Dear Arnd, Ulf:
Arnd Bergmann <arnd@xxxxxxxx> 於 2022年10月17日 週一 下午3:25寫道：
>
> On Sun, Oct 16, 2022, at 5:48 PM, Tony Huang wrote:
> > This is a patch for mmc driver for Sunplus SP7021 SOC.
> > Supports eMMC 4.41 DDR 104MB/s speed mode.
> >
> > Signed-off-by: Tony Huang <tonyhuang.sunplus@xxxxxxxxx>
>
> Looks ok to me me overall.
>
> Acked-by: Arnd Bergmann <arnd@xxxxxxxx>
>
> Just one more thing I noticed:
>
> > +#define SPMMC_TIMEOUT 500000
> ...
> > +static inline int spmmc_wait_finish(struct spmmc_host *host)
> > +{
> > + u32 state;
> > +
> > + return readl_poll_timeout_atomic(host->base + SPMMC_SD_STATE_REG,
> > state,
> > + (state & SPMMC_SDSTATE_FINISH), 1, SPMMC_TIMEOUT);
> > +}
> > +
> > +static inline int spmmc_wait_sdstatus(struct spmmc_host *host,
> > unsigned int status_bit)
> > +{
> > + u32 status;
> > +
> > + return readl_poll_timeout_atomic(host->base + SPMMC_SD_STATUS_REG,
> > status,
> > + (status & status_bit), 1, SPMMC_TIMEOUT);
> > +}
>
> 500ms seems like an awfully long time for a busy-wait, I wonder if this
> could be improved in some way. Is this always called from atomic context?
>
> If not, any callers from non-atomic context could use
> readl_poll_timeout() instead, or maybe there could be a shorter
> timeout in atomic context, with a fallback to a non-atomic
> workqueue if that times out, so only the MMC access will stall but
> not the entire system.
OK, I would use real_poll_timeout() instead.
Because I see "BUG: scheduling while atomic" issue before.
I have solved this problem.
>
> The same problem does appear to be in dw_mmc.c and mtk-sd.c but not
> in sdhci*.c, so I don't know if this is avoidable.
>
> Arnd
