From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: <4CC5F525.7040206@domain.hid> Date: Mon, 25 Oct 2010 23:22:45 +0200 From: Jan Kiszka MIME-Version: 1.0 References: <20101007115728.GA24500@domain.hid> <4CADBDC2.8080600@domain.hid> <20101008070148.GB2255@domain.hid> <1286530884.13186.109.camel@domain.hid> <20101013090353.GA6902@domain.hid> <1286961375.1759.71.camel@domain.hid> <20101013092617.GB6902@domain.hid> <1286981521.1759.83.camel@domain.hid> <1288025329.26618.132.camel@domain.hid> <4CC5C80E.2070004@domain.hid> <1288033731.26618.161.camel@domain.hid> <4CC5D742.9080307@domain.hid> <1288034435.26618.164.camel@domain.hid> <4CC5D8FF.5080109@domain.hid> <1288041166.26618.182.camel@domain.hid> In-Reply-To: <1288041166.26618.182.camel@domain.hid> Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="------------enig900302A116AC96596BFDF5CE" Sender: jan.kiszka@domain.hid Subject: Re: [Xenomai-help] kernel oopses when killing realtime task List-Id: Help regarding installation and common use of Xenomai List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Philippe Gerum Cc: xenomai@xenomai.org This is an OpenPGP/MIME signed message (RFC 2440 and 3156) --------------enig900302A116AC96596BFDF5CE Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: quoted-printable Am 25.10.2010 23:12, Philippe Gerum wrote: > On Mon, 2010-10-25 at 21:22 +0200, Jan Kiszka wrote: >> Am 25.10.2010 21:20, Philippe Gerum wrote: >>> On Mon, 2010-10-25 at 21:15 +0200, Jan Kiszka wrote: >>>> Am 25.10.2010 21:08, Philippe Gerum wrote: >>>>> On Mon, 2010-10-25 at 20:10 +0200, Jan Kiszka wrote: >>>>>> Am 25.10.2010 18:48, Philippe Gerum wrote: >>>>>>> On Wed, 2010-10-13 at 16:52 +0200, Philippe Gerum wrote:=20 >>>>>>>>> >>>>>>>>> Should we test IPIPE_STALL_FLAG on all but current CPUs? >>>>>>>> >>>>>>>> That would solve this particular issue, but we should drain the = pipeline >>>>>>>> out of any Xenomai critical section. The way it is done now may = induce a >>>>>>>> deadlock (e.g. CPU0 waiting for CPU1 to acknowledge critical ent= ry in >>>>>>>> ipipe_enter_critical when getting some IPI, and CPU1 waiting hw = IRQs off >>>>>>>> for CPU0 to release the Xenomai lock that annoys us right now). >>>>>>>> >>>>>>>> I'll come up with something hopefully better and tested in the n= ext >>>>>>>> days. >>>>>>>> >>>>>>> >>>>>>> Sorry for the lag. In case that helps, here is another approach, = based >>>>>>> on telling the pipeline to ignore the irq about to be detached, s= o that >>>>>>> it passes all further occurrences down to the next domain, withou= t >>>>>> >>>>>> Err, won't this irritate that next domain, ie. won't Linux dump wa= rnings >>>>>> about a spurious/unhandled IRQ? I think either the old handler sha= ll >>>>>> receive the last event or no one. >>>>> >>>>> Flipping the IRQ modes within a ipipe_critical_enter/exit section g= ives >>>>> you that guarantee. You are supposed to have disabled the irq line >>>>> before detaching, and critical IPIs cannot be acknowledged until al= l >>>>> CPUs have re-enabled interrupts at some point. Therefore, there are= only >>>>> two scenarii: >>>>> >>>>> - irq was disabled before delivery, and a pending interrupt is mask= ed by >>>>> the PIC and never delivered to the CPU. >>>>> >>>>> - an interrupt sneaked in before disabling, it is currently process= ed by >>>>> the pipeline in the low handler on some CPU, in which case interrup= ts >>>>> are off, so a critical IPI could be acked yet, and the irq mode bit= s >>>>> still allow dispatching to the target domain on that CPU. The assum= ption >>>>> which is happily made is that only head domains are interested in >>>>> un-virtualizing irqs, so the dispatch will happen immediately, whil= e the >>>>> handler is still valid (actually, we are not allowed to un-virtuali= ze >>>>> root irqs, and intermediate Adeos domains are already considered as= >>>>> endangered species, so this is fine). >>>>> >>>>>> >>>>>> Why this complex solution, why not simply draining (via critical_e= nter >>>>>> or whatever) - but _after_ xnintr_irq_detach, ie. while the relate= d >>>>>> resources are still valid? >>>>>> >>>>> >>>>> Because it's already too late. You have cleared the handler pointer= when >>>>> un-virtualizing via xnarch_release_irq, and the wired irq dispatche= r or >>>>> the log syncer on another CPU could then branch to eip $0. >>>> >>>> Just make ipipe_virtualize_irq install a nop handler instead of NULL= =2E >>> >>> This does not solve the issue of the last interrupt which should be >>> processed. You don't want to miss it. >> >> Don't understand. No interrupt is supposed to arrive anymore on >> deregistration, the last user is supposed to be down by now. We just >> need to catch though that slipped through. >=20 > No, we are handling the case when an interrupt is currently handled on = a > CPU which is not the one that unregisters the IRQ, and which managed to= > sneak in while the irq source was about to be masked in the PIC. This i= s > purely asynchronous for us in SMP, since we don't have irq descriptor > locks for the pipeline, we only have them at Xenomai level, which is on= e > step too far to protected our low level Adeos handler against that kind= > of race. Logically speaking, there is no reason why you would accept to= > leave that irq unhandled if it is there and known from a CPU (it was > actually the first point you raised). If that handler is already running, the IRQ will get handled, we just need to wait for the handler to finish after we returned from ipipe_virtualize_irq =3D> thus we need a barrier here. If the handler was about to run but we deregistered it a bit quicker, the IRQ need not be addressed at device level anymore. Reason: we already switched off any IRQ assertions in the device before we entered ipipe_virtualize_irq. So no harm is caused, that IRQ line is deasserted already (IOW: the IRQ became a spurious one while cleaning up). >=20 > The proper way to fix this issue would have been to fix xnintr_detach i= n > the first place, because calling ipipe_virtualize_irq() while holding a= > lock with irqs off is wrong. We could have then drained the pipeline > from the unregistering code. I'm rather going for a decent solution > which is not prone to regression for 2.x. >=20 Again, I see no reason for a more complex solution than avoiding that NULL pointer dereference at ipipe level as I suggested and adding a very simply system wide barrier right after dropping nklock in xnintr_detach. Jan --------------enig900302A116AC96596BFDF5CE Content-Type: application/pgp-signature; name="signature.asc" Content-Description: OpenPGP digital signature Content-Disposition: attachment; filename="signature.asc" -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.15 (GNU/Linux) Comment: Using GnuPG with SUSE - http://enigmail.mozdev.org/ iEYEARECAAYFAkzF9SUACgkQitSsb3rl5xTodgCeI+dcxMaEKFmlXbPTQbfCBNKP PLIAnR9s3lIzPM18WTQ/URg6ASlKxcrj =1PF/ -----END PGP SIGNATURE----- --------------enig900302A116AC96596BFDF5CE--