Re: [PATCH] perf/x86: Fix integer overflow when left shifting an integer more than 32 bits

[Peter Zijlstra]

On Tue, Apr 20, 2021 at 03:29:07PM +0100, Colin King wrote:
> From: Colin Ian King <colin.king@xxxxxxxxxxxxx>
> 
> The 64 bit value read from MSR_ARCH_PERFMON_FIXED_CTR_CTRL is being
> bit-wise masked with the value (0x03 << i*4). However, the shifted value
> is evaluated using 32 bit arithmetic, so will overflow when i > 8.
> Fix this by making 0x03 a ULL so that the shift is performed using
> 64 bit arithmetic.
> 
> Addresses-Coverity: ("Unintentional integer overflow")

Strange tag that, also inaccurate, wide shifts are UB and don't behave
consistently.

As is, we've not had hardware with that many fixed counters, but yes,
worth fixing I suppose.

> Fixes: a5ebe0ba3dff ("perf/x86: Check all MSRs before passing hw check")
> Signed-off-by: Colin Ian King <colin.king@xxxxxxxxxxxxx>
> ---
>  arch/x86/events/core.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
> index bafd93c54ffa..59c665c8c2e9 100644
> --- a/arch/x86/events/core.c
> +++ b/arch/x86/events/core.c
> @@ -261,7 +261,7 @@ static bool check_hw_exists(void)
>  		for (i = 0; i < x86_pmu.num_counters_fixed; i++) {
>  			if (fixed_counter_disabled(i))
>  				continue;
> -			if (val & (0x03 << i*4)) {
> +			if (val & (0x03ULL << i*4)) {
>  				bios_fail = 1;
>  				val_fail = val;
>  				reg_fail = reg;
> -- 
> 2.30.2
>