| Message ID | 1536165491-27813-1-git-send-email-sudeep.holla@arm.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | firmware: arm_scmi: fix divide by zero when sustained_perf_level is zero | expand |
Hi Sudeep, On Wednesday 05 Sep 2018 at 17:38:11 (+0100), Sudeep Holla wrote: > @@ -166,7 +166,12 @@ scmi_perf_domain_attributes_get(const struct scmi_handle *handle, u32 domain, > le32_to_cpu(attr->sustained_freq_khz); > dom_info->sustained_perf_level = > le32_to_cpu(attr->sustained_perf_level); > - dom_info->mult_factor = (dom_info->sustained_freq_khz * 1000) / > + if (!dom_info->sustained_freq_khz || > + !dom_info->sustained_perf_level) > + dom_info->mult_factor = 1; I'm sorry I missed that the first time I reviewed this patch, but after discussing with Ionela, we found out that there is actually a case where this could be a problem. If you have perf levels that are 1,2,3,4 (for example), then with mult_factor=1 you'll register OPPs at 1Hz, 2Hz, 3Hz, 4Hz into PM_OPP. And that will be turned into 0 KHz for all of them at the CPUFreq level when divided by 1000 in dev_pm_opp_init_cpufreq_table(). I guess a quick fix would be to have a default mult_factor of 1000 ... What do you think ? Thanks, Quentin
On 06/09/18 14:17, Quentin Perret wrote: > Hi Sudeep, > > On Wednesday 05 Sep 2018 at 17:38:11 (+0100), Sudeep Holla wrote: >> @@ -166,7 +166,12 @@ scmi_perf_domain_attributes_get(const struct scmi_handle *handle, u32 domain, >> le32_to_cpu(attr->sustained_freq_khz); >> dom_info->sustained_perf_level = >> le32_to_cpu(attr->sustained_perf_level); >> - dom_info->mult_factor = (dom_info->sustained_freq_khz * 1000) / >> + if (!dom_info->sustained_freq_khz || >> + !dom_info->sustained_perf_level) >> + dom_info->mult_factor = 1; > > I'm sorry I missed that the first time I reviewed this patch, but after > discussing with Ionela, we found out that there is actually a case where > this could be a problem. If you have perf levels that are 1,2,3,4 (for > example), then with mult_factor=1 you'll register OPPs at 1Hz, 2Hz, 3Hz, > 4Hz into PM_OPP. And that will be turned into 0 KHz for all of them at > the CPUFreq level when divided by 1000 in dev_pm_opp_init_cpufreq_table(). > Good find. > I guess a quick fix would be to have a default mult_factor of 1000 ... > I agree. > What do you think ? I will respin and send. -- Regards, Sudeep
diff --git a/drivers/firmware/arm_scmi/perf.c b/drivers/firmware/arm_scmi/perf.c index 721e6c57beae..51c08394026b 100644 --- a/drivers/firmware/arm_scmi/perf.c +++ b/drivers/firmware/arm_scmi/perf.c @@ -166,7 +166,12 @@ scmi_perf_domain_attributes_get(const struct scmi_handle *handle, u32 domain, le32_to_cpu(attr->sustained_freq_khz); dom_info->sustained_perf_level = le32_to_cpu(attr->sustained_perf_level); - dom_info->mult_factor = (dom_info->sustained_freq_khz * 1000) / + if (!dom_info->sustained_freq_khz || + !dom_info->sustained_perf_level) + dom_info->mult_factor = 1; + else + dom_info->mult_factor = + (dom_info->sustained_freq_khz * 1000) / dom_info->sustained_perf_level; memcpy(dom_info->name, attr->name, SCMI_MAX_STR_SIZE); }
Firmware can provide zero as values for sustained performance level and corresponding sustained frequency in kHz in order to hide the actual frequencies and provide only abstract values. It may endup with divide by zero scenario resulting in kernel panic. Let's set the multiplication factor to one if either one or both of them (sustained_perf_level and sustained_freq) are set to zero. Fixes: a9e3fbfaa0ff ("firmware: arm_scmi: add initial support for performance protocol") Reported-by: Ionela Voinescu <ionela.voinescu@arm.com> Signed-off-by: Sudeep Holla <sudeep.holla@arm.com> --- drivers/firmware/arm_scmi/perf.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) Hi ARM SoC team, Can you pick this patch directly ? Regards, Sudeep -- 2.7.4