| Message ID | 20150224132805.385a56a5@notabene.brown (mailing list archive) |
|---|---|
| State | Accepted, archived |
| Headers | show |
On Tuesday, February 24, 2015 01:28:05 PM NeilBrown wrote: > > --Sig_/GFnrHDc1Bs/XVRdbvRQTN5O > Content-Type: text/plain; charset=US-ASCII > Content-Transfer-Encoding: quoted-printable > > > > Nested IRQs can only fire when the parent irq fires. > So when the parent is suspended, there is no need to suspend > the child irq. > > Suspending nested irqs can cause a problem is they are suspended or > resumed in the wrong order. > If an interrupt fires while the parent is active but the child is > suspended, then the interrupt will not be acknowledged properly > and so an interrupt storm can result. > This is particularly likely if the parent is resumed before > the child, and the interrupt was raised during suspend. > > Ensuring correct ordering would be possible, but it is simpler > to just never suspend nested interrupts. > > Signed-off-by: NeilBrown <neil@brown.name> We need to make some progress here. Can you please resend it again for people to have a fresh look (and CC PeterZ this time too)? > > --- > This is a resend of a patch sent at the end of January 2015. > Rafael seemed happy with it, but I receive no other response so I'm resendi= > ng. > > Thanks, > NeilBrown > > > diff --git a/kernel/irq/pm.c b/kernel/irq/pm.c > index 3ca532592704..40cbcfb7fc43 100644 > --- a/kernel/irq/pm.c > +++ b/kernel/irq/pm.c > @@ -118,6 +118,8 @@ void suspend_device_irqs(void) > unsigned long flags; > bool sync; > =20 > + if (irq_settings_is_nested_thread(desc)) > + continue; > raw_spin_lock_irqsave(&desc->lock, flags); > sync =3D suspend_device_irq(desc, irq); > raw_spin_unlock_irqrestore(&desc->lock, flags); > @@ -158,6 +160,8 @@ static void resume_irqs(bool want_early) > =20 > if (!is_early && want_early) > continue; > + if (irq_settings_is_nested_thread(desc)) > + continue; > =20 > raw_spin_lock_irqsave(&desc->lock, flags); > resume_irq(desc, irq); > > --Sig_/GFnrHDc1Bs/XVRdbvRQTN5O > Content-Type: application/pgp-signature > Content-Description: OpenPGP digital signature > > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v2 > > iQIVAwUBVOvhtTnsnt1WYoG5AQKnXxAAjQnaCA2UPMaoXcwoP/Qx1MJpMSjW6Q6v > b7NiJTm7ltcKoPBN4y2//p0Wk6BEKT+nCbfqFacj4+MxfjrC4qtjIpsL6lVxi2Iz > sBdbhdkKdUsX6ueRKWv5ifK9qbDzBje7QUcGDrJ/9n80rvbVB5AnSH8bM0TADs2W > bfg5xWAMLfBZq4yhMTgOkPx7D9LMk6/HmlR55ZEC7JvZo6yKwifyQi18k1wfDlLZ > FobxyB6/MexRpjG9RXLpp4MJLV7BV0b6wQhZWWYSFhxVcuYxb9u7BHWumtO9fCMy > fgIzSVsZHuPxTH+ct7fbj+Pat+Y38c5RYpq4ojWJP2FnrGjEnrPt+dpw6AnbkPme > JudzHus6HhgUu5vLdT/Hu5i1qhVcQVe7OaweqhGs1BReGYX7AVPAsVzJAJ09Fwyb > vNIbfuPoTWVI1cz9VV7VgbFHwZgq9/yhk9KhvRaypaMTeuWKbcf2SgiX315w2e21 > D3Gxa6FDKc+CmnDycrx0eALtfCDp86gR2yeOeiCPAGUp8ICJ5vvAJcnxYVPYzt64 > JBBILBwzRV/7LNzbkYWEFlq4CCm9HhmkN4/HDC7tMcxCqQoVFTK0eGn5nz9kFZ6x > MaUyhhakwK+cAx1bVmTZ+gdBxnaa31p26c2OFdv2VRFnJlFC2X+vc9GFuz8KUZCD > d43mKkWNdWQ= > =vmFh > -----END PGP SIGNATURE----- > > --Sig_/GFnrHDc1Bs/XVRdbvRQTN5O-- > -- > To unsubscribe from this list: send the line "unsubscribe linux-pm" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html >
diff --git a/kernel/irq/pm.c b/kernel/irq/pm.c index 3ca532592704..40cbcfb7fc43 100644 --- a/kernel/irq/pm.c +++ b/kernel/irq/pm.c @@ -118,6 +118,8 @@ void suspend_device_irqs(void) unsigned long flags; bool sync; + if (irq_settings_is_nested_thread(desc)) + continue; raw_spin_lock_irqsave(&desc->lock, flags); sync = suspend_device_irq(desc, irq); raw_spin_unlock_irqrestore(&desc->lock, flags); @@ -158,6 +160,8 @@ static void resume_irqs(bool want_early) if (!is_early && want_early) continue; + if (irq_settings_is_nested_thread(desc)) + continue; raw_spin_lock_irqsave(&desc->lock, flags); resume_irq(desc, irq);
Nested IRQs can only fire when the parent irq fires. So when the parent is suspended, there is no need to suspend the child irq. Suspending nested irqs can cause a problem is they are suspended or resumed in the wrong order. If an interrupt fires while the parent is active but the child is suspended, then the interrupt will not be acknowledged properly and so an interrupt storm can result. This is particularly likely if the parent is resumed before the child, and the interrupt was raised during suspend. Ensuring correct ordering would be possible, but it is simpler to just never suspend nested interrupts. Signed-off-by: NeilBrown <neil@brown.name> --- This is a resend of a patch sent at the end of January 2015. Rafael seemed happy with it, but I receive no other response so I'm resending. Thanks, NeilBrown