemu-system-arm+0x3254459)

0x62b000006a92 is located 2 bytes to the right of 26768-byte region [0x62b000000200,0x62b000006a90)
allocated by thread T0 here:
    #0 0x55d68e7cbe4d in malloc (/home/alxndr/Development/qemu/general-fuzz/build/arm-softmmu/qemu-system-arm+0x32cce4d)
    #1 0x7f69a1b56500 in g_malloc (/usr/lib/x86_64-linux-gnu/libglib-2.0.so.0+0x54500)
    #2 0x55d69254f231 in object_new /home/alxndr/Development/qemu/general-fuzz/qom/object.c:708:12
    #3 0x55d69034bf01 in qdev_new /home/alxndr/Development/qemu/general-fuzz/hw/core/qdev.c:136:12
    #4 0x55d68f2b7aa4 in calxeda_init /home/alxndr/Development/qemu/general-fuzz/hw/arm/highbank.c:319:15
    #5 0x55d68f2b6466 in highbank_init /home/alxndr/Development/qemu/general-fuzz/hw/arm/highbank.c:411:5
    #6 0x55d6903d43f1 in machine_run_board_init /home/alxndr/Development/qemu/general-fuzz/hw/core/machine.c:1134:5
    #7 0x55d68f77e0ee in qemu_init /home/alxndr/Development/qemu/general-fuzz/softmmu/vl.c:4356:5
    #8 0x55d692f6f207 in main /home/alxndr/Development/qemu/general-fuzz/softmmu/main.c:48:5
    #9 0x7f69a06d6e0a in __libc_start_main /build/glibc-GwnBeO/glibc-2.30/csu/../csu/libc-start.c:308:16

Let me know if I can provide any further info.
-Alex

[Philippe Mathieu-Daudé (philmd)]

Why put all these bugs in the same ticket?

For reproducer #2:

writeq 0xfff11f00 0x613a650f0fda6555 does:

gic_dist_write dist write at 0x00000f00 size 4: 0x0fda6555

0x0fda6555 => IRQ 341, mask type 3 illegal -> DPRINTF("Bad Soft Int target filter\n");

mask = ALL_CPU_MASK = 0xff

Having:

#define GIC_NR_SGIS 16
uint8_t sgi_pending[GIC_NR_SGIS][GIC_NCPU];

s->sgi_pending[irq][target_cpu] |= (1 << cpu);
               ^^^
                  \ OOB access.

[Peter Maydell (pmaydell)]

I haven't looked at reproducer #1, but is it a fuzzer-specific variant of LP:1602247 (trying to read the "for this CPU" registers from something other than a CPU doesn't work) ?

[Alexander Bulekov (a1xndr)]

On 200730 1531, Philippe Mathieu-Daudé wrote:
> Why put all these bugs in the same ticket?

Thought they might have a similar root cause, though that is evidently
wrong..

> For reproducer #2:
>
> writeq 0xfff11f00 0x613a650f0fda6555 does:
>
> gic_dist_write dist write at 0x00000f00 size 4: 0x0fda6555
>
> 0x0fda6555 => IRQ 341, mask type 3 illegal -> DPRINTF("Bad Soft Int
> target filter\n");
>
> mask = ALL_CPU_MASK = 0xff
>
> Having:
>
> #define GIC_NR_SGIS 16
> uint8_t sgi_pending[GIC_NR_SGIS][GIC_NCPU];
>
> s->sgi_pending[irq][target_cpu] |= (1 << cpu);
> ^^^
> \ OOB access.
>
> ** Changed in: qemu
> Status: New => Confirmed
>
> ** Tags added: arm
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/1889621
>
> Title:
> ARM Highbank Crashes Realted to GIC
>
> Status in QEMU:
> Confirmed
>
> Bug description:
> Hello,
> Here are some QTest reproducers for crashes on ARM Highbank that all seem to be related to the gic device.
>
> Reproducer 1:
> cat << EOF | ./arm-softmmu/qemu-system-arm -machine highbank \
> -nographic -monitor none -serial none -qtest stdio
> writel 0xfff11f00 0x8405f559
> writel 0xfff117fd 0x5c057bd8
> EOF
>
> ==10595==ERROR: AddressSanitizer: SEGV on unknown address 0x62b000013e01 (pc 0x55b6ab85cc91 bp 0x7fff60bd4d70 sp 0x7fff60bd4ce0 T0)
> ==10595==The signal is caused by a READ memory access.
> #0 0x55b6ab85cc91 in gic_get_current_cpu /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:60:12
> #1 0x55b6ab85e1bd in gic_dist_writeb /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:1182:11
> #2 0x55b6ab855a97 in gic_dist_write /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:1514:9
> #3 0x55b6aa1650d4 in memory_region_write_with_attrs_accessor /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:503:12
> #4 0x55b6aa163ac6 in access_with_adjusted_size /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:544:18
> #5 0x55b6aa161f35 in memory_region_dispatch_write /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:1473:13
> #6 0x55b6a9313949 in flatview_write_continue /home/alxndr/Development/qemu/general-fuzz/exec.c:3176:23
> #7 0x55b6a92fca11 in flatview_write /home/alxndr/Development/qemu/general-fuzz/exec.c:3216:14
> #8 0x55b6a92fc54e in address_space_write /home/alxndr/Development/qemu/general-fuzz/exec.c:3308:18
> =================================================================
>
> Reproducer 2:
> cat << EOF | ./arm-softmmu/qemu-system-arm -machine highbank \
> -nographic -monitor none -serial none -qtest stdio
> writeq 0xfff11f00 0x613a650f0fda6555
> EOF
>
> ==1375==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x608000001c80 at pc 0x5618928c486e bp 0x7ffe22c4ee10 sp 0x7ffe22c4ee08
> READ of size 8 at 0x608000001c80 thread T0
> #0 0x5618928c486d in address_space_translate_iommu /home/alxndr/Development/qemu/general-fuzz/exec.c:451:23
> #1 0x561892850acc in flatview_do_translate /home/alxndr/Development/qemu/general-fuzz/e...

[Alexander Bulekov (a1xndr)]

On 200730 1550, Peter Maydell wrote:
> I haven't looked at reproducer #1, but is it a fuzzer-specific variant
> of LP:1602247 (trying to read the "for this CPU" registers from
> something other than a CPU doesn't work) ?

That was my initial suspicion as well, but it looks like the SEGV
happens here:
if (s->num_cpu > 1) {
rather than here:
    return current_cpu->cpu_index;

-Alex

>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/1889621
>
> Title:
> ARM Highbank Crashes Realted to GIC
>
> Status in QEMU:
> Confirmed
>
> Bug description:
> Hello,
> Here are some QTest reproducers for crashes on ARM Highbank that all seem to be related to the gic device.
>
> Reproducer 1:
> cat << EOF | ./arm-softmmu/qemu-system-arm -machine highbank \
> -nographic -monitor none -serial none -qtest stdio
> writel 0xfff11f00 0x8405f559
> writel 0xfff117fd 0x5c057bd8
> EOF
>
> ==10595==ERROR: AddressSanitizer: SEGV on unknown address 0x62b000013e01 (pc 0x55b6ab85cc91 bp 0x7fff60bd4d70 sp 0x7fff60bd4ce0 T0)
> ==10595==The signal is caused by a READ memory access.
> #0 0x55b6ab85cc91 in gic_get_current_cpu /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:60:12
> #1 0x55b6ab85e1bd in gic_dist_writeb /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:1182:11
> #2 0x55b6ab855a97 in gic_dist_write /home/alxndr/Development/qemu/general-fuzz/hw/intc/arm_gic.c:1514:9
> #3 0x55b6aa1650d4 in memory_region_write_with_attrs_accessor /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:503:12
> #4 0x55b6aa163ac6 in access_with_adjusted_size /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:544:18
> #5 0x55b6aa161f35 in memory_region_dispatch_write /home/alxndr/Development/qemu/general-fuzz/softmmu/memory.c:1473:13
> #6 0x55b6a9313949 in flatview_write_continue /home/alxndr/Development/qemu/general-fuzz/exec.c:3176:23
> #7 0x55b6a92fca11 in flatview_write /home/alxndr/Development/qemu/general-fuzz/exec.c:3216:14
> #8 0x55b6a92fc54e in address_space_write /home/alxndr/Development/qemu/general-fuzz/exec.c:3308:18
> =================================================================
>
> Reproducer 2:
> cat << EOF | ./arm-softmmu/qemu-system-arm -machine highbank \
> -nographic -monitor none -serial none -qtest stdio
> writeq 0xfff11f00 0x613a650f0fda6555
> EOF
>
> ==1375==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x608000001c80 at pc 0x5618928c486e bp 0x7ffe22c4ee10 sp 0x7ffe22c4ee08
> READ of size 8 at 0x608000001c80 thread T0
> #0 0x5618928c486d in address_space_translate_iommu /home/alxndr/Development/qemu/general-fuzz/exec.c:451:23
> #1 0x561892850acc in flatview_do_translate /home/alxndr/Development/qemu/general-fuzz/exec.c:524:16
> #2 0x5618928514ad in flatview_translate /home/alxndr/Development/qemu/general-fuzz/exec.c:584:15
> #3 0x5618928b1e14 in flatview_write_continue /home/alxndr/Development/qemu/general-fuzz/exec.c:3199:14
> #4 0x56189289aa11 in flatview_write /home/alxndr/Development/qemu/general-fuzz/exec.c:...

[Thomas Huth (th-huth)]

Can you still reproduce one of these issues with the current master branch of QEMU? For me, all three reproduces do not seem to cause any trouble anymore...

[Alexander Bulekov (a1xndr)]

I believe these were all taken care of by
edfe2eb436 ("hw/intc/arm_gic: Fix interrupt ID in GICD_SGIR register")
09bbdb89bc ("hw/intc/arm_gic: Allow to use QTest without crashing")

[Thomas Huth (th-huth)]

Ok, thanks, then let's close this (and open new tickets on gitlab if it happens again)