On 3/6/2020 3:50 AM, Doug Anderson wrote:
> Hi,
>
> On Thu, Mar 5, 2020 at 9:07 AM Maulik Shah <mkshah@xxxxxxxxxxxxxx> wrote:
>> TCSes have previously programmed data when rpmh_flush is called.
>> This can cause old data to trigger along with newly flushed.
>>
>> Fix this by cleaning SLEEP and WAKE TCSes before new data is flushed.
>>
>> Fixes: 600513dfeef3 ("drivers: qcom: rpmh: cache sleep/wake state requests")
>> Signed-off-by: Maulik Shah <mkshah@xxxxxxxxxxxxxx>
>> ---
>> drivers/soc/qcom/rpmh.c | 5 +++++
>> 1 file changed, 5 insertions(+)
>>
>> diff --git a/drivers/soc/qcom/rpmh.c b/drivers/soc/qcom/rpmh.c
>> index 1951f6a..63364ce 100644
>> --- a/drivers/soc/qcom/rpmh.c
>> +++ b/drivers/soc/qcom/rpmh.c
>> @@ -472,6 +472,11 @@ int rpmh_flush(struct rpmh_ctrlr *ctrlr)
>> return 0;
>> }
>>
>> + /* Invalidate the TCSes first to avoid stale data */
>> + do {
>> + ret = rpmh_rsc_invalidate(ctrlr_to_drv(ctrlr));
>> + } while (ret == -EAGAIN);
>> +
>> /* First flush the cached batch requests */
>> ret = flush_batch(ctrlr);
>> if (ret)
> I think you should make this patch 3/4 instead of 4/4, and then:
>
> 1. In this patch remove the call to rpmh_rsc_invalidate() in
> rpmh_invalidate(). You've already marked things "dirty" in
> invalidate_batch() so no need to actually program the hardware--it'll
> happen in the flush.
Done.
>
> 2. In patch 4/4 (the flushing patch) add a call to rpmh_flush() to
> rpmh_invalidate() if you're in non-OSI mode. Presumably you'll need a
> spinlock around the rpmh_flush() call?
With (1) addressed and rpmh_start_transaction and rpmh_end_transaction introduced in v13, this is not required.
Thanks,
Maulik
>
>
> The end result of that will be that rpmh_invalidate() will properly
> leave the non-batch sleep/wake sets programmed.
>
>
> -Doug
--
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation
