When runtime_pm was originally added, it was done in rather confusing
way: omap2430_musb_init() (called from musb_init_controller) would do
runtime_pm_get_sync() and musb_init_controller() itself would do
runtime_pm_put to balance it out. This is not only confusing but also
wrong if non-omap2430 glue layer is used.
This confusion resulted in commit
772aed45b604 "usb: musb: fix
pm_runtime mismatch", that removed runtime_pm_put() from
musb_init_controller as that looked unbalanced, and also happened to
fix unrelated isp1704_charger crash. However this broke runtime PM
functionality (musb is now always powered, even without gadget active).
Avoid these confusing runtime pm dependences by making
musb_init_controller() and omap2430_musb_init() do their own runtime
get/put pairs; also cover error paths. Remove unneeded runtime_pm_put
in omap2430_remove too. isp1704_charger crash that motivated
772aed45b604 will be fixed by following patch.
Cc: Felipe Contreras <felipe.contreras@gmail.com>
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
if (!musb->isr) {
status = -ENODEV;
- goto fail3;
+ goto fail2;
}
if (!musb->xceiv->io_ops) {
musb->xceiv->io_ops = &musb_ulpi_access;
}
+ pm_runtime_get_sync(musb->controller);
+
#ifndef CONFIG_MUSB_PIO_ONLY
if (use_dma && dev->dma_mask) {
struct dma_controller *c;
goto fail5;
#endif
+ pm_runtime_put(musb->controller);
+
dev_info(dev, "USB %s mode controller at %p using %s, IRQ %d\n",
({char *s;
switch (musb->board_mode) {
musb_gadget_cleanup(musb);
fail3:
+ pm_runtime_put_sync(musb->controller);
+
+fail2:
if (musb->irq_wake)
device_init_wakeup(dev, 0);
musb_platform_exit(musb);
setup_timer(&musb_idle_timer, musb_do_idle, (unsigned long) musb);
+ pm_runtime_put_noidle(musb->controller);
return 0;
err1:
platform_device_del(glue->musb);
platform_device_put(glue->musb);
- pm_runtime_put(&pdev->dev);
kfree(glue);
return 0;