Skip to contentSkip to conversation

UBSAN: Undefined behaviour in kernel/time/tick-sched.c:600:9

DISCUSSION Meelis Roos 3 participants 5 messages Mar 10, 2017 – Mar 11, 2017

lore
Share & tools
lore
#0 lore raw

(Retry with fixed address ox 86 list)

This HP Netserver LT6000R, quad P3 Xeon. Trying 4.11.0-rc1-00088-gec3b93a, I got the following UBSAN warning that was not there in 4.10:

[   14.820437] ================================================================================
[   14.820534] UBSAN: Undefined behaviour in kernel/time/tick-sched.c:600:9
[   14.820594] signed integer overflow:
[   14.820645] 11567464533 + 3985952 cannot be represented in type 'long long int'
[   14.820722] CPU: 3 PID: 135 Comm: startpar Not tainted 4.11.0-rc1-00088-gec3b93a #61
[   14.820793] Hardware name: Hewlett Packard HP NetServer/HP System Board, BIOS 4.06.46 PW 06/25/2003
[   14.820863] Call Trace:
[   14.820950]  ? dump_stack+0x45/0x6e
[   14.821022]  ? ubsan_epilogue+0xb/0x40
[   14.821081]  ? handle_overflow+0xb6/0xe0
[   14.821146]  ? smp_trace_apic_timer_interrupt+0x85/0x100
[   14.821208]  ? __ubsan_handle_add_overflow+0xf/0x20
[   14.821281]  ? get_cpu_idle_time_us+0x204/0x2b0
[   14.821350]  ? get_idle_time+0x57/0xb0
[   14.821405]  ? show_stat+0x217/0x10c0
[   14.821463]  ? seq_open+0x33/0xb0
[   14.821519]  ? __d_alloc+0x149/0x2c0
[   14.821574]  ? get_iowait_time+0xb0/0xb0
[   14.821628]  ? single_open+0x4d/0xb0
[   14.821685]  ? single_open_size+0x37/0xc0
[   14.821748]  ? path_openat+0x7d5/0x1bd0
[   14.821804]  ? seq_read+0xca/0x6f0
[   14.821872]  ? _copy_to_user+0x5e/0x90
[   14.821937]  ? cp_new_stat64+0x108/0x130
[   14.821991]  ? __seq_open_private+0x80/0x80
[   14.822062]  ? proc_reg_read+0x7c/0x120
[   14.822117]  ? proc_reg_write+0x120/0x120
[   14.822172]  ? __vfs_read+0x3e/0x240
[   14.822228]  ? vfs_read+0x92/0x1b0
[   14.822282]  ? SyS_read+0x45/0xa0
[   14.822338]  ? do_fast_syscall_32+0xb9/0x220
[   14.822395]  ? entry_SYSENTER_32+0x47/0x71
[   14.822447] 

================================================================================

Signature · 1 line
Meelis Roos ([email protected])
#1 lore raw

On Fri, 10 Mar 2017, Meelis Roos wrote:

Cc'ing KASAN folks

Quoted 10 lines

(Retry with fixed address ox 86 list)

This HP Netserver LT6000R, quad P3 Xeon. Trying 4.11.0-rc1-00088-gec3b93a, I got the following UBSAN warning that was not there in 4.10:

[   14.820437] ================================================================================
[   14.820534] UBSAN: Undefined behaviour in kernel/time/tick-sched.c:600:9
[   14.820594] signed integer overflow:
[   14.820645] 11567464533 + 3985952 cannot be represented in type 'long long int'

This does not make any sense:

