| Message ID | 20140425081815.10258.36201.stgit@srivatsabhat.in.ibm.com (mailing list archive) |
|---|---|
| State | Superseded, archived |
| Headers | show |
On 25 April 2014 13:48, Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> wrote: > During frequency transitions, the cpufreq core takes the responsibility of > invoking cpufreq_freq_transition_begin() and cpufreq_freq_transition_end() > for those cpufreq drivers that define the ->target_index callback but don't > set the ASYNC_NOTIFICATION flag. > > The powernow-k6 cpufreq driver falls under this category, but this driver was > invoking the _begin() and _end() APIs itself around frequency transitions, > which led to double invocation of the _begin() API. The _begin API makes > contending callers wait until the previous invocation is complete. Hence, > the powernow-k6 driver ended up waiting on itself, leading to system hangs > during boot. > > Fix this by removing the calls to the _begin() and _end() APIs from the > powernow-k6 driver, since they rightly belong to the cpufreq core. > > (Note that during ->exit(), the powernow-k6 driver sets the frequency > without any help from the cpufreq core. So add explicit calls to the > _begin() and _end() APIs around that frequency transition alone, to take > care of that special case. Also, add a missing 'break' statement there.) > > Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> > --- > > drivers/cpufreq/powernow-k6.c | 20 +++++++++++--------- > 1 file changed, 11 insertions(+), 9 deletions(-) > > diff --git a/drivers/cpufreq/powernow-k6.c b/drivers/cpufreq/powernow-k6.c > index 49f120e..6a4c34e 100644 > --- a/drivers/cpufreq/powernow-k6.c > +++ b/drivers/cpufreq/powernow-k6.c > @@ -138,22 +138,14 @@ static void powernow_k6_set_cpu_multiplier(unsigned int best_i) > static int powernow_k6_target(struct cpufreq_policy *policy, > unsigned int best_i) > { > - struct cpufreq_freqs freqs; > > if (clock_ratio[best_i].driver_data > max_multiplier) { > printk(KERN_ERR PFX "invalid target frequency\n"); > return -EINVAL; > } > > - freqs.old = busfreq * powernow_k6_get_cpu_multiplier(); > - freqs.new = busfreq * clock_ratio[best_i].driver_data; > - > - cpufreq_freq_transition_begin(policy, &freqs); > - > powernow_k6_set_cpu_multiplier(best_i); > > - cpufreq_freq_transition_end(policy, &freqs, 0); > - > return 0; > } > > @@ -228,8 +220,18 @@ static int powernow_k6_cpu_exit(struct cpufreq_policy *policy) > { > unsigned int i; > for (i = 0; i < 8; i++) { > - if (i == max_multiplier) > + if (i == max_multiplier) { This wouldn't work, it has to be: (clock_ratio[i].driver_data == max_multiplier) > + struct cpufreq_freqs freqs; > + > + freqs.old = policy->cur; > + freqs.new = clock_ratio[i].frequency; > + freqs.flags = 0; > + > + cpufreq_freq_transition_begin(policy, &freqs); > powernow_k6_target(policy, i); > + cpufreq_freq_transition_end(policy, &freqs, 0); > + break; > + } > } > return 0; > } Acked-by: Viresh Kumar <viresh.kumar@linaro.org> -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 04/25/2014 01:58 PM, Viresh Kumar wrote: > On 25 April 2014 13:48, Srivatsa S. Bhat > <srivatsa.bhat@linux.vnet.ibm.com> wrote: >> During frequency transitions, the cpufreq core takes the responsibility of >> invoking cpufreq_freq_transition_begin() and cpufreq_freq_transition_end() >> for those cpufreq drivers that define the ->target_index callback but don't >> set the ASYNC_NOTIFICATION flag. >> >> The powernow-k6 cpufreq driver falls under this category, but this driver was >> invoking the _begin() and _end() APIs itself around frequency transitions, >> which led to double invocation of the _begin() API. The _begin API makes >> contending callers wait until the previous invocation is complete. Hence, >> the powernow-k6 driver ended up waiting on itself, leading to system hangs >> during boot. >> >> Fix this by removing the calls to the _begin() and _end() APIs from the >> powernow-k6 driver, since they rightly belong to the cpufreq core. >> >> (Note that during ->exit(), the powernow-k6 driver sets the frequency >> without any help from the cpufreq core. So add explicit calls to the >> _begin() and _end() APIs around that frequency transition alone, to take >> care of that special case. Also, add a missing 'break' statement