On Tue, Mar 29, 2022 at 4:42 PM Tony Huang <[email protected]> wrote:
>
> Add mmc driver for Sunplus SP7021
>
> Signed-off-by: Tony Huang <[email protected]>
There should be a description of the device in the changelog, not just the same
text as the subject.
> +static void spmmc_request(struct mmc_host *mmc, struct mmc_request *mrq)
> +{
> + struct spmmc_host *host = mmc_priv(mmc);
> + struct mmc_data *data;
> + struct mmc_command *cmd;
> + int ret;
> +
> + ret = mutex_lock_interruptible(&host->mrq_lock);
> + if (ret)
> + return;
I don't think it's valid to just return here when you get a signal. If
nothing can
handle the signal, doesn't it just hang?
It also appears that you don't release the mutex until the tasklet runs,
but it is not valid to release a mutex from a different context.
You should get a warning about this when running a kernel with lockdep
enabled at compile time. Please rework the locking to make this work.
> +#endif /* ifdef CONFIG_PM_RUNTIME */
> +
> +static const struct dev_pm_ops spmmc_pm_ops = {
> + SET_SYSTEM_SLEEP_PM_OPS(spmmc_pm_suspend, spmmc_pm_resume)
> +#ifdef CONFIG_PM_RUNTIME
> + SET_RUNTIME_PM_OPS(spmmc_pm_runtime_suspend, spmmc_pm_runtime_resume, NULL)
> +#endif
> +};
> +#endif /* ifdef CONFIG_PM */
It's better to use SYSTEM_SLEEP_PM_OPS/RUNTIME_PM_OPS instead
of the SET_ version, then you can remove all the #ifdef checks.
Arnd
Dear Arnd:
> > Add mmc driver for Sunplus SP7021
> >
> > Signed-off-by: Tony Huang <[email protected]>
>
> There should be a description of the device in the changelog, not just the same
> text as the subject.
OK, I will add description.
> > +static void spmmc_request(struct mmc_host *mmc, struct mmc_request
> > +*mrq) {
> > + struct spmmc_host *host = mmc_priv(mmc);
> > + struct mmc_data *data;
> > + struct mmc_command *cmd;
> > + int ret;
> > +
> > + ret = mutex_lock_interruptible(&host->mrq_lock);
> > + if (ret)
> > + return;
>
> I don't think it's valid to just return here when you get a signal. If nothing can
> handle the signal, doesn't it just hang?
>
> It also appears that you don't release the mutex until the tasklet runs, but it is
> not valid to release a mutex from a different context.
>
> You should get a warning about this when running a kernel with lockdep
> enabled at compile time. Please rework the locking to make this work.
>
Reomve code:
ret = mutex_lock_interruptible(&host->mrq_lock);
if (ret)
return;
Below is my modification:
. mutex_lock(&host->mrq_lock);
> > +#endif /* ifdef CONFIG_PM_RUNTIME */
> > +
> > +static const struct dev_pm_ops spmmc_pm_ops = {
> > + SET_SYSTEM_SLEEP_PM_OPS(spmmc_pm_suspend,
> spmmc_pm_resume)
> > +#ifdef CONFIG_PM_RUNTIME
> > + SET_RUNTIME_PM_OPS(spmmc_pm_runtime_suspend,
> > +spmmc_pm_runtime_resume, NULL) #endif }; #endif /* ifdef CONFIG_PM
> */
>
> It's better to use SYSTEM_SLEEP_PM_OPS/RUNTIME_PM_OPS instead of the
> SET_ version, then you can remove all the #ifdef checks.
>
I use SYSTEM_SLEEP_PM_OPS/RUNTIME_PM_OPS.
Compile shows error. Error: implicit declaration of function ? ? SYSTEM_SLEEP_PM_OPS? ? Did you mean ? ? SET_SYSTEM_SLEEP_PM_OPS? ? [-Werror=implicit-function-declaration]
I reference other mmc driver.
Below is my modification:
Compiler is pass.
