remoteproc: qcom: pas: fix error handling in adsp_pds_enable

[ Upstream commit c0a6e5ee1ecfe4c3a5799cfd30820748eff5dfab ]

If the pm_runtime_get_sync failed in adsp_pds_enable when
loop (i), The unroll_pd_votes will start from (i - 1), and
it will resulted in following problems:

  1) pm_runtime_get_sync will increment pm usage counter even it
     failed. Forgetting to pm_runtime_put_noidle will result in
     reference leak.

  2) Have not reset pds[i] performance state.

Then we fix it.

Fixes: 17ee2fb4e8567 ("remoteproc: qcom: pas: Vote for active/proxy power domains")
Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com>
Link: https://lore.kernel.org/r/20201102143554.144707-1-zhangqilong3@huawei.com
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>

diff --git a/drivers/remoteproc/qcom_q6v5_pas.c b/drivers/remoteproc/qcom_q6v5_pas.c
index 3837f23..0678b41 100644 (file)
--- a/drivers/remoteproc/qcom_q6v5_pas.c
+++ b/drivers/remoteproc/qcom_q6v5_pas.c
@@ -90,8 +90,11 @@ static int adsp_pds_enable(struct qcom_adsp *adsp, struct device **pds,
        for (i = 0; i < pd_count; i++) {
                dev_pm_genpd_set_performance_state(pds[i], INT_MAX);
                ret = pm_runtime_get_sync(pds[i]);
-               if (ret < 0)
+               if (ret < 0) {
+                       pm_runtime_put_noidle(pds[i]);
+                       dev_pm_genpd_set_performance_state(pds[i], 0);
                        goto unroll_pd_votes;
+               }
        }
 
        return 0;