Re: [PATCH] take 2 of the tr-based current

Re: [PATCH] take 2 of the tr-based current

[Benjamin LaHaise]

On Thu, Nov 08, 2001 at 07:05:46PM -0500, Benjamin LaHaise wrote:
> If other people could bang on this a bit and post any problems, I'd 
> appreciate it.  tia,

Ooops.  A slight typo is fixed below.

		-ben (who shouldn't edit patches before hitting send)


diff -ur ./v2.4.13-ac8/include/asm-i386/smp.h ../toomuch-v2.4.13-ac8+tr/include/asm-i386/smp.h
--- ./v2.4.13-ac8/include/asm-i386/smp.h	Thu Nov  8 21:07:47 2001
+++ toomuch-v2.4.13-ac8+tr/include/asm-i386/smp.h	Thu Nov  8 21:06:25 2001
@@ -102,7 +102,8 @@
  * so this is correct in the x86 case.
  */
 
-static unsigned get_TR(void) __attribute__ ((pure))
+static unsigned get_TR(void) __attribute__ ((pure));
+static unsigned get_TR(void)
 {
 	unsigned tr;
 	__asm__("str %w0" : "=g" (tr));

Re: [PATCH] take 2 of the tr-based current

[Jakub Jelinek]

On Thu, Nov 08, 2001 at 09:11:43PM -0500, Benjamin LaHaise wrote:
> On Thu, Nov 08, 2001 at 07:05:46PM -0500, Benjamin LaHaise wrote:
> > If other people could bang on this a bit and post any problems, I'd 
> > appreciate it.  tia,
> 
> Ooops.  A slight typo is fixed below.
> 
> 		-ben (who shouldn't edit patches before hitting send)
> 
> 
> diff -ur ./v2.4.13-ac8/include/asm-i386/smp.h ../toomuch-v2.4.13-ac8+tr/include/asm-i386/smp.h
> --- ./v2.4.13-ac8/include/asm-i386/smp.h	Thu Nov  8 21:07:47 2001
> +++ toomuch-v2.4.13-ac8+tr/include/asm-i386/smp.h	Thu Nov  8 21:06:25 2001
> @@ -102,7 +102,8 @@
>   * so this is correct in the x86 case.
>   */
>  
> -static unsigned get_TR(void) __attribute__ ((pure))
> +static unsigned get_TR(void) __attribute__ ((pure));
> +static unsigned get_TR(void)
>  {
>  	unsigned tr;
>  	__asm__("str %w0" : "=g" (tr));

Why not
static inline unsigned __attribute__ ((const)) get_TR(void)
{
}
?
If TR register only ever changes during cpu_init, I don't see why you
cannot use const. Using pure would mean if you do get_TR, then store
something into global memory and do get_TR again, it will be done twice.
Also, I wonder why you don't inline it.

	Jakub

Re: [PATCH] take 2 of the tr-based current

[Manfred Spraul]

Jakub Jelinek wrote:
> 
> On Thu, Nov 08, 2001 at 09:11:43PM -0500, Benjamin LaHaise wrote:
> > -static unsigned get_TR(void) __attribute__ ((pure))
> > +static unsigned get_TR(void) __attribute__ ((pure));
> > +static unsigned get_TR(void)
> >  {
> >       unsigned tr;
> >       __asm__("str %w0" : "=g" (tr));
> 
> Why not
> static inline unsigned __attribute__ ((const)) get_TR(void)
> {
> }
> ?
> If TR register only ever changes during cpu_init, I don't see why you
> cannot use const.

The task register is only pure, not const. It's true that it's only
initialized during cpu_init(), but different cpus in the system have
different task register values.
	get_TR();
	schedule();
	get_TR();
must reload the task register.

But the value of "current" can be considered as const: It never changes
for a running thread. The exception are fork() and clone(), but both
functions directly call asm code and run on a fresh stack after forking.

