From mboxrd@z Thu Jan 1 00:00:00 1970 From: ebiederm@xmission.com (Eric W. Biederman) Date: Tue, 14 Nov 2006 18:08:12 +0000 Subject: Re: [patch] ia64: use generic_handle_irq() Message-Id: List-Id: References: <1163495289.4311.68.camel@ymzhang-perf.sh.intel.com> <20061114020517.2222dd08.akpm@osdl.org> <20061114124638.GA12848@elte.hu> In-Reply-To: <20061114124638.GA12848@elte.hu> (Ingo Molnar's message of "Tue, 14 Nov 2006 13:46:38 +0100") MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: Ingo Molnar Cc: Andrew Morton , "Zhang, Yanmin" , LKML , "linux-ia64@vger.kernel.org" , Thomas Gleixner Ingo Molnar writes: > * Andrew Morton wrote: > >> On Tue, 14 Nov 2006 17:08:10 +0800 >> "Zhang, Yanmin" wrote: >> >> > I got an oops when booting 2.6.19-rc5-mm1 on my ia64 machine. >> > >> > >> > The root cause is that some irq_chip variables, especially ia64_msi_chip, >> > initiate their memeber end to point to NULL. __do_IRQ doesn't check >> > if irq_chip->end is null and just calls it after processing the interrupt. >> > >> > As irq_chip->end is called at many places, so I fix it by reinitiating >> > irq_chip->end to dummy_irq_chip.end, e.g., a noop function. >> > >> > Below patch against 2.6.19-rc5-mm1 fixes it. >> > >> > Signed-off-by: Zhang Yanmin >> > >> > --- >> > >> > --- linux-2.6.19-rc5-mm1/kernel/irq/chip.c 2006-11-14 14:16:16.000000000 > +0800 >> > +++ linux-2.6.19-rc5-mm1_fix/kernel/irq/chip.c 2006-11-14 14:14:25.000000000 > +0800 >> > @@ -233,6 +233,8 @@ void irq_chip_set_defaults(struct irq_ch >> > chip->shutdown = chip->disable; >> > if (!chip->name) >> > chip->name = chip->typename; >> > + if (!chip->end) >> > + chip->end = dummy_irq_chip.end; >> > } >> > >> >> The same bug should be hitting in mainline, shouldn't it? > > correct. > > this bug comes from a 'mixed' IRQ setup on ia64: half of it is still > old-style, half of it (the MSI stuff) is new-style irqchip code. But the > ia64 lowlevel code unconditionally calls __do_IRQ(), which is a bug. Now that we hare half converted yes it is a bug. > the genirq code has all the right helpers for such a mixed situation: so > a better fix might be the one below: use generic_handle_irq() instead of > unconditionally calling into __do_IRQ(). But i have not tested it - > Yanmin, can you confirm this too fixes your bug? > > similarly, any architecture that makes use of the new generic MSI > infrastructure should not be calling __do_IRQ() directly. (but i'm not > aware of any other besides ia64 - i386 and x86_64 is now fully irq-chip > converted.) > > Eric, do you agree? > This is true in practice. It isn't necessarily true, as all of irq_chip structures are arch specific. It would be silly to start using msi interrupts without converting to genirq though. So I think this covers it for ia64. Your patch only fixes 2 spots and there is a third in irq_ia64.c that needs to be fixed as well. At least in linus's tree. > > Index: linux/arch/ia64/kernel/irq.c > =================================> --- linux.orig/arch/ia64/kernel/irq.c > +++ linux/arch/ia64/kernel/irq.c > @@ -197,7 +197,7 @@ void fixup_irqs(void) > struct pt_regs *old_regs = set_irq_regs(NULL); > > vectors_in_migration[irq]=0; > - __do_IRQ(irq); > + generic_handle_irq(irq); > set_irq_regs(old_regs); > } > } > Index: linux/arch/ia64/kernel/irq_ia64.c > =================================> --- linux.orig/arch/ia64/kernel/irq_ia64.c > +++ linux/arch/ia64/kernel/irq_ia64.c > @@ -186,7 +186,7 @@ ia64_handle_irq (ia64_vector vector, str > ia64_setreg(_IA64_REG_CR_TPR, vector); > ia64_srlz_d(); > > - __do_IRQ(local_vector_to_irq(vector)); > + generic_handle_irq(local_vector_to_irq(vector)); > > /* > * Disable interrupts and send EOI: