Message ID | 20190205110728.11451-1-miquel.raynal@bootlin.com |
---|---|
State | Accepted |
Delegated to: | David Miller |
Headers | show |
Series | [net-next,v3] net: dsa: mv88e6xxx: Prevent suspend to RAM | expand |
On Tue, Feb 05, 2019 at 12:07:28PM +0100, Miquel Raynal wrote: > On one hand, the mv88e6xxx driver has a work queue called in loop > which will attempt register accesses after MDIO bus suspension, that > entirely freezes the platform during suspend. > > On the other hand, the DSA core is not ready yet to support suspend to > RAM operation because so far there is no way to recover reliably the > switch configuration. > > To avoid the kernel to freeze when suspending with a switch driven by > the mv88e6xxx driver, we choose to prevent the driver suspension and > in the same way, the whole platform. > > Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com> Reviewed-by: Andrew Lunn <andrew@lunn.ch> Andrew
Hi Miquel, On Tue, 5 Feb 2019 12:07:28 +0100, Miquel Raynal <miquel.raynal@bootlin.com> wrote: > +/* There is no suspend to RAM support at DSA level yet, the switch configuration > + * would be lost after a power cycle so prevent it to be suspended. > + */ > +static int __maybe_unused mv88e6xxx_suspend(struct device *dev) > +{ > + return -EOPNOTSUPP; > +} > + > +static int __maybe_unused mv88e6xxx_resume(struct device *dev) > +{ > + return 0; > +} The code looks good but my only concern is -EOPNOTSUPP. In this context this code is specific to callbacks targeting bridge and switchdev, while the dev_pm_ops are completely parallel to DSA. It is intuitive but given Documentation/power/runtime_pm.txt, this will default to being interpreted as a fatal error, while -EBUSY seems to keep the device in an 'active' state in a saner way. I don't understand yet how to properly tell PM core that suspend to RAM isn't supported. If an error code different from -EAGAIN or -EBUSY is the way to go, I'm good with it: Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com> Thanks, Vivien
Hi Vivien, Vivien Didelot <vivien.didelot@gmail.com> wrote on Tue, 5 Feb 2019 11:28:57 -0500: > Hi Miquel, > > On Tue, 5 Feb 2019 12:07:28 +0100, Miquel Raynal <miquel.raynal@bootlin.com> wrote: > > > +/* There is no suspend to RAM support at DSA level yet, the switch configuration > > + * would be lost after a power cycle so prevent it to be suspended. > > + */ > > +static int __maybe_unused mv88e6xxx_suspend(struct device *dev) > > +{ > > + return -EOPNOTSUPP; > > +} > > + > > +static int __maybe_unused mv88e6xxx_resume(struct device *dev) > > +{ > > + return 0; > > +} > > The code looks good but my only concern is -EOPNOTSUPP. In this > context this code is specific to callbacks targeting bridge and > switchdev, while the dev_pm_ops are completely parallel to DSA. > > It is intuitive but given Documentation/power/runtime_pm.txt, this > will default to being interpreted as a fatal error, while -EBUSY > seems to keep the device in an 'active' state in a saner way. > > I don't understand yet how to properly tell PM core that suspend to RAM > isn't supported. If an error code different from -EAGAIN or -EBUSY > is the way to go, I'm good with it: I do share your concern and I went through the Documentation but I did not find a unified way to tell the PM core the feature is unsupported. By grepping code, I realized returning -EOPNOTSUPP was a recurrent alternative so here we are. I also considered -EBUSY but it seems more like a "I cannot right now" and -EAGAIN which is more a "try again soon". Anyway, no matter the error code returned, I'm not sure if the PM core actually cares? > Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com> > > > Thanks, > > Vivien Thanks, Miquèl
From: Miquel Raynal <miquel.raynal@bootlin.com> Date: Tue, 5 Feb 2019 12:07:28 +0100 > On one hand, the mv88e6xxx driver has a work queue called in loop > which will attempt register accesses after MDIO bus suspension, that > entirely freezes the platform during suspend. > > On the other hand, the DSA core is not ready yet to support suspend to > RAM operation because so far there is no way to recover reliably the > switch configuration. > > To avoid the kernel to freeze when suspending with a switch driven by > the mv88e6xxx driver, we choose to prevent the driver suspension and > in the same way, the whole platform. > > Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com> Applied to net-next, thanks.