> Using pure would mean if you do get_TR, then store
> something into global memory and do get_TR again, it will be done twice.
> Also, I wonder why you don't inline it.
>

ben?
I think get_current should be inline, const. get_TR() and
smp_processor_id would be inline, pure.

--
	Manfred

Re: [PATCH] take 2 of the tr-based current

[David S. Miller]

   From: Manfred Spraul <manfred@colorfullife.com>
   Date: Fri, 09 Nov 2001 15:54:03 +0100

   Jakub Jelinek wrote:
   > If TR register only ever changes during cpu_init, I don't see why you
   > cannot use const.
   
   The task register is only pure, not const.

As far as what the compiler can see or care about, it is
const.

Franks a lot,
David S. Miller
davem@redhat.com

Re: [PATCH] take 2 of the tr-based current

[Manfred Spraul]

"David S. Miller" wrote:
> 
>    From: Manfred Spraul <manfred@colorfullife.com>
>    Date: Fri, 09 Nov 2001 15:54:03 +0100
> 
>    Jakub Jelinek wrote:
>    > If TR register only ever changes during cpu_init, I don't see why you
>    > cannot use const.
> 
>    The task register is only pure, not const.
> 
> As far as what the compiler can see or care about, it is
> const.
>
No. const == never changes.
get_TR changes if a task calls schedule, and return on another cpu.

<<<
+static unsigned get_TR(void) __attribute__ ((pure))
+{
+       unsigned tr;
+       __asm__("str %w0" : "=g" (tr));
+       return tr;
+}
+
+#define smp_processor_id()     ( ((get_TR() >> 3) - __FIRST_TSS_ENTRY)
>> 2 )
<<<
smp_processor_id() is definitively not const.

OTHO 'current' is const.
--
	Manfred

Re: [PATCH] take 2 of the tr-based current

[Richard B. Johnson]

On Fri, 9 Nov 2001, Manfred Spraul wrote:

> "David S. Miller" wrote:
> > 
> >    From: Manfred Spraul <manfred@colorfullife.com>
> >    Date: Fri, 09 Nov 2001 15:54:03 +0100
> > 
[SNIPPED...]

> smp_processor_id() is definitively not const.
> 
> OTHO 'current' is const.
> --

Not as far as the compiler is concerned! You don't want the compiler
to generate code that reads 'current' once upon startup, and never
again. Further, you don't want that pointer to exist in .rodata, you
want it with the changable variables in .data or .bss .

Cheers,
Dick Johnson

Penguin : Linux version 2.4.1 on an i686 machine (799.53 BogoMips).

    I was going to compile a list of innovations that could be
    attributed to Microsoft. Once I realized that Ctrl-Alt-Del
    was handled in the BIOS, I found that there aren't any.

Re: [PATCH] take 2 of the tr-based current

[Anton Blanchard]

Hi,

> No. const == never changes.
> get_TR changes if a task calls schedule, and return on another cpu.

Yes, I found this exact problem on ppc64 where we would cache the
processor data area across a schedule(). What was interesting was that
__attribute__ ((pure)) was not enough to fix this.

static inline struct Paca *get_paca(void) __attribute__ ((pure));
static inline struct Paca *get_paca(void)
{
	struct Paca *rval;
	__asm__ ("mfspr %0,0x113" : "=r" (rval));
	return rval;
}

Alan Modra came to the rescue and found that gcc was optimising too much
and since the function did not touch any global variables, it would
upgrade the pure to const. This was on gcc 3.0.X.

Anton

Re: [PATCH] take 2 of the tr-based current

[Benjamin LaHaise]

On Sun, Nov 11, 2001 at 11:01:08AM +1100, Anton Blanchard wrote:
> static inline struct Paca *get_paca(void) __attribute__ ((pure));
> static inline struct Paca *get_paca(void)
> {
> 	struct Paca *rval;
> 	__asm__ ("mfspr %0,0x113" : "=r" (rval));
> 	return rval;
> }
> 
> Alan Modra came to the rescue and found that gcc was optimising too much
> and since the function did not touch any global variables, it would
> upgrade the pure to const. This was on gcc 3.0.X.

