On 4.09.2024 3:56 PM, Konrad Dybcio wrote:
> On 4.09.2024 5:13 AM, Sibi Sankar wrote:
>> Ensure that the bad duplicates reported by the platform firmware doesn't
>> get added to the opp-tables.
>>
>> Signed-off-by: Sibi Sankar <quic_sibis@xxxxxxxxxxx>
>> ---
>> drivers/firmware/arm_scmi/perf.c | 13 +++++++++++--
>> 1 file changed, 11 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/firmware/arm_scmi/perf.c b/drivers/firmware/arm_scmi/perf.c
>> index 2d77b5f40ca7..114c3dd70ede 100644
>> --- a/drivers/firmware/arm_scmi/perf.c
>> +++ b/drivers/firmware/arm_scmi/perf.c
>> @@ -386,9 +386,11 @@ process_response_opp(struct device *dev, struct perf_dom_info *dom,
>> le16_to_cpu(r->opp[loop_idx].transition_latency_us);
>>
>> ret = xa_insert(&dom->opps_by_lvl, opp->perf, opp, GFP_KERNEL);
>> - if (ret)
>> + if (ret) {
>> dev_warn(dev, "Failed to add opps_by_lvl at %d for %s - ret:%d\n",
>> opp->perf, dom->info.name, ret);
>> + opp->perf = 0;
>> + }
>> }
>>
>> static inline void
>> @@ -404,9 +406,12 @@ process_response_opp_v4(struct device *dev, struct perf_dom_info *dom,
>> le16_to_cpu(r->opp[loop_idx].transition_latency_us);
>>
>> ret = xa_insert(&dom->opps_by_lvl, opp->perf, opp, GFP_KERNEL);
>> - if (ret)
>> + if (ret) {
>> dev_warn(dev, "Failed to add opps_by_lvl at %d for %s - ret:%d\n",
>> opp->perf, dom->info.name, ret);
>> + opp->perf = 0;
>> + return;
>> + }
>>
>> /* Note that PERF v4 reports always five 32-bit words */
>> opp->indicative_freq = le32_to_cpu(r->opp[loop_idx].indicative_freq);
>> @@ -871,6 +876,10 @@ static int scmi_dvfs_device_opps_add(const struct scmi_protocol_handle *ph,
>> else
>> freq = dom->opp[idx].indicative_freq * dom->mult_factor;
>>
>> + /* Skip all invalid frequencies reported by the firmware */
>> + if (!freq)
>> + continue;
>
> Maybe something like this instead? (not tested)
>
> diff --git a/drivers/firmware/arm_scmi/perf.c b/drivers/firmware/arm_scmi/perf.c
> index 2d77b5f40ca7..530692119c79 100644
> --- a/drivers/firmware/arm_scmi/perf.c
> +++ b/drivers/firmware/arm_scmi/perf.c
> @@ -431,8 +431,14 @@ iter_perf_levels_process_response(const struct scmi_protocol_handle *ph,
> {
> struct scmi_opp *opp;
> struct scmi_perf_ipriv *p = priv;
> + unsigned int idx = st->desc_index + st->loop_idx;
> +
> + opp = &p->perf_dom->opp[idx];
> +
> + /* Avoid duplicate entries coming from buggy firmware */
> + if (idx > 0 && opp->perf && p->perf_dom->opp[idx - 1].perf)
> + return 0;
>
> - opp = &p->perf_dom->opp[st->desc_index + st->loop_idx];
> if (PROTOCOL_REV_MAJOR(p->version) <= 0x3)
> process_response_opp(ph->dev, p->perf_dom, opp, st->loop_idx,
> response);
No that won't work, perf_dom->opp has all the entries and that's used
in e.g. scmi_dvfs_device_opps_add :/
Konrad
