On 24 June 2014 17:52, Tomasz Figa <t.figa@xxxxxxxxxxx> wrote: > Hi Daniel, > > [adding Ulf, Chris and Mike to the discussion] > > On 24.06.2014 11:48, Daniel Drake wrote: >> sdhci_s3c_set_clock is called from sdhci_do_set_ios with interrupts >> disabled, and this calls into sdhci_s3c_consider_clock(). >> >> The patch "mmc: sdhci-s3c: Cache bus clock rates" addressed some >> scheduling while atomic in this function, but there are more issues >> here, seen while testing 3.16-rc2 on exynos4412: >> >> BUG: sleeping function called from invalid context at kernel/locking/mutex.c:103 >> in_atomic(): 1, irqs_disabled(): 128, pid: 75, name: mmcqd/0 >> Preemption disabled at:[< (null)>] (null) >> >> CPU: 0 PID: 75 Comm: mmcqd/0 Not tainted 3.16.0-rc2-00028-ge9fe7eb-dirty #77 >> [<c0016140>] (unwind_backtrace) from [<c0011e14>] (show_stack+0x10/0x14) >> [<c0011e14>] (show_stack) from [<c05ce8a8>] (dump_stack+0x84/0xc4) >> [<c05ce8a8>] (dump_stack) from [<c05d2de0>] (mutex_lock+0x1c/0x3c) >> [<c05d2de0>] (mutex_lock) from [<c046214c>] (clk_prepare_lock+0x6c/0xf4) >> [<c046214c>] (clk_prepare_lock) from [<c04625ac>] (clk_round_rate+0x10/0x2c) >> [<c04625ac>] (clk_round_rate) from [<c0447628>] (sdhci_s3c_set_clock+0x4c/0x1e8) >> [<c0447628>] (sdhci_s3c_set_clock) from [<c0447818>] >> (sdhci_cmu_set_clock+0x54/0x140) >> [<c0447818>] (sdhci_cmu_set_clock) from [<c0443a38>] >> (sdhci_do_set_ios+0x138/0x58c) >> [<c0443a38>] (sdhci_do_set_ios) from [<c0443864>] (sdhci_set_ios+0x28/0x34) >> >> clk_round_rate cannot be called here because it takes a mutex. >> >> sdhci_s3c_set_clock() also calls into clk_prepare_enable() which looks >> like it could trigger this problem too. > > Apparently this is related to generic sdhci code calling .set_clock() > under spin_lock_irqsave(). For reference, see sdhci_do_set_ios() and > sdhci_tasklet_finish(). > > Chris, Ulf, any suggestions? Well, the approach I think of is simply to invoke spin_unlock_irqrestore() prior the ->set_clock callbacks are invoked. The spinlock then needs to be handled by each ->set_clock callback for each driver. To simplify for ->set_clock callbacks, we might also want to add spinlock handling to the exported library function, sdhci_set_clock(). Kind regards Uffe -- 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
