linux-kernel - Re: [PATCH] intel_pstate: Change busy calculation to use fixed point math.

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Wed, 26 Feb 2014 01:39:25 +0100
From:	"Rafael J. Wysocki" <rjw@...ysocki.net>
To:	dirk.brandewie@...il.com
Cc:	linux-pm@...r.kernel.org, linux-kernel@...r.kernel.org,
	gregkh@...uxfoundation.org, rafael.j.wysocki@...el.com,
	gf435@....net, bp@...en8.de,
	Dirk Brandewie <dirk.j.brandewie@...el.com>
Subject: Re: [PATCH] intel_pstate: Change busy calculation to use fixed point math.

On Tuesday, February 25, 2014 10:35:37 AM dirk.brandewie@...il.com wrote:
> From: Dirk Brandewie <dirk.j.brandewie@...el.com>
> 
> Commit fcb6a15c2e Take core C0 time into account for core busy calculation.
> 
> Introduced a regression on some processor SKUs supported by
> intel_pstate. This was caused by the truncation caused by using
> integer math to calculate core busy and C0 percentages.
> 
> On a i7-4770K processor operating at 800Mhz going to 100% utilization
> the percent busy of the CPU using integer math is 22% it actually is
> 22.85%.  This value scaled to the current frequency returned 97 which
> the PID interpreted as no error and did not adjust the P state.
> 
> Tested on i7-4770K, i7-2600, i5-3230M
> 
> References:
>    https://lkml.org/lkml/2014/2/19/626
>    https://bugzilla.kernel.org/show_bug.cgi?id=70941
> 
> Signed-off-by: Dirk Brandewie <dirk.j.brandewie@...el.com>

Queued up as a fix for 3.14.

Thanks!

> ---
>  drivers/cpufreq/intel_pstate.c | 28 ++++++++++++++++++----------
>  1 file changed, 18 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
> index e908161..2cd36b9 100644
> --- a/drivers/cpufreq/intel_pstate.c
> +++ b/drivers/cpufreq/intel_pstate.c
> @@ -39,9 +39,10 @@
>  #define BYT_TURBO_RATIOS	0x66c
>  
>  
> -#define FRAC_BITS 8
> +#define FRAC_BITS 6
>  #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
>  #define fp_toint(X) ((X) >> FRAC_BITS)
> +#define FP_ROUNDUP(X) ((X) += 1 << FRAC_BITS)
>  
>  static inline int32_t mul_fp(int32_t x, int32_t y)
>  {
> @@ -556,18 +557,20 @@ static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
>  static inline void intel_pstate_calc_busy(struct cpudata *cpu,
>  					struct sample *sample)
>  {
> -	u64 core_pct;
> -	u64 c0_pct;
> +	int32_t core_pct;
> +	int32_t c0_pct;
>  
> -	core_pct = div64_u64(sample->aperf * 100, sample->mperf);
> +	core_pct = div_fp(int_tofp((sample->aperf)),
> +			int_tofp((sample->mperf)));
> +	core_pct = mul_fp(core_pct, int_tofp(100));
> +	FP_ROUNDUP(core_pct);
> +
> +	c0_pct = div_fp(int_tofp(sample->mperf), int_tofp(sample->tsc));
>  
> -	c0_pct = div64_u64(sample->mperf * 100, sample->tsc);
>  	sample->freq = fp_toint(
> -		mul_fp(int_tofp(cpu->pstate.max_pstate),
> -			int_tofp(core_pct * 1000)));
> +		mul_fp(int_tofp(cpu->pstate.max_pstate * 1000), core_pct));
>  
> -	sample->core_pct_busy = mul_fp(int_tofp(core_pct),
> -				div_fp(int_tofp(c0_pct + 1), int_tofp(100)));
> +	sample->core_pct_busy = mul_fp(core_pct, c0_pct);
>  }
>  
>  static inline void intel_pstate_sample(struct cpudata *cpu)
> @@ -579,6 +582,10 @@ static inline void intel_pstate_sample(struct cpudata *cpu)
>  	rdmsrl(MSR_IA32_MPERF, mperf);
>  	tsc = native_read_tsc();
>  
> +	aperf = aperf >> FRAC_BITS;
> +	mperf = mperf >> FRAC_BITS;
> +	tsc = tsc >> FRAC_BITS;
> +
>  	cpu->sample_ptr = (cpu->sample_ptr + 1) % SAMPLE_COUNT;
>  	cpu->samples[cpu->sample_ptr].aperf = aperf;
>  	cpu->samples[cpu->sample_ptr].mperf = mperf;
> @@ -610,7 +617,8 @@ static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
>  	core_busy = cpu->samples[cpu->sample_ptr].core_pct_busy;
>  	max_pstate = int_tofp(cpu->pstate.max_pstate);
>  	current_pstate = int_tofp(cpu->pstate.current_pstate);
> -	return mul_fp(core_busy, div_fp(max_pstate, current_pstate));
> +	core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
> +	return FP_ROUNDUP(core_busy);
>  }
>  
>  static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
> 

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/