     11567464533 + 3985952 = 0x2b1798055 + 0x3cd220 = 0x2b1b65275

which definitely fits into a 'long long int' aka 64 bit

Quoted 35 lines
[   14.820722] CPU: 3 PID: 135 Comm: startpar Not tainted 4.11.0-rc1-00088-gec3b93a #61
[   14.820793] Hardware name: Hewlett Packard HP NetServer/HP System Board, BIOS 4.06.46 PW 06/25/2003
[   14.820863] Call Trace:
[   14.820950]  ? dump_stack+0x45/0x6e
[   14.821022]  ? ubsan_epilogue+0xb/0x40
[   14.821081]  ? handle_overflow+0xb6/0xe0
[   14.821146]  ? smp_trace_apic_timer_interrupt+0x85/0x100
[   14.821208]  ? __ubsan_handle_add_overflow+0xf/0x20
[   14.821281]  ? get_cpu_idle_time_us+0x204/0x2b0
[   14.821350]  ? get_idle_time+0x57/0xb0
[   14.821405]  ? show_stat+0x217/0x10c0
[   14.821463]  ? seq_open+0x33/0xb0
[   14.821519]  ? __d_alloc+0x149/0x2c0
[   14.821574]  ? get_iowait_time+0xb0/0xb0
[   14.821628]  ? single_open+0x4d/0xb0
[   14.821685]  ? single_open_size+0x37/0xc0
[   14.821748]  ? path_openat+0x7d5/0x1bd0
[   14.821804]  ? seq_read+0xca/0x6f0
[   14.821872]  ? _copy_to_user+0x5e/0x90
[   14.821937]  ? cp_new_stat64+0x108/0x130
[   14.821991]  ? __seq_open_private+0x80/0x80
[   14.822062]  ? proc_reg_read+0x7c/0x120
[   14.822117]  ? proc_reg_write+0x120/0x120
[   14.822172]  ? __vfs_read+0x3e/0x240
[   14.822228]  ? vfs_read+0x92/0x1b0
[   14.822282]  ? SyS_read+0x45/0xa0
[   14.822338]  ? do_fast_syscall_32+0xb9/0x220
[   14.822395]  ? entry_SYSENTER_32+0x47/0x71
[   14.822447] 

================================================================================

Signature · 1 line
Meelis Roos ([email protected])
#2 lore raw
Quoted 15 lines · Thomas Gleixner

On Fri, 10 Mar 2017, Meelis Roos wrote:

Cc'ing KASAN folks

Quoted 10 lines

(Retry with fixed address ox 86 list)

This HP Netserver LT6000R, quad P3 Xeon. Trying 4.11.0-rc1-00088-gec3b93a, I got the following UBSAN warning that was not there in 4.10:

[   14.820437] ================================================================================
[   14.820534] UBSAN: Undefined behaviour in kernel/time/tick-sched.c:600:9
[   14.820594] signed integer overflow:
[   14.820645] 11567464533 + 3985952 cannot be represented in type 'long long int'
Quoted 5 lines

This does not make any sense:

     11567464533 + 3985952 = 0x2b1798055 + 0x3cd220 = 0x2b1b65275

which definitely fits into a 'long long int' aka 64 bit

Is there a reproducer? Or can you print the values right in tick-sched.c with pr_err?

The code in ubsan.c looks more complex than I would expect from that code. There is probably some potential for truncation:

static s_max get_signed_val(struct type_descriptor *type, unsigned long val)

{
        if (is_inline_int(type)) {
                unsigned extra_bits = sizeof(s_max)*8 - type_bit_width(type);
                return ((s_max)val) << extra_bits >> extra_bits;
        }
        if (type_bit_width(type) == 64)
                return *(s64 *)val;
        return *(s_max *)val;
}
Quoted 30 lines
Quoted 30 lines
[   14.820722] CPU: 3 PID: 135 Comm: startpar Not tainted 4.11.0-rc1-00088-gec3b93a #61
[   14.820793] Hardware name: Hewlett Packard HP NetServer/HP System Board, BIOS 4.06.46 PW 06/25/2003
[   14.820863] Call Trace:
[   14.820950]  ? dump_stack+0x45/0x6e
[   14.821022]  ? ubsan_epilogue+0xb/0x40
[   14.821081]  ? handle_overflow+0xb6/0xe0
[   14.821146]  ? smp_trace_apic_timer_interrupt+0x85/0x100
[   14.821208]  ? __ubsan_handle_add_overflow+0xf/0x20
[   14.821281]  ? get_cpu_idle_time_us+0x204/0x2b0
[   14.821350]  ? get_idle_time+0x57/0xb0
[   14.821405]  ? show_stat+0x217/0x10c0
[   14.821463]  ? seq_open+0x33/0xb0
[   14.821519]  ? __d_alloc+0x149/0x2c0
[   14.821574]  ? get_iowait_time+0xb0/0xb0
[   14.821628]  ? single_open+0x4d/0xb0
[   14.821685]  ? single_open_size+0x37/0xc0
[   14.821748]  ? path_openat+0x7d5/0x1bd0
[   14.821804]  ? seq_read+0xca/0x6f0
[   14.821872]  ? _copy_to_user+0x5e/0x90
[   14.821937]  ? cp_new_stat64+0x108/0x130
[   14.821991]  ? __seq_open_private+0x80/0x80
[   14.822062]  ? proc_reg_read+0x7c/0x120
[   14.822117]  ? proc_reg_write+0x120/0x120
[   14.822172]  ? __vfs_read+0x3e/0x240
[   14.822228]  ? vfs_read+0x92/0x1b0
[   14.822282]  ? SyS_read+0x45/0xa0
[   14.822338]  ? do_fast_syscall_32+0xb9/0x220
[   14.822395]  ? entry_SYSENTER_32+0x47/0x71
[   14.822447]