there.) >> >> Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> >> --- >> >> drivers/cpufreq/powernow-k6.c | 20 +++++++++++--------- >> 1 file changed, 11 insertions(+), 9 deletions(-) >> >> diff --git a/drivers/cpufreq/powernow-k6.c b/drivers/cpufreq/powernow-k6.c >> index 49f120e..6a4c34e 100644 >> --- a/drivers/cpufreq/powernow-k6.c >> +++ b/drivers/cpufreq/powernow-k6.c >> @@ -138,22 +138,14 @@ static void powernow_k6_set_cpu_multiplier(unsigned int best_i) >> static int powernow_k6_target(struct cpufreq_policy *policy, >> unsigned int best_i) >> { >> - struct cpufreq_freqs freqs; >> >> if (clock_ratio[best_i].driver_data > max_multiplier) { >> printk(KERN_ERR PFX "invalid target frequency\n"); >> return -EINVAL; >> } >> >> - freqs.old = busfreq * powernow_k6_get_cpu_multiplier(); >> - freqs.new = busfreq * clock_ratio[best_i].driver_data; >> - >> - cpufreq_freq_transition_begin(policy, &freqs); >> - >> powernow_k6_set_cpu_multiplier(best_i); >> >> - cpufreq_freq_transition_end(policy, &freqs, 0); >> - >> return 0; >> } >> >> @@ -228,8 +220,18 @@ static int powernow_k6_cpu_exit(struct cpufreq_policy *policy) >> { >> unsigned int i; >> for (i = 0; i < 8; i++) { >> - if (i == max_multiplier) >> + if (i == max_multiplier) { > > This wouldn't work, it has to be: (clock_ratio[i].driver_data == max_multiplier) > Hmm, looks like an existing bug in the driver. Will fix this in the next version. Regards, Srivatsa S. Bhat >> + struct cpufreq_freqs freqs; >> + >> + freqs.old = policy->cur; >> + freqs.new = clock_ratio[i].frequency; >> + freqs.flags = 0; >> + >> + cpufreq_freq_transition_begin(policy, &freqs); >> powernow_k6_target(policy, i); >> + cpufreq_freq_transition_end(policy, &freqs, 0); >> + break; >> + } >> } >> return 0; >> } > > Acked-by: Viresh Kumar <viresh.kumar@linaro.org> > -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/cpufreq/powernow-k6.c b/drivers/cpufreq/powernow-k6.c index 49f120e..6a4c34e 100644 --- a/drivers/cpufreq/powernow-k6.c +++ b/drivers/cpufreq/powernow-k6.c @@ -138,22 +138,14 @@ static void powernow_k6_set_cpu_multiplier(unsigned int best_i) static int powernow_k6_target(struct cpufreq_policy *policy, unsigned int best_i) { - struct cpufreq_freqs freqs; if (clock_ratio[best_i].driver_data > max_multiplier) { printk(KERN_ERR PFX "invalid target frequency\n"); return -EINVAL; } - freqs.old = busfreq * powernow_k6_get_cpu_multiplier(); - freqs.new = busfreq * clock_ratio[best_i].driver_data; - - cpufreq_freq_transition_begin(policy, &freqs); - powernow_k6_set_cpu_multiplier(best_i); - cpufreq_freq_transition_end(policy, &freqs, 0); - return 0; } @@ -228,8 +220,18 @@ static int powernow_k6_cpu_exit(struct cpufreq_policy *policy) { unsigned int i; for (i = 0; i < 8; i++) { - if (i == max_multiplier) + if (i == max_multiplier) { + struct cpufreq_freqs freqs; + + freqs.old = policy->cur; + freqs.new = clock_ratio[i].frequency; + freqs.flags = 0; + + cpufreq_freq_transition_begin(policy, &freqs); powernow_k6_target(policy, i); + cpufreq_freq_transition_end(policy, &freqs, 0); + break; + } } return 0; }
During frequency transitions, the cpufreq core takes the responsibility of invoking cpufreq_freq_transition_begin() and cpufreq_freq_transition_end() for those cpufreq drivers that define the ->target_index callback but don't set the ASYNC_NOTIFICATION flag. The powernow-k6 cpufreq driver falls under this category, but this driver was invoking the _begin() and _end() APIs itself around frequency transitions, which led to double invocation of the _begin() API. The _begin API makes contending callers wait until the previous invocation is complete. Hence, the powernow-k6 driver ended up waiting on itself, leading to system hangs during boot. Fix this by removing the calls to the _begin() and _end() APIs from the powernow-k6 driver, since they rightly belong to the cpufreq core. (Note that during ->exit(), the powernow-k6 driver sets the frequency without any help from the cpufreq core. So add explicit calls to the _begin() and _end() APIs around that frequency transition alone, to take care of that special case. Also, add a missing 'break' statement there.) Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> --- drivers/cpufreq/powernow-k6.c | 20 +++++++++++--------- 1 file changed, 11 insertions(+), 9 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html