[patch 0/2] Immediate Values - in kernel users

[patch 0/2] Immediate Values - in kernel users

[Mathieu Desnoyers]

Here are the first users of the immediate values : f00f bug handler and
profiling.

It applies to 2.6.22-rc6-mm1 and depends on the Immediate Values.

Mathieu

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68

[patch 1/2] F00F bug fixup for i386 - use immediate values

[Mathieu Desnoyers]

[-- Attachment #1: f00f-bug-use-immediate-values.patch --]
[-- Type: text/plain, Size: 2749 bytes --]

Use the faster immediate values for F00F bug handling in do_page_fault.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
---
 arch/i386/kernel/traps.c     |    4 ++++
 arch/i386/mm/fault.c         |    3 ++-
 include/asm-i386/processor.h |    3 +++
 3 files changed, 9 insertions(+), 1 deletion(-)

Index: linux-2.6-lttng/arch/i386/kernel/traps.c
===================================================================
--- linux-2.6-lttng.orig/arch/i386/kernel/traps.c	2007-07-13 19:30:53.000000000 -0400
+++ linux-2.6-lttng/arch/i386/kernel/traps.c	2007-07-13 19:31:21.000000000 -0400
@@ -31,6 +31,7 @@
 #include <linux/uaccess.h>
 #include <linux/nmi.h>
 #include <linux/bug.h>
+#include <linux/immediate.h>
 
 #ifdef CONFIG_EISA
 #include <linux/ioport.h>
@@ -1158,6 +1159,8 @@
 #endif /* CONFIG_MATH_EMULATION */
 
 #ifdef CONFIG_X86_F00F_BUG
+immediate_char_t f00f_bug_fix __read_mostly;
+
 void __init trap_init_f00f_bug(void)
 {
 	__set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
@@ -1168,6 +1171,7 @@
 	 */
 	idt_descr.address = fix_to_virt(FIX_F00F_IDT);
 	load_idt(&idt_descr);
+	immediate_set_early(&f00f_bug_fix, 1);
 }
 #endif
 
Index: linux-2.6-lttng/arch/i386/mm/fault.c
===================================================================
--- linux-2.6-lttng.orig/arch/i386/mm/fault.c	2007-07-13 19:25:38.000000000 -0400
+++ linux-2.6-lttng/arch/i386/mm/fault.c	2007-07-13 19:30:55.000000000 -0400
@@ -25,6 +25,7 @@
 #include <linux/kprobes.h>
 #include <linux/uaccess.h>
 #include <linux/kdebug.h>
+#include <linux/immediate.h>
 
 #include <asm/system.h>
 #include <asm/desc.h>
@@ -492,7 +493,7 @@
 	/*
 	 * Pentium F0 0F C7 C8 bug workaround.
 	 */
-	if (boot_cpu_data.f00f_bug) {
+	immediate_if (&f00f_bug_fix) {
 		unsigned long nr;
 		
 		nr = (address - idt_descr.address) >> 3;
Index: linux-2.6-lttng/include/asm-i386/processor.h
===================================================================
--- linux-2.6-lttng.orig/include/asm-i386/processor.h	2007-07-13 19:25:38.000000000 -0400
+++ linux-2.6-lttng/include/asm-i386/processor.h	2007-07-13 19:30:55.000000000 -0400
@@ -21,6 +21,7 @@
 #include <asm/percpu.h>
 #include <linux/cpumask.h>
 #include <linux/init.h>
+#include <linux/immediate.h>
 #include <asm/processor-flags.h>
 
 /* flag for disabling the tsc */
@@ -102,6 +103,8 @@
 extern struct tss_struct doublefault_tss;
 DECLARE_PER_CPU(struct tss_struct, init_tss);
 
+extern immediate_char_t f00f_bug_fix;
+
 #ifdef CONFIG_SMP
 extern struct cpuinfo_x86 cpu_data[];
 #define current_cpu_data cpu_data[smp_processor_id()]

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68

Re: [patch 1/2] F00F bug fixup for i386 - use immediate values

[Alexey Dobriyan]

On Fri, Jul 13, 2007 at 09:26:43PM -0400, Mathieu Desnoyers wrote:
> Use the faster immediate values for F00F bug handling in do_page_fault.

> --- linux-2.6-lttng.orig/arch/i386/mm/fault.c
> +++ linux-2.6-lttng/arch/i386/mm/fault.c
> @@ -492,7 +493,7 @@
>  	/*
>  	 * Pentium F0 0F C7 C8 bug workaround.
>  	 */
> -	if (boot_cpu_data.f00f_bug) {
> +	immediate_if (&f00f_bug_fix) {

This code is not called during normal pagefaults and even during invalid
userspace accesses.