Hmmm.  Would adding a fake global input help with that?  Something like a 
"g" (aligned_data) input.

		-ben
-- 
Fish.

Re: [PATCH] take 2 of the tr-based current

[Anton Blanchard]

 
> Hmmm.  Would adding a fake global input help with that?  Something like a 
> "g" (aligned_data) input.

Yep thats what Alan ended up doing, it needed a dummy memory constraint
to prevent gcc over optimising:

        __asm__ ("mfspr %0,0x113; #%1"
	: "=r" (rval)
	: "m" ((int *) 0));

Anton

Re: [PATCH] take 2 of the tr-based current

[Manfred Spraul]

[-- Attachment #1: Type: text/plain, Size: 1053 bytes --]

Anton Blanchard wrote:

>Hi,
>
>>No. const == never changes.
>>get_TR changes if a task calls schedule, and return on another cpu.
>>
>
>Yes, I found this exact problem on ppc64 where we would cache the
>processor data area across a schedule(). What was interesting was that
>__attribute__ ((pure)) was not enough to fix this.
>
>static inline struct Paca *get_paca(void) __attribute__ ((pure));
>static inline struct Paca *get_paca(void)
>{
>	struct Paca *rval;
>	__asm__ ("mfspr %0,0x113" : "=r" (rval));
>	return rval;
>}
>
>Alan Modra came to the rescue and found that gcc was optimising too much
>and since the function did not touch any global variables, it would
>upgrade the pure to const. This was on gcc 3.0.X.
>
But the function called schedule - mustn't gcc assume that schedule 
writes into global variables?
As far as I can see that sounds like a gcc bug.

Could you try how many get_cpu calls are generated by the attached testapp?
the i386 RH compilers generate correct code:
gcc-2.96-98: 4 calls.
gcc3-3.0.1-3: 2 calls.

--
    Manfred

[-- Attachment #2: test.c --]
[-- Type: application/msword, Size: 487 bytes --]

Re: [PATCH] take 2 of the tr-based current

[Anton Blanchard]

Hi,

> But the function called schedule - mustn't gcc assume that schedule 
> writes into global variables?
> As far as I can see that sounds like a gcc bug.

Yes gcc knows we need to reload across a function call, but it also
knows that the get_cpu function uses no global variables.

> Could you try how many get_cpu calls are generated by the attached testapp?

I changed the code a bit so that get_cpu() is now inline - this
represents our situation better. I think it is valid for gcc to cache
get_cpu across a function call in the below example because it knows
that get_cpu does not refer to any global variables.

I brought it up in case gcc optimises your get_tr the same way (I cant
remember what the operand constraints on it were now, if it was only a
register then you might see it).

(The disassembly of the below has only one mfspr and it caches the
result across schedule).

Anton

int cpu;

static void schedule(void);

static inline int get_cpu(void) __attribute__((pure));
static inline int get_cpu(void)
{
	int ret;
	__asm__ ("mfspr %0, 0x113"
			: "=r" (ret)
			:);

	return ret;
}

int main(void)
{
	int cpu1, cpu2, cpu3, cpu4;
	cpu1 = get_cpu();
	cpu2 = get_cpu();
	schedule();
	cpu3 = get_cpu();
	cpu4 = get_cpu();
	printf("the cpu values were %d %d %d %d.\n",
		cpu1, cpu2, cpu3, cpu4);
	return 0;
}

static void schedule(void)
{
	cpu = 2;
	printf("schedule called .\n");
}

Re: [PATCH] take 2 of the tr-based current

[Manfred Spraul]

[-- Attachment #1: Type: text/plain, Size: 1057 bytes --]

Anton Blanchard wrote:

 >I changed the code a bit so that get_cpu() is now inline - this
 >represents our situation better. I think it is valid for gcc to cache
 >get_cpu across a function call in the below example because it knows
 >that get_cpu does not refer to any global variables.
 >
You are right: it seems that gcc internally decides if an inline
function is const or pure.

static inline int get_TR(void)
{
     int reg;
     __asm__ (
         "rep; nop;\n\t"
         "xor %0,%0\n\t"
         : "=r" (reg));
     return ret;
}

is always optimized as if get_TR is a function with __attribute__((const)).