On Tue 2019-02-05 12:07:28, Miquel Raynal wrote: > On one hand, the mv88e6xxx driver has a work queue called in loop > which will attempt register accesses after MDIO bus suspension, that > entirely freezes the platform during suspend. > > On the other hand, the DSA core is not ready yet to support suspend to > RAM operation because so far there is no way to recover reliably the > switch configuration. > > To avoid the kernel to freeze when suspending with a switch driven by > the mv88e6xxx driver, we choose to prevent the driver suspension and > in the same way, the whole platform. > > Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com> Could we at least do printk() so that user knows what went wrong? Debugging s2ram is usually not easy :-(. Pavel
Hi Pavel, Pavel Machek <pavel@ucw.cz> wrote on Mon, 8 Apr 2019 23:55:41 +0200: > On Tue 2019-02-05 12:07:28, Miquel Raynal wrote: > > On one hand, the mv88e6xxx driver has a work queue called in loop > > which will attempt register accesses after MDIO bus suspension, that > > entirely freezes the platform during suspend. > > > > On the other hand, the DSA core is not ready yet to support suspend to > > RAM operation because so far there is no way to recover reliably the > > switch configuration. > > > > To avoid the kernel to freeze when suspending with a switch driven by > > the mv88e6xxx driver, we choose to prevent the driver suspension and > > in the same way, the whole platform. > > > > Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com> > > Could we at least do printk() so that user knows what went wrong? > > Debugging s2ram is usually not easy :-(. I suppose you will be told that suspend was refused by a driver (probably without stating which one though). You may send a patch to add a trace if you think it is important, as this change as already been merged. Thanks, Miquèl
On Tue, Apr 09, 2019 at 09:06:43AM +0200, Miquel Raynal wrote: > Hi Pavel, > > Pavel Machek <pavel@ucw.cz> wrote on Mon, 8 Apr 2019 23:55:41 +0200: > > > On Tue 2019-02-05 12:07:28, Miquel Raynal wrote: > > > On one hand, the mv88e6xxx driver has a work queue called in loop > > > which will attempt register accesses after MDIO bus suspension, that > > > entirely freezes the platform during suspend. > > > > > > On the other hand, the DSA core is not ready yet to support suspend to > > > RAM operation because so far there is no way to recover reliably the > > > switch configuration. > > > > > > To avoid the kernel to freeze when suspending with a switch driven by > > > the mv88e6xxx driver, we choose to prevent the driver suspension and > > > in the same way, the whole platform. > > > > > > Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com> > > > > Could we at least do printk() so that user knows what went wrong? > > > > Debugging s2ram is usually not easy :-(. > > I suppose you will be told that suspend was refused by a driver > (probably without stating which one though). You may send a patch to add > a trace if you think it is important, as this change as already been > merged. Hi Miquel, Pavel I don't know the s2ram code at all, but this seems like a generic problem. The core should be reporting which driver returned an error, not the driver itself. Andrew
=================== * After having discussed with Vivien and Andrew, it seems that preventing the mv88e6xxx driver to suspend is the best option we have as of today. I removed the code saving the switch rules at driver level, I forgot about saving them at DSA level, so here are just two dummy PM hooks that prevent suspension. For people interested in picking-up what has been written and continue working on it, my initial proposal is available there: https://lore.kernel.org/netdev/20190130104606.31639abb@xps13/T/ drivers/net/dsa/mv88e6xxx/chip.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) diff --git a/drivers/net/dsa/mv88e6xxx/chip.c b/drivers/net/dsa/mv88e6xxx/chip.c index 8a517d8fb9d1..28764d6d7388 100644 --- a/drivers/net/dsa/mv88e6xxx/chip.c +++ b/drivers/net/dsa/mv88e6xxx/chip.c @@ -4651,6 +4651,21 @@ static const void *pdata_device_get_match_data(struct device *dev) return NULL; } +/* There is no suspend to RAM support at DSA level yet, the switch configuration + * would be lost after a power cycle so prevent it to be suspended. + */ +static int __maybe_unused mv88e6xxx_suspend(struct device *dev) +{ + return -EOPNOTSUPP; +} + +static int __maybe_unused mv88e6xxx_resume(struct device *dev) +{ + return 0; +} + +static SIMPLE_DEV_PM_OPS(mv88e6xxx_pm_ops, mv88e6xxx_suspend, mv88e6xxx_resume); + static int mv88e6xxx_probe(struct mdio_device *mdiodev) { struct dsa_mv88e6xxx_pdata *pdata = mdiodev->dev.platform_data; @@ -4835,6 +4850,7 @@ static struct mdio_driver mv88e6xxx_driver = { .mdiodrv.driver = { .name = "mv88e6085", .of_match_table = mv88e6xxx_of_match, + .pm = &mv88e6xxx_pm_ops, }, };
On one hand, the mv88e6xxx driver has a work queue called in loop which will attempt register accesses after MDIO bus suspension, that entirely freezes the platform during suspend. On the other hand, the DSA core is not ready yet to support suspend to RAM operation because so far there is no way to recover reliably the switch configuration. To avoid the kernel to freeze when suspending with a switch driven by the mv88e6xxx driver, we choose to prevent the driver suspension and in the same way, the whole platform. Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com> --- Changes since v1/v2