From: Jack Steiner <steiner@sgi.com>
To: Ingo Molnar <mingo@elte.hu>
Cc: tglx@linutronix.de, hpa@zytor.com, x86@kernel.org,
linux-kernel@vger.kernel.org,
Peter Zijlstra <a.p.zijlstra@chello.nl>,
Cyrill Gorcunov <gorcunov@gmail.com>
Subject: Re: [PATCH] x86, UV: Fix NMI handler for UV platforms
Date: Mon, 21 Mar 2011 11:56:08 -0500 [thread overview]
Message-ID: <20110321165608.GA12718@sgi.com> (raw)
In-Reply-To: <20110321161425.GC23614@elte.hu>
On Mon, Mar 21, 2011 at 05:14:25PM +0100, Ingo Molnar wrote:
>
> * Jack Steiner <steiner@sgi.com> wrote:
>
> > This fixes a problem seen on UV systems handling NMIs from the node controller.
> > The original code used the DIE notifier as the hook to get to the UV NMI
> > handler. This does not work if performance counters are active - the hw_perf
> > code consumes the NMI and the UV handler is not called.
>
> Sigh:
Agree. X86 architecture does not make it easy to use NMIs from multiple sources.
>
> > --- linux.orig/arch/x86/kernel/traps.c 2011-03-21 09:05:43.000000000 -0500
> > +++ linux/arch/x86/kernel/traps.c 2011-03-21 09:13:01.306555675 -0500
> > @@ -57,6 +57,7 @@
> > #include <asm/mce.h>
> >
> > #include <asm/mach_traps.h>
> > +#include <asm/uv/uv.h>
> >
> > #ifdef CONFIG_X86_64
> > #include <asm/x86_init.h>
> > @@ -397,13 +398,16 @@ unknown_nmi_error(unsigned char reason,
> > static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
> > {
> > unsigned char reason = 0;
> > + int handled;
> >
> > /*
> > * CPU-specific NMI must be processed before non-CPU-specific
> > * NMI, otherwise we may lose it, because the CPU-specific
> > * NMI can not be detected/processed on other CPUs.
> > */
> > - if (notify_die(DIE_NMI, "nmi", regs, 0, 2, SIGINT) == NOTIFY_STOP)
> > + handled = uv_handle_nmi(regs, reason);
> > + if (notify_die(DIE_NMI, "nmi", regs, 0, 2, SIGINT) == NOTIFY_STOP ||
> > + handled)
> > return;
>
> Such code is extremely ugly. Please *reduce* the number of is_uv_system() type
> of hacks in core x86 code, not increase it!
>
> Any reason why a higher priority for the UV NMI handler cannot solve the 'perf
> eats the NMI' problem?
Yes. I tried that.
If the UV handler needs to know if hwperf is active in order to know whether or not
to return NOTIFY_STOP:
- if the UV NMI handler returns NOTIFY_STOP and hw_perf is active, hw_perf will miss
and NMI & counter sometimes stop working.
- if the UV NMI handler does not return NOTIFY_STOP and hw_perf is not active,
we get the "dazed" messages.
A cleaner solution would be to hide the platform specific NMI action in a x86_platform_ops
such as (untested):
Index: linux/arch/x86/include/asm/x86_init.h
===================================================================
--- linux.orig/arch/x86/include/asm/x86_init.h 2011-03-18 11:29:08.000000000 -0500
+++ linux/arch/x86/include/asm/x86_init.h 2011-03-21 11:52:36.413496546 -0500
@@ -153,6 +153,7 @@ struct x86_platform_ops {
void (*iommu_shutdown)(void);
bool (*is_untracked_pat_range)(u64 start, u64 end);
void (*nmi_init)(void);
+ int (*nmi_handler)(void *regs);
int (*i8042_detect)(void);
};
Index: linux/arch/x86/kernel/apic/x2apic_uv_x.c
===================================================================
--- linux.orig/arch/x86/kernel/apic/x2apic_uv_x.c 2011-03-21 11:40:36.000000000 -0500
+++ linux/arch/x86/kernel/apic/x2apic_uv_x.c 2011-03-21 11:45:14.134555108 -0500
@@ -115,6 +115,7 @@ static int __init uv_acpi_madt_oem_check
early_get_apic_pnode_shift();
x86_platform.is_untracked_pat_range = uv_is_untracked_pat_range;
x86_platform.nmi_init = uv_nmi_init;
+ x86_platform.nmi_handler = uv_nmi_handler;
if (!strcmp(oem_table_id, "UVL"))
uv_system_type = UV_LEGACY_APIC;
else if (!strcmp(oem_table_id, "UVX"))
Index: linux/arch/x86/kernel/traps.c
===================================================================
--- linux.orig/arch/x86/kernel/traps.c 2011-03-21 11:40:36.000000000 -0500
+++ linux/arch/x86/kernel/traps.c 2011-03-21 11:52:21.057498053 -0500
@@ -55,6 +55,8 @@
#include <asm/desc.h>
#include <asm/i387.h>
#include <asm/mce.h>
+#include <asm/x86_init.h>
+
#include <asm/mach_traps.h>
@@ -397,13 +399,16 @@ unknown_nmi_error(unsigned char reason,
static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
{
unsigned char reason = 0;
+ int handled;
/*
* CPU-specific NMI must be processed before non-CPU-specific
* NMI, otherwise we may lose it, because the CPU-specific
* NMI can not be detected/processed on other CPUs.
