Hi,
On 2025. 01. 10. 8:56, Alexander Dahl wrote:
Hello,
found another suspicious area regarding pm_runtime in
atmel_qspi_probe(), see below (irrelevant stuff dropped this time) …
Am Thu, Nov 28, 2024 at 06:43:15PM +0100 schrieb Csókás, Bence:
From: Tudor Ambarus <tudor.ambarus@xxxxxxxxxxxxx>
[…]
pm_runtime_set_autosuspend_delay(&pdev->dev, 500);
pm_runtime_use_autosuspend(&pdev->dev);
@@ -745,7 +1444,9 @@ static int atmel_qspi_probe(struct platform_device *pdev)
pm_runtime_enable(&pdev->dev);
pm_runtime_get_noresume(&pdev->dev);
- atmel_qspi_init(aq);
+ err = atmel_qspi_init(aq);
+ if (err)
+ goto dma_release;
This error handling leads to unbalanced pm_runtime right? The block
below rolls back pm_runtime before jumping to 'dma_release', this here
does not.
err = spi_register_controller(ctrl);
if (err) {
@@ -753,13 +1454,16 @@ static int atmel_qspi_probe(struct platform_device *pdev)
pm_runtime_disable(&pdev->dev);
pm_runtime_set_suspended(&pdev->dev);
pm_runtime_dont_use_autosuspend(&pdev->dev);
- goto disable_qspick;
+ goto dma_release;
}
This hunk got it right?
Hm, I think this would be best handled by switching to
`devm_pm_runtime_enable()`... I'll add that to my upcoming "make
everything use devm_*()"-series... (but I'm waiting on new functions to
be added to dmaengine for that).
Bence
