On 27-08-20, 11:48, Arnd Bergmann wrote:
[ 3.680477] dev_pm_opp_put_clkname+0x30/0x58 [ 3.683431] sdhci_msm_probe+0x284/0x9a0
dev_pm_opp_put_clkname() is part of the error handling in the probe function, so I would deduct there are two problems:
- something failed during the probe and the driver is trying to unwind
- the error handling it self is buggy and tries to undo something again that has already been undone.
Right.
This points to Viresh's d05a7238fe1c mmc: sdhci-msm: Unconditionally call dev_pm_opp_of_remove_table()
I completely forgot that Ulf already pushed this patch and I was wondering on which of the OPP core changes I wrote have done this :(
Most likely this is not the entire problem but it uncovered a preexisting bug.
I think this is.
Naresh: Can you please test with this diff ?
diff --git a/drivers/mmc/host/sdhci-msm.c b/drivers/mmc/host/sdhci-msm.c index b7e47107a31a..401839a97b57 100644 --- a/drivers/mmc/host/sdhci-msm.c +++ b/drivers/mmc/host/sdhci-msm.c @@ -2286,7 +2286,7 @@ static int sdhci_msm_probe(struct platform_device *pdev) ret = dev_pm_opp_of_add_table(&pdev->dev); if (ret != -ENODEV) { dev_err(&pdev->dev, "Invalid OPP table in Device tree\n"); - goto opp_cleanup; + goto opp_put_clkname; }
/* Vote for maximum clock rate for maximum performance */ @@ -2451,6 +2451,7 @@ static int sdhci_msm_probe(struct platform_device *pdev) msm_host->bulk_clks); opp_cleanup: dev_pm_opp_of_remove_table(&pdev->dev); +opp_put_clkname: dev_pm_opp_put_clkname(msm_host->opp_table); bus_clk_disable: if (!IS_ERR(msm_host->bus_clk))