On Mon, Dec 23, 2013 at 08:47:01PM +0800, Shawn Guo wrote: > On Mon, Dec 23, 2013 at 07:02:23PM +0800, Dong Aisheng wrote: > > Sometimes we may meet the following lockdep issue. > > The root cause is .set_clock callback is executed with spin_lock_irqsave > > in sdhci_do_set_ios. However, the IMX set_clock callback will try to access > > clk_get_rate which is using a mutex lock. > > > > The fix avoids access mutex in .set_clock callback by initializing the > > pltfm_host->clock at probe time and use it later instead of calling > > clk_get_rate again in atomic context. > > While I agree this is the way less intrusive, I'm also wondering why > sdhci_do_set_ios() needs a spinlock at all, or at least it's > questionable if the lock should necessarily be held for such a long > running section. Per my understanding, it's for protecting the host structure including register r/w where can also be accessed from ISR, so we need it. > > Nevertheless, as long as we can guarantee that the rate of > pltfm_host->clk does not change after the driver is probed, I'm fine > with the patch. > We does not change the host parent clock always currently. Regards Dong Aisheng > Shawn > -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
