Messages in this thread | | | Date | Mon, 10 Mar 2008 10:41:54 -0700 | From | Andrew Morton <> | Subject | Re: [PATCH] sysrq show-all-cpus |
| |
On Mon, 10 Mar 2008 13:32:42 -0400 Rik van Riel <riel@redhat.com> wrote:
> SysRQ-P is not always useful on SMP systems, since it usually ends up showing the > backtrace of a CPU that is doing just fine, instead of the backtrace of the CPU > that is having problems. > > This patch adds SysRQ show-all-cpus(L), which shows the backtrace of every active > CPU in the system. It skips idle CPUs because some SMP systems are just too large > and we already know what the backtrace of the idle task looks like. > > Signed-off-by: Rik van Riel <riel@redhat.com> > > diff -up linux-2.6.25-rc3-mm1/drivers/char/sysrq.c.multicpu linux-2.6.25-rc3-mm1/drivers/char/sysrq.c > --- linux-2.6.25-rc3-mm1/drivers/char/sysrq.c.multicpu 2008-03-09 20:22:17.000000000 -0400 > +++ linux-2.6.25-rc3-mm1/drivers/char/sysrq.c 2008-03-10 13:03:00.000000000 -0400 > @@ -196,6 +196,44 @@ static struct sysrq_key_op sysrq_showloc > #define sysrq_showlocks_op (*(struct sysrq_key_op *)0) > #endif > > +#ifdef CONFIG_SMP > +static DEFINE_SPINLOCK(show_lock); > +static void showacpu(void *dummy) > +{ > + unsigned long flags; > + > + /* Idle CPUs have no interesting backtrace. */ > + if (idle_cpu(smp_processor_id())) > + return; > + > + spin_lock_irqsave(&show_lock, flags); > + printk("CPU%d:\n", smp_processor_id()); > + show_stack(NULL, NULL); > + spin_unlock_irqrestore(&show_lock, flags); > +} > +static void sysrq_showregs_othercpus(struct work_struct *dummy) > +{ > + smp_call_function(showacpu, NULL, 0, 0); > +} > +static DECLARE_WORK(sysrq_showallcpus, sysrq_showregs_othercpus); > + > +static void sysrq_handle_showallcpus(int key, struct tty_struct *tty) > +{ > + struct pt_regs *regs = get_irq_regs(); > + if (regs) { > + printk("CPU%d:\n", smp_processor_id()); > + show_regs(regs); > + } > + schedule_work(&sysrq_showallcpus); > +} > +static struct sysrq_key_op sysrq_showallcpus_op = { > + .handler = sysrq_handle_showallcpus, > + .help_msg = "show-all-cpus(L)", > + .action_msg = "Show backtrace of all active CPUs", > + .enable_mask = SYSRQ_ENABLE_DUMP, > +}; > +#endif
So the CPU which took the interrupt is always the first to display - that's nice.
Even if it's idle. I guess that's OK.
How come the sysrq-handling CPU uses show_regs() and the others use show_stack()? Do we not set up the get_irq_regs() data for IPIs?
> static void sysrq_handle_showregs(int key, struct tty_struct *tty) > { > struct pt_regs *regs = get_irq_regs(); > @@ -340,7 +378,11 @@ static struct sysrq_key_op *sysrq_key_ta > &sysrq_kill_op, /* i */ > NULL, /* j */ > &sysrq_SAK_op, /* k */ > +#ifdef CONFIG_SMP > + &sysrq_showallcpus_op, /* l */ > +#else > NULL, /* l */ > +#endif > &sysrq_showmem_op, /* m */ > &sysrq_unrt_op, /* n */ > /* o: This will often be registered as 'Off' at init time */
| |