================================================================================

#3 lore raw
Quoted 28 lines · Dmitry Vyukov
Quoted 14 lines

Cc'ing KASAN folks

Quoted 10 lines

(Retry with fixed address ox 86 list)

This HP Netserver LT6000R, quad P3 Xeon. Trying 4.11.0-rc1-00088-gec3b93a, I got the following UBSAN warning that was not there in 4.10:

[   14.820437] ================================================================================
[   14.820534] UBSAN: Undefined behaviour in kernel/time/tick-sched.c:600:9
[   14.820594] signed integer overflow:
[   14.820645] 11567464533 + 3985952 cannot be represented in type 'long long int'
Quoted 5 lines

This does not make any sense:

     11567464533 + 3985952 = 0x2b1798055 + 0x3cd220 = 0x2b1b65275

which definitely fits into a 'long long int' aka 64 bit

Is there a reproducer? Or can you print the values right in tick-sched.c with pr_err?

The code in ubsan.c looks more complex than I would expect from that code. There is probably some potential for truncation:

static s_max get_signed_val(struct type_descriptor *type, unsigned long val)

Quoted 9 lines
{
        if (is_inline_int(type)) {
                unsigned extra_bits = sizeof(s_max)*8 - type_bit_width(type);
                return ((s_max)val) << extra_bits >> extra_bits;
        }
        if (type_bit_width(type) == 64)
                return *(s64 *)val;
        return *(s_max *)val;
}

Also what compiler is this?

Quoted 30 lines
Quoted 30 lines
Quoted 30 lines
[   14.820722] CPU: 3 PID: 135 Comm: startpar Not tainted 4.11.0-rc1-00088-gec3b93a #61
[   14.820793] Hardware name: Hewlett Packard HP NetServer/HP System Board, BIOS 4.06.46 PW 06/25/2003
[   14.820863] Call Trace:
[   14.820950]  ? dump_stack+0x45/0x6e
[   14.821022]  ? ubsan_epilogue+0xb/0x40
[   14.821081]  ? handle_overflow+0xb6/0xe0
[   14.821146]  ? smp_trace_apic_timer_interrupt+0x85/0x100
[   14.821208]  ? __ubsan_handle_add_overflow+0xf/0x20
[   14.821281]  ? get_cpu_idle_time_us+0x204/0x2b0
[   14.821350]  ? get_idle_time+0x57/0xb0
[   14.821405]  ? show_stat+0x217/0x10c0
[   14.821463]  ? seq_open+0x33/0xb0
[   14.821519]  ? __d_alloc+0x149/0x2c0
[   14.821574]  ? get_iowait_time+0xb0/0xb0
[   14.821628]  ? single_open+0x4d/0xb0
[   14.821685]  ? single_open_size+0x37/0xc0
[   14.821748]  ? path_openat+0x7d5/0x1bd0
[   14.821804]  ? seq_read+0xca/0x6f0
[   14.821872]  ? _copy_to_user+0x5e/0x90
[   14.821937]  ? cp_new_stat64+0x108/0x130
[   14.821991]  ? __seq_open_private+0x80/0x80
[   14.822062]  ? proc_reg_read+0x7c/0x120
[   14.822117]  ? proc_reg_write+0x120/0x120
[   14.822172]  ? __vfs_read+0x3e/0x240
[   14.822228]  ? vfs_read+0x92/0x1b0
[   14.822282]  ? SyS_read+0x45/0xa0
[   14.822338]  ? do_fast_syscall_32+0xb9/0x220
[   14.822395]  ? entry_SYSENTER_32+0x47/0x71
[   14.822447]