Ben, is it intentional that get_TR is _not_ marked as inline? Your
version produces explicit function calls with -O2, and incorrect code
with -O99 (gcc decides to inline get_TR even without an inline
directive, and then optimizes away the calls after schedule.)

It seems that Anton's version generates the best code.
I've tested the attached version with egcs-1.1.2, gcc-2.96-98 and
gcc3-3.0.1-3, with -O0, -O2 and -O99.

--
     Manfred


[-- Attachment #2: test.c --]
[-- Type: application/msword, Size: 661 bytes --]

Re: [PATCH] take 2 of the tr-based current

[Benjamin LaHaise]

On Sun, Nov 11, 2001 at 03:02:28PM +0100, Manfred Spraul wrote:
> Ben, is it intentional that get_TR is _not_ marked as inline? Your
> version produces explicit function calls with -O2, and incorrect code
> with -O99 (gcc decides to inline get_TR even without an inline
> directive, and then optimizes away the calls after schedule.)

No, I'd taken the inline keyword out during testing.  It's now back in.

> It seems that Anton's version generates the best code.
> I've tested the attached version with egcs-1.1.2, gcc-2.96-98 and
> gcc3-3.0.1-3, with -O0, -O2 and -O99.

Done and tested:

#APP
        str %bx
#NO_APP
        call    schedule
        subl    $12, %esp
#APP
        str %ax
#NO_APP

with no duplication of the inline asm for cases where it is valid to 
optimize.  Updated patch is below.

		-ben
-- 
Fish.

... v2.4.13-ac8+tr.5.diff ...
diff -urN kernels/2.4/v2.4.13-ac8/arch/i386/kernel/entry.S v2.4.13-ac8+tr.5/arch/i386/kernel/entry.S
--- kernels/2.4/v2.4.13-ac8/arch/i386/kernel/entry.S	Tue Nov  6 20:43:22 2001
+++ v2.4.13-ac8+tr.5/arch/i386/kernel/entry.S	Thu Nov  8 22:27:30 2001
@@ -45,6 +45,7 @@
 #include <linux/linkage.h>
 #include <asm/segment.h>
 #include <asm/smp.h>
+#include <asm/current_asm.h>
 
 EBX		= 0x00
 ECX		= 0x04
@@ -134,9 +135,6 @@
 	.long 3b,6b;	\
 .previous
 
-#define GET_CURRENT(reg) \
-	movl %cr2, reg
-
 ENTRY(lcall7)
 	pushfl			# We get a different stack layout with call gates,
 	pushl %eax		# which has to be cleaned up later..
@@ -149,7 +147,7 @@
 	movl %ecx,CS(%esp)	#
 	movl %esp,%ebx
 	pushl %ebx
-	GET_CURRENT(%ebx)
+	GET_CURRENT(%ebx,%bx)
 	movl exec_domain(%ebx),%edx	# Get the execution domain
 	movl 4(%edx),%edx	# Get the lcall7 handler for the domain
 	pushl $0x7
@@ -170,7 +168,7 @@
 	movl %ecx,CS(%esp)	#
 	movl %esp,%ebx
 	pushl %ebx
-	GET_CURRENT(%ebx)
+	GET_CURRENT(%ebx,%bx)
 	movl exec_domain(%ebx),%edx	# Get the execution domain
 	movl 4(%edx),%edx	# Get the lcall7 handler for the domain
 	pushl $0x27
@@ -184,7 +182,7 @@
 	pushl %ebx
 	call SYMBOL_NAME(schedule_tail)
 	addl $4, %esp
-	GET_CURRENT(%ebx)
+	GET_CURRENT(%ebx,%bx)
 	testb $0x02,tsk_ptrace(%ebx)	# PT_TRACESYS
 	jne tracesys_exit
 	jmp	ret_from_sys_call
@@ -199,7 +197,7 @@
 ENTRY(system_call)
 	pushl %eax			# save orig_eax
 	SAVE_ALL
-	GET_CURRENT(%ebx)
+	GET_CURRENT(%ebx,%bx)
 	cmpl $(NR_syscalls),%eax
 	jae badsys
 	testb $0x02,tsk_ptrace(%ebx)	# PT_TRACESYS
@@ -251,7 +249,7 @@
 