Out of curiosity, I inserted printk() at this place to see where I was
wrong. I got only two hits:

	Checking if this processor honours the WP bit even in supervisor mode... do_page_fault:
	Freeing unused kernel memory: 116k freed
	do_page_fault:

Resume: nobody gives a fuck about performance of this particular if,
	so conversion it totally pointless.

Re: [patch 1/2] F00F bug fixup for i386 - use immediate values

[Mathieu Desnoyers]

* Alexey Dobriyan (adobriyan@gmail.com) wrote:
> On Fri, Jul 13, 2007 at 09:26:43PM -0400, Mathieu Desnoyers wrote:
> > Use the faster immediate values for F00F bug handling in do_page_fault.
> 
> > --- linux-2.6-lttng.orig/arch/i386/mm/fault.c
> > +++ linux-2.6-lttng/arch/i386/mm/fault.c
> > @@ -492,7 +493,7 @@
> >  	/*
> >  	 * Pentium F0 0F C7 C8 bug workaround.
> >  	 */
> > -	if (boot_cpu_data.f00f_bug) {
> > +	immediate_if (&f00f_bug_fix) {
> 
> This code is not called during normal pagefaults and even during invalid
> userspace accesses.
> 
> Out of curiosity, I inserted printk() at this place to see where I was
> wrong. I got only two hits:
> 
> 	Checking if this processor honours the WP bit even in supervisor mode... do_page_fault:
> 	Freeing unused kernel memory: 116k freed
> 	do_page_fault:
> 
> Resume: nobody gives a fuck about performance of this particular if,
> 	so conversion it totally pointless.
> 

Interesting investigation, let's push it further:

instrumenting the f00f test site with a printk, I get:

[    0.000000] Checking if this processor honours the WP bit even in
supervisor mode... TEST: would test f00f bug at vadd ffecc000, eip c011928e
[    0.000000] Ok.
... and (whenever xdm restarts) :
[   64.768165] TEST: would test f00f bug at vadd 00000000, eip c0237596
[   64.787136] TEST: would test f00f bug at vadd 0000004c, eip c02375a2

Those EIPs are:

0xc011928e <do_test_wp_bit+20>: mov    %cl,0xffecd000(%edx)
-> Will trigger fixup_exception.

0xc0237596 <__copy_from_user_ll+53>:    rep movsl %ds:(%esi),%es:(%edi)
0xc02375a2 <__copy_from_user_ll+65>:    mov    0x20(%esi),%eax
-> Those look like user-space programs that gave NULL pointers to kernel
system calls.

I agree with you that this is not a "hot path". It was mostly a
straightforward test conversion.

Mathieu

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68

Re: [patch 1/2] F00F bug fixup for i386 - use immediate values

[Mathieu Desnoyers]

* Mathieu Desnoyers (mathieu.desnoyers@polymtl.ca) wrote:
> * Alexey Dobriyan (adobriyan@gmail.com) wrote:
> > On Fri, Jul 13, 2007 at 09:26:43PM -0400, Mathieu Desnoyers wrote:
> > > Use the faster immediate values for F00F bug handling in do_page_fault.
> > 
> > > --- linux-2.6-lttng.orig/arch/i386/mm/fault.c
> > > +++ linux-2.6-lttng/arch/i386/mm/fault.c
> > > @@ -492,7 +493,7 @@
> > >  	/*
> > >  	 * Pentium F0 0F C7 C8 bug workaround.
> > >  	 */
> > > -	if (boot_cpu_data.f00f_bug) {
> > > +	immediate_if (&f00f_bug_fix) {
> > 
> > This code is not called during normal pagefaults and even during invalid
> > userspace accesses.
> > 
> > Out of curiosity, I inserted printk() at this place to see where I was
> > wrong. I got only two hits:
> > 
> > 	Checking if this processor honours the WP bit even in supervisor mode... do_page_fault:
> > 	Freeing unused kernel memory: 116k freed
> > 	do_page_fault:
> > 
> > Resume: nobody gives a fuck about performance of this particular if,
> > 	so conversion it totally pointless.
> > 
> 
> Interesting investigation, let's push it further:
> 
> instrumenting the f00f test site with a printk, I get:
> 
> [    0.000000] Checking if this processor honours the WP bit even in
> supervisor mode... TEST: would test f00f bug at vadd ffecc000, eip c011928e
> [    0.000000] Ok.
> ... and (whenever xdm restarts) :
> [   64.768165] TEST: would test f00f bug at vadd 00000000, eip c0237596
> [   64.787136] TEST: would test f00f bug at vadd 0000004c, eip c02375a2
> 
> Those EIPs are:
> 
> 0xc011928e <do_test_wp_bit+20>: mov    %cl,0xffecd000(%edx)
> -> Will trigger fixup_exception.
> 
> 0xc0237596 <__copy_from_user_ll+53>:    rep movsl %ds:(%esi),%es:(%edi)
> 0xc02375a2 <__copy_from_user_ll+65>:    mov    0x20(%esi),%eax
> -> Those look like user-space programs that gave NULL pointers to kernel
> system calls.
> 
> I agree with you that this is not a "hot path". It was mostly a
> straightforward test conversion.
> 
> Mathieu

Actually, what is even more weird is that my system is configured to
output the segfaults, i.e. a small user-space program that does a memory
read of *(char*)NULL causes:
[ 1413.164212] null[3740]: segfault at 00000000 eip 0804836a esp
bffde2c0 error 4

But there is no message for the __copy_from_user_ll fault, so I wonder
1 - has the process been killed ?
2 - Is it just the printk that is missing ?

