Re: FW: [4.12 regression] "cpufreq: intel_pstate: Active mode P-state limits rework"

[Rafael J. Wysocki]

On Thursday, June 01, 2017 01:08:16 AM Jongman Heo wrote:
> > 
> >--------- Original Message ---------
> >Sender : Rafael J. Wysocki <rjw@xxxxxxxxxxxxx>
> >Date   : 2017-06-01 08:58 (GMT+9)
> >Title  : Re: FW: [4.12 regression] "cpufreq: intel_pstate: Active mode P-state limits rework"
> > 
> >On Monday, May 29, 2017 02:18:52 AM Jongman Heo wrote:
> >> 
> >> Hi,
> >> 
> >> FYI, 4.12-rc3 still has this issue.
> >> 
> >>  
> >> --------- Original Message ---------
> >> Sender : íìë <jongman.heo@xxxxxxxxxxx>
> >> Date   : 2017-05-16 13:25 (GMT+9)
> >> Title  : [4.12 regression] "cpufreq: intel_pstate: Active mode P-state limits rework"
> >>  
> >>  
> >> Hi,
> >>  
> >> With 4.12-rc1 (Linus git), booting fails due to kernel panic, at intel_pstate_register_driver+0x56/0x110.
> >> I can't copy the whole trace from the graphic console, it looks like below.
> >>  
> >> Call Trace:
> >>   intel_pstate_init
> >>   intel_pstate_setup
> >>   do_one_initcall
> >>   set_debug_rodata
> >>   kernel_init_freeable
> >>   rest_init
> >>   kernel_init
> >>   ret_from_fork
> >>     Code: ...
> >>   RIP: intel_pstate_register_driver+0x56/0x110 RSP: ffffa3a98000bd80
> >>   ...
> > 
> >I guess this is a division by 0 due to the unmet assumption that the denominator
> >in min_perf_pct_min() will always be nonzero.
> > 
> >If this guess is correct, the patch below should help, so can you please test it?
> > 
> >Thanks,
> >Rafael
> > 
> > 
> >---
> > drivers/cpufreq/intel_pstate.c |    5 +++--
> > 1 file changed, 3 insertions(+), 2 deletions(-)
> > 
> >Index: linux-pm/drivers/cpufreq/intel_pstate.c
> >===================================================================
> >--- linux-pm.orig/drivers/cpufreq/intel_pstate.c
> >+++ linux-pm/drivers/cpufreq/intel_pstate.c
> >@@ -571,9 +571,10 @@ static inline void update_turbo_state(vo
> > static int min_perf_pct_min(void)
> > {
> >         struct cpudata *cpu = all_cpu_data[0];
> >+        int turbo_pstate = cpu->pstate.turbo_pstate;
> > 
> >-        return DIV_ROUND_UP(cpu->pstate.min_pstate * 100,
> >-                            cpu->pstate.turbo_pstate);
> >+        return turbo_pstate ?
> >+                DIV_ROUND_UP(cpu->pstate.min_pstate * 100, turbo_pstate) : 0;
> > }
> > 
> > static s16 intel_pstate_get_epb(struct cpudata *cpu_data)
> > 
> > 
> > 
> 
> Hi,
> 
> Your patch fixes the issue.

OK, thanks for the confirmation.

Here it goes again with a changelog and all.

---
From: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
Subject: [PATCH] cpufreq: intel_pstate: Avoid division by 0 in min_perf_pct_min()

Commit c5a2ee7dde89 (cpufreq: intel_pstate: Active mode P-state
limits rework) incorrectly assumed that pstate.turbo_pstate will
always be nonzero for CPU0 in min_perf_pct_min() if
cpufreq_register_driver() has succeeded which may not be the case
in virtualized environments.

If that is not the case, it leads to an early crash on boot in
intel_pstate_register_driver(), so add a sanity check to
min_perf_pct_min() to prevent the crash from happening.

Fixes: c5a2ee7dde89 (cpufreq: intel_pstate: Active mode P-state limits rework)
Reported-and-tested-by: Jongman Heo <jongman.heo@xxxxxxxxxxx>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
---
 drivers/cpufreq/intel_pstate.c |    5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

Index: linux-pm/drivers/cpufreq/intel_pstate.c
===================================================================
--- linux-pm.orig/drivers/cpufreq/intel_pstate.c
+++ linux-pm/drivers/cpufreq/intel_pstate.c
@@ -571,9 +571,10 @@ static inline void update_turbo_state(vo
 static int min_perf_pct_min(void)
 {
 	struct cpudata *cpu = all_cpu_data[0];
+	int turbo_pstate = cpu->pstate.turbo_pstate;
 
-	return DIV_ROUND_UP(cpu->pstate.min_pstate * 100,
-			    cpu->pstate.turbo_pstate);
+	return turbo_pstate ?
+		DIV_ROUND_UP(cpu->pstate.min_pstate * 100, turbo_pstate) : 0;
 }
 
 static s16 intel_pstate_get_epb(struct cpudata *cpu_data)