*/
- if (notify_die(DIE_NMI, "nmi", regs, 0, 2, SIGINT) == NOTIFY_STOP)
+ handled = x86_platform.nmi_handler(regs);
+ if (notify_die(DIE_NMI, "nmi", regs, 0, 2, SIGINT) == NOTIFY_STOP ||
+ handled)
return;
/* Non-CPU-specific NMI: NMI sources can be processed on any CPU */
Index: linux/arch/x86/kernel/x86_init.c
===================================================================
--- linux.orig/arch/x86/kernel/x86_init.c 2011-03-18 11:29:08.000000000 -0500
+++ linux/arch/x86/kernel/x86_init.c 2011-03-21 11:53:26.849496085 -0500
@@ -89,6 +89,7 @@ struct x86_cpuinit_ops x86_cpuinit __cpu
};
static void default_nmi_init(void) { };
+static int default_nmi_handler(void *regs) { return 1; };
static int default_i8042_detect(void) { return 1; };
struct x86_platform_ops x86_platform = {
@@ -98,6 +99,7 @@ struct x86_platform_ops x86_platform = {
.iommu_shutdown = iommu_shutdown_noop,
.is_untracked_pat_range = is_ISA_range,
.nmi_init = default_nmi_init,
+ .nmi_handler = default_nmi_handler,
.i8042_detect = default_i8042_detect
};
next prev parent reply other threads:[~2011-03-21 16:57 UTC|newest]
Thread overview: 38+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-03-21 16:01 [PATCH] x86, UV: Fix NMI handler for UV platforms Jack Steiner
2011-03-21 16:14 ` Ingo Molnar
2011-03-21 16:26 ` Cyrill Gorcunov
2011-03-21 16:43 ` Cyrill Gorcunov
2011-03-21 17:00 ` Cyrill Gorcunov
2011-03-21 17:08 ` Jack Steiner
2011-03-21 17:19 ` Cyrill Gorcunov
2011-03-21 17:34 ` Jack Steiner
2011-03-21 17:48 ` Cyrill Gorcunov
2011-03-21 17:55 ` Cyrill Gorcunov
2011-03-21 18:15 ` Cyrill Gorcunov
2011-03-21 18:24 ` Jack Steiner
2011-03-21 17:53 ` Don Zickus
2011-03-21 17:51 ` Don Zickus
2011-03-21 18:00 ` Cyrill Gorcunov
2011-03-21 18:22 ` Jack Steiner
2011-03-21 19:37 ` Don Zickus
2011-03-21 20:37 ` Jack Steiner
2011-03-22 17:11 ` Jack Steiner
2011-03-22 18:44 ` Don Zickus
2011-03-22 20:02 ` Jack Steiner
2011-03-22 21:25 ` Jack Steiner
2011-03-22 22:02 ` Cyrill Gorcunov
2011-03-23 13:36 ` Jack Steiner
2011-03-22 22:05 ` Don Zickus
2011-03-23 16:32 ` Jack Steiner
2011-03-23 17:53 ` Don Zickus
2011-03-23 20:00 ` Don Zickus
2011-03-23 20:41 ` Cyrill Gorcunov
2011-03-23 20:45 ` Cyrill Gorcunov
2011-03-23 21:22 ` Don Zickus
2011-03-23 20:46 ` Jack Steiner
2011-03-23 21:23 ` Don Zickus
2011-03-24 17:09 ` Jack Steiner
2011-03-24 18:43 ` Don Zickus
2011-03-21 16:56 ` Jack Steiner [this message]
2011-03-21 18:05 ` Ingo Molnar
2011-03-21 19:23 ` [PATCH V2] " Jack Steiner
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20110321165608.GA12718@sgi.com \
--to=steiner@sgi.com \
--cc=a.p.zijlstra@chello.nl \
--cc=gorcunov@gmail.com \
--cc=hpa@zytor.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=tglx@linutronix.de \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).