linux-mips
[Top] [All Lists]

Re: [PATCH RESEND 16/17] MIPS: make oprofile use cp0_perfcount_irq if it

To: Sergei Shtylyov <sshtylyov@mvista.com>
Subject: Re: [PATCH RESEND 16/17] MIPS: make oprofile use cp0_perfcount_irq if it is set
From: Ralf Baechle <ralf@linux-mips.org>
Date: Thu, 12 Jan 2012 18:02:51 +0100
Cc: John Crispin <blogic@openwrt.org>, linux-mips@linux-mips.org, Felix Fietkau <nbd@openwrt.org>
In-reply-to: <4F0EFE6E.3080503@mvista.com>
References: <1326314674-9899-1-git-send-email-blogic@openwrt.org> <1326314674-9899-16-git-send-email-blogic@openwrt.org> <4F0EFE6E.3080503@mvista.com>
Sender: linux-mips-bounce@linux-mips.org
User-agent: Mutt/1.5.21 (2010-09-15)
On Thu, Jan 12, 2012 at 06:38:22PM +0300, Sergei Shtylyov wrote:

> >@@ -374,6 +379,10 @@ static int __init mipsxx_init(void)
> >     save_perf_irq = perf_irq;
> >     perf_irq = mipsxx_perfcount_handler;
> >
> >+    if (cp0_perfcount_irq>= 0)
> 
>    BTW, I just noticed. IRQ0 is not a valid IRQ in Linux,
> request_irq() should fail when passed 0, so this and following check
> should be '> 0'.

In a normal configuration that is in a discrete processor or in a MIPS
core where the performance IRQ is just routed back into the core the
lowest sensible value for cp0_perfcount_irq is 2, so there is no
immediate problem there.

IRQ 0 is ok for static use; dynamic use is problematic.  This case is
even more problematic because the interrupt might be shared with the
timer and the timer interrupt is allocated statically (see cevt-r4k.c)
but the performance counter interrupt later allocated dynamically with
IRQF_SHARED.

  Ralf

<Prev in Thread] Current Thread [Next in Thread>