Re: [PATCH v2] drm: Add the new api to install irq

[Maxime Ripard]

Hi!

On Tue, Nov 03, 2020 at 11:55:08AM +0100, Daniel Vetter wrote:
> On Tue, Nov 03, 2020 at 11:38:32AM +0100, Maxime Ripard wrote:
> > On Tue, Nov 03, 2020 at 11:10:27AM +0100, Thomas Zimmermann wrote:
> > > Hi
> > > 
> > > Am 03.11.20 um 10:52 schrieb Maxime Ripard:
> > > > On Tue, Nov 03, 2020 at 10:10:41AM +0800, Tian Tao wrote:
> > > >> Add new api devm_drm_irq_install() to register interrupts,
> > > >> no need to call drm_irq_uninstall() when the drm module is removed.
> > > >>
> > > >> v2:
> > > >> fixed the wrong parameter.
> > > >>
> > > >> Signed-off-by: Tian Tao <tiantao6@xxxxxxxxxxxxx>
> > > >> ---
> > > >>  drivers/gpu/drm/drm_drv.c | 23 +++++++++++++++++++++++
> > > >>  include/drm/drm_drv.h     |  3 ++-
> > > >>  2 files changed, 25 insertions(+), 1 deletion(-)
> > > >>
> > > >> diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c
> > > >> index cd162d4..0fe5243 100644
> > > >> --- a/drivers/gpu/drm/drm_drv.c
> > > >> +++ b/drivers/gpu/drm/drm_drv.c
> > > >> @@ -39,6 +39,7 @@
> > > >>  #include <drm/drm_color_mgmt.h>
> > > >>  #include <drm/drm_drv.h>
> > > >>  #include <drm/drm_file.h>
> > > >> +#include <drm/drm_irq.h>
> > > >>  #include <drm/drm_managed.h>
> > > >>  #include <drm/drm_mode_object.h>
> > > >>  #include <drm/drm_print.h>
> > > >> @@ -678,6 +679,28 @@ static int devm_drm_dev_init(struct device *parent,
> > > >>  	return ret;
> > > >>  }
> > > >>  
> > > >> +static void devm_drm_dev_irq_uninstall(void *data)
> > > >> +{
> > > >> +	drm_irq_uninstall(data);
> > > >> +}
> > > >> +
> > > >> +int devm_drm_irq_install(struct device *parent,
> > > >> +			 struct drm_device *dev, int irq)
> > > >> +{
> > > >> +	int ret;
> > > >> +
> > > >> +	ret = drm_irq_install(dev, irq);
> > > >> +	if (ret)
> > > >> +		return ret;
> > > >> +
> > > >> +	ret = devm_add_action(parent, devm_drm_dev_irq_uninstall, dev);
> > > >> +	if (ret)
> > > >> +		devm_drm_dev_irq_uninstall(dev);
> > > >> +
> > > >> +	return ret;
> > > >> +}
> > > >> +EXPORT_SYMBOL(devm_drm_irq_install);
> > > >> +
> > > > 
> > > > Shouldn't we tie the IRQ to the drm device (so with drmm_add_action)
> > > > instead of tying it to the underlying device?
> > > 
> > > If the HW device goes away, there won't be any more interrupts. So it's
> > > similar to devm_ functions for I/O memory. Why would you use the drmm_
> > > interface?
> > 
> > drm_irq_install/uninstall do more that just calling request_irq and
> > free_irq though, they will also run (among other things) the irq-related
> > hooks in the drm driver (irq_preinstall, irq_postinstall irq_uninstall)
> > and wake anything waiting for a vblank to occur, so we need the DRM
> > device and driver to still be around when we run drm_irq_uninstall.
> > That's why (I assume) you have to pass the drm_device as an argument and
> > not simply the device.
> 
> drm_device is guaranteed to outlive devm_, plus the hooks are meant to
> shut down hw. hw isn't around anymore when we do drmm_ cleanup, at least
> not in full generality.

drm_dev_put is either called through devm or in remove / unbind, and the
drm_device takes a reference on its parent device, so how can the
drm_device outlive its parent device?

> > This probably works in most case since you would allocate the drm_device
> > with devm_drm_dev_alloc, and then run drm_irq_install, so in the undoing
> > phase you would have first drm_irq_uninstall to run, and everything is
> > fine.
> > 
> > However, if one doesn't use devm_drm_dev_alloc but would use
> > devm_drm_irq_install, you would have first remove being called that
> > would free the drm device, and then drm_irq_uninstall which will use a
> > free'd pointer.
> 
> Don't do that, it's broken :-)

Well, yeah it's usually a pretty bad situation, but if we can fix it for
free it doesn't hurt?

Maxime

Attachment: signature.asc
Description: PGP signature