From mboxrd@z Thu Jan  1 00:00:00 1970
From: will.deacon@arm.com (Will Deacon)
Date: Tue, 8 May 2012 19:04:30 +0100
Subject: [PATCH] Prevent process migration during vfp_init()
In-Reply-To: <CAOsG7tqr__yWp1zmMfy3myRGFmY690f8i8emuXq5d7E=o3E7Vw@mail.gmail.com>
References: <CAOsG7tqktqxxU9gtyqojP7KP87cy1jpgT8H==wOe0cQj8=LV+Q@mail.gmail.com>
 <20120508111332.GG2263@mudshark.cambridge.arm.com>
 <CAOsG7tqr__yWp1zmMfy3myRGFmY690f8i8emuXq5d7E=o3E7Vw@mail.gmail.com>
Message-ID: <20120508180430.GL2263@mudshark.cambridge.arm.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

On Tue, May 08, 2012 at 06:24:27PM +0100, Hyungwoo Yang wrote:
> I think we don't need preempt_disable()/preempt_enable() call in your proposal.
> I mean, since we have enabled on every VFPs in cores online(
> on_each_cpu() ), we don't need to worry about accessing a VFP which is
> disabled. So we don't need to worry about migration after
> "on_each_cpu()", right?

Yes, that sounds reasonable to me since any thread migration will imply the
barriers we need for the VFP exception vector to be correct. In which case the
patch can be further reduced to what I've got below.

It seems happy enough on my quad A9 running a bunch of paranoia FP tests --
do you have a particular testcase which was exhibiting this failure when you
reported the issue?

Cheers,

Will


diff --git a/arch/arm/vfp/vfpmodule.c b/arch/arm/vfp/vfpmodule.c
index bc683b8..c5767b5 100644
--- a/arch/arm/vfp/vfpmodule.c
+++ b/arch/arm/vfp/vfpmodule.c
@@ -11,6 +11,7 @@
 #include <linux/types.h>
 #include <linux/cpu.h>
 #include <linux/cpu_pm.h>
+#include <linux/hardirq.h>
 #include <linux/kernel.h>
 #include <linux/notifier.h>
 #include <linux/signal.h>
@@ -432,7 +433,10 @@ void VFP_bounce(u32 trigger, u32 fpexc, struct pt_regs *regs)
 
 static void vfp_enable(void *unused)
 {
-       u32 access = get_copro_access();
+       u32 access;
+
+       BUG_ON(preemptible());
+       access = get_copro_access();
 
        /*
         * Enable full access to VFP (cp10 and cp11)
@@ -657,7 +661,7 @@ static int __init vfp_init(void)
        unsigned int cpu_arch = cpu_architecture();
 
        if (cpu_arch >= CPU_ARCH_ARMv6)
-               vfp_enable(NULL);
+               on_each_cpu(vfp_enable, NULL, 1);
 
        /*
         * First check that there is a VFP that we can use.
@@ -678,8 +682,6 @@ static int __init vfp_init(void)
        } else {
                hotcpu_notifier(vfp_hotplug, 0);
 
-               smp_call_function(vfp_enable, NULL, 1);
-
                VFP_arch = (vfpsid & FPSID_ARCH_MASK) >> FPSID_ARCH_BIT;  /* Extract the architecture version */
                printk("implementor %02x architecture %d part %02x variant %x rev %x\n",
                        (vfpsid & FPSID_IMPLEMENTER_MASK) >> FPSID_IMPLEMENTER_BIT,