#ifdef CONFIG_PM_SLEEP
static int spmmc_pm_suspend(struct device *dev)
{
pm_runtime_force_suspend(dev);
return 0;
}
static int spmmc_pm_resume(struct device *dev)
{
pm_runtime_force_resume(dev);
return 0;
}
#endif
#ifdef CONFIG_PM
static int spmmc_pm_runtime_suspend(struct device *dev)
{
struct spmmc_host *host;
host = dev_get_drvdata(dev);
clk_disable(host->clk);
return 0;
}
static int spmmc_pm_runtime_resume(struct device *dev)
{
struct spmmc_host *host;
host = dev_get_drvdata(dev);
return clk_enable(host->clk);
}
#endif
static const struct dev_pm_ops spmmc_pm_ops = {
SET_SYSTEM_SLEEP_PM_OPS(spmmc_pm_suspend, spmmc_pm_resume)
SET_RUNTIME_PM_OPS(spmmc_pm_runtime_suspend, spmmc_pm_runtime_resume, NULL)
};
On Thu, Mar 31, 2022 at 11:27 AM Tony Huang 黃懷厚 <[email protected]> wrote:
> > > +static void spmmc_request(struct mmc_host *mmc, struct mmc_request
> > > +*mrq) {
> > > + struct spmmc_host *host = mmc_priv(mmc);
> > > + struct mmc_data *data;
> > > + struct mmc_command *cmd;
> > > + int ret;
> > > +
> > > + ret = mutex_lock_interruptible(&host->mrq_lock);
> > > + if (ret)
> > > + return;
> >
> > I don't think it's valid to just return here when you get a signal. If nothing can
> > handle the signal, doesn't it just hang?
> >
> > It also appears that you don't release the mutex until the tasklet runs, but it is
> > not valid to release a mutex from a different context.
> >
> > You should get a warning about this when running a kernel with lockdep
> > enabled at compile time. Please rework the locking to make this work.
> >
> Reomve code:
> ret = mutex_lock_interruptible(&host->mrq_lock);
> if (ret)
> return;
>
> Below is my modification:
> . mutex_lock(&host->mrq_lock);
That addresses the problem with the signal handling, but not the lock
imbalance. Please fix that as well.
> >
> > It's better to use SYSTEM_SLEEP_PM_OPS/RUNTIME_PM_OPS instead of the
> > SET_ version, then you can remove all the #ifdef checks.
> >
>
> I use SYSTEM_SLEEP_PM_OPS/RUNTIME_PM_OPS.
> Compile shows error. Error: implicit declaration of function ? ? SYSTEM_SLEEP_PM_OPS? ? Did you mean ? ? SET_SYSTEM_SLEEP_PM_OPS? ? [-Werror=implicit-function-declaration]
Maybe you are on an old kernel release?
> I reference other mmc driver.
> Below is my modification:
> Compiler is pass.
>
> #ifdef CONFIG_PM_SLEEP
> static int spmmc_pm_suspend(struct device *dev)
> {
> pm_runtime_force_suspend(dev);
>
> return 0;
> }
We should fix the other drivers as well. For the moment, just do it
the right way now
instead of copying the #ifdefs.
Arnd
Dear Arnd:
> <[email protected]> wrote:
> > > > +static void spmmc_request(struct mmc_host *mmc, struct
> > > > +mmc_request
> > > > +*mrq) {
> > > > + struct spmmc_host *host = mmc_priv(mmc);
> > > > + struct mmc_data *data;
> > > > + struct mmc_command *cmd;
> > > > + int ret;
> > > > +
> > > > + ret = mutex_lock_interruptible(&host->mrq_lock);
> > > > + if (ret)
> > > > + return;
> > >
> > > I don't think it's valid to just return here when you get a signal.
> > > If nothing can handle the signal, doesn't it just hang?
> > >
> > > It also appears that you don't release the mutex until the tasklet
> > > runs, but it is not valid to release a mutex from a different context.
> > >
> > > You should get a warning about this when running a kernel with
> > > lockdep enabled at compile time. Please rework the locking to make this
> work.
> > >
> > Reomve code:
> > ret = mutex_lock_interruptible(&host->mrq_lock);
> > if (ret)
> > return;
> >
> > Below is my modification:
> > . mutex_lock(&host->mrq_lock);
>
> That addresses the problem with the signal handling, but not the lock
> imbalance. Please fix that as well.
Ok, I will modify lock imbalance issue.
> > >
> > > It's better to use SYSTEM_SLEEP_PM_OPS/RUNTIME_PM_OPS instead of
> the
> > > SET_ version, then you can remove all the #ifdef checks.
> > >
> >
> > I use SYSTEM_SLEEP_PM_OPS/RUNTIME_PM_OPS.
> > Compile shows error. Error: implicit declaration of function ? ?
> SYSTEM_SLEEP_PM_OPS? ? Did you mean ? ? SET_SYSTEM_SLEEP_PM_OPS? ?
> [-Werror=implicit-function-declaration]
>
> Maybe you are on an old kernel release?
>
OK,I will use new kernel release to compiler.
> > I reference other mmc driver.
> > Below is my modification:
> > Compiler is pass.
> >
> > #ifdef CONFIG_PM_SLEEP
> > static int spmmc_pm_suspend(struct device *dev)
> > {
> > pm_runtime_force_suspend(dev);
> >
> > return 0;
> > }
>
> We should fix the other drivers as well. For the moment, just do it the right
> way now instead of copying the #ifdefs.
>
OK, I will follow right way to remove #ifdef.