Mathieu

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68

[patch 2/2] Scheduler profiling - Use immediate values

[Mathieu Desnoyers]

[-- Attachment #1: profiling-use-immediate-values.patch --]
[-- Type: text/plain, Size: 6570 bytes --]

Use immediate values with lower d-cache hit in optimized version as a
condition for scheduler profiling call.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>

---
 drivers/kvm/svm.c       |    2 +-
 drivers/kvm/vmx.c       |    2 +-
 include/linux/profile.h |   10 ++++------
 kernel/profile.c        |   38 ++++++++++++++++++++++++++------------
 kernel/sched.c          |    3 ++-
 5 files changed, 34 insertions(+), 21 deletions(-)

Index: linux-2.6-lttng/kernel/profile.c
===================================================================
--- linux-2.6-lttng.orig/kernel/profile.c	2007-07-13 19:30:58.000000000 -0400
+++ linux-2.6-lttng/kernel/profile.c	2007-07-13 19:32:49.000000000 -0400
@@ -42,9 +42,6 @@
 static atomic_t *prof_buffer;
 static unsigned long prof_len, prof_shift;
 
-int prof_on __read_mostly;
-EXPORT_SYMBOL_GPL(prof_on);
-
 static cpumask_t prof_cpu_mask = CPU_MASK_ALL;
 #ifdef CONFIG_SMP
 static DEFINE_PER_CPU(struct profile_hit *[2], cpu_profile_hits);
@@ -52,6 +49,14 @@
 static DEFINE_MUTEX(profile_flip_mutex);
 #endif /* CONFIG_SMP */
 
+/* Immediate values */
+immediate_char_t sleep_profiling __read_mostly,
+			sched_profiling __read_mostly,
+			kvm_profiling __read_mostly,
+			cpu_profiling __read_mostly;
+EXPORT_SYMBOL_GPL(kvm_profiling);
+EXPORT_SYMBOL_GPL(cpu_profiling);
+
 static int __init profile_setup(char * str)
 {
 	static char __initdata schedstr[] = "schedule";
@@ -60,7 +65,7 @@
 	int par;
 
 	if (!strncmp(str, sleepstr, strlen(sleepstr))) {
-		prof_on = SLEEP_PROFILING;
+		immediate_set_early(&sleep_profiling, 1);
 		if (str[strlen(sleepstr)] == ',')
 			str += strlen(sleepstr) + 1;
 		if (get_option(&str, &par))
@@ -69,7 +74,7 @@
 			"kernel sleep profiling enabled (shift: %ld)\n",
 			prof_shift);
 	} else if (!strncmp(str, schedstr, strlen(schedstr))) {
-		prof_on = SCHED_PROFILING;
+		immediate_set_early(&sched_profiling, 1);
 		if (str[strlen(schedstr)] == ',')
 			str += strlen(schedstr) + 1;
 		if (get_option(&str, &par))
@@ -78,7 +83,7 @@
 			"kernel schedule profiling enabled (shift: %ld)\n",
 			prof_shift);
 	} else if (!strncmp(str, kvmstr, strlen(kvmstr))) {
-		prof_on = KVM_PROFILING;
+		immediate_set_early(&kvm_profiling, 1);
 		if (str[strlen(kvmstr)] == ',')
 			str += strlen(kvmstr) + 1;
 		if (get_option(&str, &par))
@@ -88,7 +93,7 @@
 			prof_shift);
 	} else if (get_option(&str, &par)) {
 		prof_shift = par;
-		prof_on = CPU_PROFILING;
+		immediate_set_early(&cpu_profiling, 1);
 		printk(KERN_INFO "kernel profiling enabled (shift: %ld)\n",
 			prof_shift);
 	}
@@ -99,7 +104,10 @@
 
 void __init profile_init(void)
 {
-	if (!prof_on) 
+	if (!_immediate_read(&sleep_profiling) &&
+		!_immediate_read(&sched_profiling) &&
+		!_immediate_read(&kvm_profiling) &&
+		!_immediate_read(&cpu_profiling))
 		return;
  
 	/* only text is profiled */
@@ -288,7 +296,7 @@
 	int i, j, cpu;
 	struct profile_hit *hits;
 