 	ALIGN
 ENTRY(ret_from_intr)
-	GET_CURRENT(%ebx)
+	GET_CURRENT(%ebx,%bx)
 ret_from_exception:
 	movl EFLAGS(%esp),%eax		# mix EFLAGS and CS
 	movb CS(%esp),%al
@@ -297,7 +295,7 @@
 	movl %edx,%ds
 2:	call *%edi
 	addl $8,%esp
-	GET_CURRENT(%ebx)
+	GET_CURRENT(%ebx,%bx)
 	jmp ret_from_exception
 
 ENTRY(coprocessor_error)
@@ -313,7 +311,7 @@
 ENTRY(device_not_available)
 	pushl $-1		# mark this as an int
 	SAVE_ALL
-	GET_CURRENT(%ebx)
+	GET_CURRENT(%ebx,%bx)
 	movl %cr0,%eax
 	testl $0x4,%eax			# EM (math emulation bit)
 	jne device_not_available_emulate
diff -urN kernels/2.4/v2.4.13-ac8/arch/i386/kernel/ldt.c v2.4.13-ac8+tr.5/arch/i386/kernel/ldt.c
--- kernels/2.4/v2.4.13-ac8/arch/i386/kernel/ldt.c	Thu Nov  1 16:39:57 2001
+++ v2.4.13-ac8+tr.5/arch/i386/kernel/ldt.c	Thu Nov  8 18:25:56 2001
@@ -12,11 +12,13 @@
 #include <linux/smp.h>
 #include <linux/smp_lock.h>
 #include <linux/vmalloc.h>
+#include <linux/per_cpu.h>
 
 #include <asm/uaccess.h>
 #include <asm/system.h>
 #include <asm/ldt.h>
 #include <asm/desc.h>
+#include <asm/descfn.h>
 
 /*
  * read_ldt() is not really atomic - this is not a problem since
diff -urN kernels/2.4/v2.4.13-ac8/arch/i386/kernel/nmi.c v2.4.13-ac8+tr.5/arch/i386/kernel/nmi.c
--- kernels/2.4/v2.4.13-ac8/arch/i386/kernel/nmi.c	Tue Nov  6 20:43:22 2001
+++ v2.4.13-ac8+tr.5/arch/i386/kernel/nmi.c	Sat Nov 10 14:00:33 2001
@@ -264,7 +264,7 @@
 	/*
 	 * NMI can interrupt page faults, use hard_get_current.
 	 */
-	int sum, cpu = hard_get_current()->processor;
+	int sum, cpu = hard_smp_processor_id();
 
 	sum = apic_timer_irqs[cpu];
 