================================================================================

#4 lore raw
In reply to Dmitry Vyukov
Quoted 15 lines
Quoted 15 lines

Cc'ing KASAN folks

Quoted 10 lines

(Retry with fixed address ox 86 list)

This HP Netserver LT6000R, quad P3 Xeon. Trying 4.11.0-rc1-00088-gec3b93a, I got the following UBSAN warning that was not there in 4.10:

[   14.820437] ================================================================================
[   14.820534] UBSAN: Undefined behaviour in kernel/time/tick-sched.c:600:9
[   14.820594] signed integer overflow:
[   14.820645] 11567464533 + 3985952 cannot be represented in type 'long long int'

This does not make any sense:

Quoted 7 lines
Quoted 3 lines
     11567464533 + 3985952 = 0x2b1798055 + 0x3cd220 = 0x2b1b65275

which definitely fits into a 'long long int' aka 64 bit

Is there a reproducer? Or can you print the values right in tick-sched.c with pr_err?

No, it's not reproducible at will - subsequent reboots did not cause it, only the KASAN warning from drivers/acpi/acpica/dsutils.c:640 and arch/x86/events/intel/p6.c:115:29 are there as usual.

gcc version 6.2.0 20160927 (Debian 6.2.0-5)

Quoted 48 lines

The code in ubsan.c looks more complex than I would expect from that code. There is probably some potential for truncation:

static s_max get_signed_val(struct type_descriptor *type, unsigned long val)

{
        if (is_inline_int(type)) {
                unsigned extra_bits = sizeof(s_max)*8 - type_bit_width(type);
                return ((s_max)val) << extra_bits >> extra_bits;
        }
        if (type_bit_width(type) == 64)
                return *(s64 *)val;
        return *(s_max *)val;
}
Quoted 30 lines
Quoted 30 lines
[   14.820722] CPU: 3 PID: 135 Comm: startpar Not tainted 4.11.0-rc1-00088-gec3b93a #61
[   14.820793] Hardware name: Hewlett Packard HP NetServer/HP System Board, BIOS 4.06.46 PW 06/25/2003
[   14.820863] Call Trace:
[   14.820950]  ? dump_stack+0x45/0x6e
[   14.821022]  ? ubsan_epilogue+0xb/0x40
[   14.821081]  ? handle_overflow+0xb6/0xe0
[   14.821146]  ? smp_trace_apic_timer_interrupt+0x85/0x100
[   14.821208]  ? __ubsan_handle_add_overflow+0xf/0x20
[   14.821281]  ? get_cpu_idle_time_us+0x204/0x2b0
[   14.821350]  ? get_idle_time+0x57/0xb0
[   14.821405]  ? show_stat+0x217/0x10c0
[   14.821463]  ? seq_open+0x33/0xb0
[   14.821519]  ? __d_alloc+0x149/0x2c0
[   14.821574]  ? get_iowait_time+0xb0/0xb0
[   14.821628]  ? single_open+0x4d/0xb0
[   14.821685]  ? single_open_size+0x37/0xc0
[   14.821748]  ? path_openat+0x7d5/0x1bd0
[   14.821804]  ? seq_read+0xca/0x6f0
[   14.821872]  ? _copy_to_user+0x5e/0x90
[   14.821937]  ? cp_new_stat64+0x108/0x130
[   14.821991]  ? __seq_open_private+0x80/0x80
[   14.822062]  ? proc_reg_read+0x7c/0x120
[   14.822117]  ? proc_reg_write+0x120/0x120
[   14.822172]  ? __vfs_read+0x3e/0x240
[   14.822228]  ? vfs_read+0x92/0x1b0
[   14.822282]  ? SyS_read+0x45/0xa0
[   14.822338]  ? do_fast_syscall_32+0xb9/0x220
[   14.822395]  ? entry_SYSENTER_32+0x47/0x71
[   14.822447]

================================================================================

Signature · 1 line
Meelis Roos ([email protected])      http://www.cs.ut.ee/~mroos/