答复: [外部邮件] Re: [PATCH v2 1/2] cpufreq: acpi-cpufreq: fix P-state index mismatch in extract_io()

[Li,Rongqing]

> > When policy->freq_table is built in acpi_cpufreq_cpu_init(), entries
> > with duplicate frequencies are skipped. The original P-state index for
> > each remaining entry is stored in freq_table[].driver_data, so the
> > index space of freq_table no longer matches perf->states[].
> >
> > extract_io() walks perf->states[] with index i and uses the same i to
> > index policy->freq_table[i]. This causes two problems when duplicate
> > frequencies exist:
> >
> >    - Returning the frequency of the wrong P-state
> >    - Returning 0 from a zeroed tail entry, or even CPUFREQ_TABLE_END
> >      (~1u) reported as 0xfffffffe kHz
> 
> AFAICT, It may be worth expanding the changelog to cover the consequences
> described below , particularly the resulting divergence between the hardware
> state and the cpufreq core's view of the current
> frequency:
> 
> Furthermore, extract_io() is only reachable on ACPI_ADR_SPACE_SYSTEM_IO
> platforms, where cpufreq_driver->get is not installed, so its only caller in
> practice is check_freqs(), i.e. only when the acpi_pstate_strict module
> parameter is set.
> 
> There, the mismatched lookup makes the frequency comparison fail even
> though drv_write() has already switched the hardware to the requested P-state.
> So check_freqs() sleeps through all 100 iterations - at least
> ~1 ms of usleep_range() plus 100 cross-CPU calls and I/O port reads, all with
> policy->rwsem held - and ->target_index() returns -EAGAIN. perf
> ->state is therefore left at its previous value while the hardware sits
> at the new one.
> 
> The core then restores policy->cur to the old frequency, and because
> __cpufreq_driver_target() returns early when the requested frequency equals
> policy->cur, the driver is not called again for it - so the control register is not
> rewritten and the CPU is left running at a frequency the core does not know
> about.
> 
> 
> >
> > Fix it by walking policy->freq_table with cpufreq_for_each_entry() and
> > using perf->states[pos->driver_data].status, aligning with extract_msr().
> >
> > Fixes: 8cee1eed8e78 ("cpufreq: ACPI: Remove freq_table from
> > acpi_cpufreq_data")
> 
> The real tag should be fe27cb358835 ("[CPUFREQ][2/8] acpi:
> reorganize code to make MSR support addition easier")
> 
> That commit added both the entry-skipping loop with the freq_table[].index
> (now .driver_data) back-pointer, which is what makes the two indices diverge,
> and the faulty lookup itself -- back then in a function called extract_freq().
> dde9f7ba60ad ("[CPUFREQ][3/8] acpi
> cpufreq: Pull in MSR based transition support") merely renamed it to
> extract_io().
> 

Thanks, I will send v3

[Li,Rongqing] 


> 
> > Signed-off-by: Li RongQing <lirongqing@xxxxxxxxx>
> > ---
> >   drivers/cpufreq/acpi-cpufreq.c | 9 ++++-----
> >   1 file changed, 4 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/cpufreq/acpi-cpufreq.c
> > b/drivers/cpufreq/acpi-cpufreq.c index 21639d9..1abe9ab 100644
> > --- a/drivers/cpufreq/acpi-cpufreq.c
> > +++ b/drivers/cpufreq/acpi-cpufreq.c
> > @@ -197,14 +197,13 @@ static unsigned extract_io(struct cpufreq_policy
> *policy, u32 value)
> >   {
> >   	struct acpi_cpufreq_data *data = policy->driver_data;
> >   	struct acpi_processor_performance *perf;
> > -	int i;
> > +	struct cpufreq_frequency_table *pos;
> >
> >   	perf = to_perf_data(data);
> >
> > -	for (i = 0; i < perf->state_count; i++) {
> > -		if (value == perf->states[i].status)
> > -			return policy->freq_table[i].frequency;
> > -	}
> > +	cpufreq_for_each_entry(pos, policy->freq_table)
> > +		if (value == perf->states[pos->driver_data].status)
> > +			return pos->frequency;
> >   	return 0;
> >   }
> >
> 
> 
> --
> Thx and BRs,
> Zhongqiu Han