Re: kernel: clockevents: shift out-of-bounds

[Peter Zijlstra]

On Mon, Oct 20, 2014 at 03:07:50PM +0400, Andrey Ryabinin wrote:
> 
> On kernel with UBSan enabled I've got following:
> 
>      UBSan: Undefined behaviour in ../kernel/time/clockevents.c:75:34
>      shift exponent 32 is to large for 32-bit type 'unsigned int'
>      CPU: 0 PID: 0 Comm: swapper/0 Not tainted 3.17.0-rc7+ #39
>      Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.7.5-0-ge51488c-20140602_164612-nilsson.home.kraxel.org 04/01/2014
>       0000000000000000 0000000000000000 0000000000000001 ffffffff83003db0
>       ffffffff82a30940 0000000000000020 ffffffff83003dc0 ffffffff819502e9
>       ffffffff83003e40 ffffffff81950735 ffff88013f003233 0000000000000000
>      Call Trace:
>      dump_stack (/home/andrew/linux/lib/dump_stack.c:52)
>      ubsan_epilogue (/home/andrew/linux/lib/ubsan.c:122)
>      __ubsan_handle_shift_out_of_bounds (/home/andrew/linux/lib/ubsan.c:390)
>      ? hpet_enable (/home/andrew/linux/arch/x86/kernel/hpet.c:862)
>      cev_delta2ns (/home/andrew/linux/kernel/time/clockevents.c:75 (discriminator 1))
>      clockevents_config.part.2 (/home/andrew/linux/kernel/time/clockevents.c:421)
>      ? __clocksource_select (/home/andrew/linux/kernel/time/clocksource.c:607 /home/andrew/linux/kernel/time/clocksource.c:631)
>      clockevents_config_and_register (/home/andrew/linux/kernel/time/clockevents.c:440)
>      hpet_enable (/home/andrew/linux/arch/x86/kernel/hpet.c:305 /home/andrew/linux/arch/x86/kernel/hpet.c:891)
>      hpet_time_init (/home/andrew/linux/arch/x86/kernel/time.c:79)
>      x86_late_time_init (/home/andrew/linux/arch/x86/kernel/time.c:87)
>      start_kernel (/home/andrew/linux/init/main.c:637)
>      ? early_idt_handlers (/home/andrew/linux/arch/x86/kernel/head_64.S:344)
>      x86_64_start_reservations (/home/andrew/linux/arch/x86/kernel/head64.c:194)
>      x86_64_start_kernel (/home/andrew/linux/arch/x86/kernel/head64.c:183)
> 
> I guess it should be 1ULL here instead of 1U:
>             (!ismax || evt->mult <= (1U << evt->shift)))

Probably so indeed, clocks_calc_mult_shift() has max return value of 32,
which will indeed trigger that overflow.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/