-	if (prof_on != type || !prof_buffer)
+	if (!prof_buffer)
 		return;
 	pc = min((pc - (unsigned long)_stext) >> prof_shift, prof_len - 1);
 	i = primary = (pc & (NR_PROFILE_GRP - 1)) << PROFILE_GRPSHIFT;
@@ -398,7 +406,7 @@
 {
 	unsigned long pc;
 
-	if (prof_on != type || !prof_buffer)
+	if (!prof_buffer)
 		return;
 	pc = ((unsigned long)__pc - (unsigned long)_stext) >> prof_shift;
 	atomic_add(nr_hits, &prof_buffer[min(pc, prof_len - 1)]);
@@ -555,7 +563,10 @@
 	}
 	return 0;
 out_cleanup:
-	prof_on = 0;
+	immediate_set_early(&sleep_profiling, 0);
+	immediate_set_early(&sched_profiling, 0);
+	immediate_set_early(&kvm_profiling, 0);
+	immediate_set_early(&cpu_profiling, 0);
 	smp_mb();
 	on_each_cpu(profile_nop, NULL, 0, 1);
 	for_each_online_cpu(cpu) {
@@ -582,7 +593,10 @@
 {
 	struct proc_dir_entry *entry;
 
-	if (!prof_on)
+	if (!_immediate_read(&sleep_profiling) &&
+		!_immediate_read(&sched_profiling) &&
+		!_immediate_read(&kvm_profiling) &&
+		!_immediate_read(&cpu_profiling))
 		return 0;
 	if (create_hash_tables())
 		return -1;
Index: linux-2.6-lttng/include/linux/profile.h
===================================================================
--- linux-2.6-lttng.orig/include/linux/profile.h	2007-07-13 19:30:58.000000000 -0400
+++ linux-2.6-lttng/include/linux/profile.h	2007-07-13 19:32:00.000000000 -0400
@@ -7,10 +7,12 @@
 #include <linux/init.h>
 #include <linux/cpumask.h>
 #include <linux/cache.h>
+#include <linux/immediate.h>
 
 #include <asm/errno.h>
 
-extern int prof_on __read_mostly;
+extern immediate_char_t sleep_profiling, sched_profiling, kvm_profiling,
+		cpu_profiling;
 
 #define CPU_PROFILING	1
 #define SCHED_PROFILING	2
@@ -35,11 +37,7 @@
  */
 static inline void profile_hit(int type, void *ip)
 {
-	/*
-	 * Speedup for the common (no profiling enabled) case:
-	 */
-	if (unlikely(prof_on == type))
-		profile_hits(type, ip, 1);
+	profile_hits(type, ip, 1);
 }
 
 #ifdef CONFIG_PROC_FS
Index: linux-2.6-lttng/kernel/sched.c
===================================================================
--- linux-2.6-lttng.orig/kernel/sched.c	2007-07-13 19:30:58.000000000 -0400
+++ linux-2.6-lttng/kernel/sched.c	2007-07-13 19:32:00.000000000 -0400
@@ -3241,7 +3241,8 @@
 	if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state))
 		__schedule_bug(prev);
 
-	profile_hit(SCHED_PROFILING, __builtin_return_address(0));
+	immediate_if (&sched_profiling)
+		profile_hit(SCHED_PROFILING, __builtin_return_address(0));
 
 	schedstat_inc(this_rq(), sched_cnt);
 }
Index: linux-2.6-lttng/drivers/kvm/svm.c
===================================================================
--- linux-2.6-lttng.orig/drivers/kvm/svm.c	2007-07-13 19:30:58.000000000 -0400
+++ linux-2.6-lttng/drivers/kvm/svm.c	2007-07-13 19:32:00.000000000 -0400
@@ -1654,7 +1654,7 @@
 	/*
 	 * Profile KVM exit RIPs:
 	 */
-	if (unlikely(prof_on == KVM_PROFILING))
+	immediate_if (&kvm_profiling)
 		profile_hit(KVM_PROFILING,
 			(void *)(unsigned long)vcpu->svm->vmcb->save.rip);
 
Index: linux-2.6-lttng/drivers/kvm/vmx.c
===================================================================
--- linux-2.6-lttng.orig/drivers/kvm/vmx.c	2007-07-13 19:30:58.000000000 -0400
+++ linux-2.6-lttng/drivers/kvm/vmx.c	2007-07-13 19:32:00.000000000 -0400
@@ -2156,7 +2156,7 @@
 	/*
 	 * Profile KVM exit RIPs:
 	 */
-	if (unlikely(prof_on == KVM_PROFILING))
+	immediate_if (&kvm_profiling)
 		profile_hit(KVM_PROFILING, (void *)vmcs_readl(GUEST_RIP));
 
 	vcpu->launched = 1;

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68