Hi Geert, sorry for the mess! On Tue, Feb 20, 2024 at 10:03 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote: > sh_mobile_sdhi ee120000.mmc: mmc1 base at 0xee120000, max clock rate 12 MHz > mmc2: new high speed MMC card at address 0001 > sh_mobile_sdhi ee100000.mmc: mmc0 base at 0xee100000, max clock rate 88 MHz > mmcblk2: mmc2:0001 MMC08G 7.33 GiB Hey it reads some blocks... > BUG: sleeping function called from invalid context at kernel/workqueue.c:3347 > in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 35, name: irq/151-ee20000 (...) > __might_resched from __flush_work+0x20c/0x2e4 > __flush_work from __cancel_work_timer+0x118/0x198 > __cancel_work_timer from sh_mmcif_irqt+0x38/0x8f8 > sh_mmcif_irqt from irq_thread_fn+0x1c/0x58 Actually that is the thread so the message is a bit confusing, the irq thread isn't atomic. I wonder if it is caused by this: > > + sg_miter_start(&host->sg_miter, data->sg, data->sg_len, > > + SG_MITER_ATOMIC | SG_MITER_TO_SG); ...because I don't need to ask for atomic miter here, since the poll functions are actually called in process context. I've sent a patch, can you test? https://lore.kernel.org/linux-mmc/20240220-fix-sh-mmcif-v1-1-b9d08a787c1f@xxxxxxxxxx/T/#u Yours, Linus Walleij