diff -urN kernels/2.4/v2.4.13-ac8/arch/i386/kernel/setup.c v2.4.13-ac8+tr.5/arch/i386/kernel/setup.c
--- kernels/2.4/v2.4.13-ac8/arch/i386/kernel/setup.c	Tue Nov  6 20:43:22 2001
+++ v2.4.13-ac8+tr.5/arch/i386/kernel/setup.c	Sat Nov 10 16:56:23 2001
@@ -108,6 +108,7 @@
 #include <asm/cobalt.h>
 #include <asm/msr.h>
 #include <asm/desc.h>
+#include <asm/descfn.h>
 #include <asm/e820.h>
 #include <asm/dma.h>
 #include <asm/mpspec.h>
@@ -2852,7 +2853,8 @@
  */
 void __init cpu_init (void)
 {
-	int nr = smp_processor_id();
+	struct task_struct *cur = hard_get_current();
+	int nr = cur->processor;
 	struct tss_struct * t = &init_tss[nr];
 
 	if (test_and_set_bit(nr, &cpu_initialized)) {
@@ -2884,17 +2886,19 @@
 	 * set up and load the per-CPU TSS and LDT
 	 */
 	atomic_inc(&init_mm.mm_count);
-	current->active_mm = &init_mm;
-	if(current->mm)
+	cur->active_mm = &init_mm;
+	if(cur->mm)
 		BUG();
-	enter_lazy_tlb(&init_mm, current, nr);
+	enter_lazy_tlb(&init_mm, cur, nr);
 
-	t->esp0 = current->thread.esp0;
+	t->esp0 = cur->thread.esp0;
 	set_tss_desc(nr,t);
 	gdt_table[__TSS(nr)].b &= 0xfffffdff;
 	load_TR(nr);
 	load_LDT(&init_mm);
 
+	set_current(cur);
+
 	/*
 	 * Clear all 6 debug registers:
 	 */
@@ -2908,8 +2912,8 @@
 	/*
 	 * Force FPU initialization:
 	 */
-	current->flags &= ~PF_USEDFPU;
-	current->used_math = 0;
+	cur->flags &= ~PF_USEDFPU;
+	cur->used_math = 0;
 	stts();
 }
 
diff -urN kernels/2.4/v2.4.13-ac8/arch/i386/kernel/smpboot.c v2.4.13-ac8+tr.5/arch/i386/kernel/smpboot.c
--- kernels/2.4/v2.4.13-ac8/arch/i386/kernel/smpboot.c	Tue Nov  6 20:43:22 2001
+++ v2.4.13-ac8+tr.5/arch/i386/kernel/smpboot.c	Sat Nov 10 15:56:44 2001
@@ -507,14 +507,14 @@
 }
 
 /* which physical APIC ID maps to which logical CPU number */
-volatile int physical_apicid_2_cpu[MAX_APICID];
+volatile int physical_apicid_to_cpu[MAX_APICID];
 /* which logical CPU number maps to which physical APIC ID */
-volatile int cpu_2_physical_apicid[NR_CPUS];
+volatile int cpu_to_physical_apicid[NR_CPUS];
 
 /* which logical APIC ID maps to which logical CPU number */
-volatile int logical_apicid_2_cpu[MAX_APICID];
+volatile int logical_apicid_to_cpu[MAX_APICID];
 /* which logical CPU number maps to which logical APIC ID */
-volatile int cpu_2_logical_apicid[NR_CPUS];
+volatile int cpu_to_logical_apicid[NR_CPUS];
 
 static inline void init_cpu_to_apicid(void)
 /* Initialize all maps between cpu number and apicids */
@@ -522,12 +522,12 @@
 	int apicid, cpu;
 
 	for (apicid = 0; apicid < MAX_APICID; apicid++) {
-		physical_apicid_2_cpu[apicid] = -1;
-		logical_apicid_2_cpu[apicid] = -1;
+		physical_apicid_to_cpu[apicid] = -1;
+		logical_apicid_to_cpu[apicid] = -1;
 	}
 	for (cpu = 0; cpu < NR_CPUS; cpu++) {
-		cpu_2_physical_apicid[cpu] = -1;
-		cpu_2_logical_apicid[cpu] = -1;
+		cpu_to_physical_apicid[cpu] = -1;
+		cpu_to_logical_apicid[cpu] = -1;
 	}
 }
 
@@ -538,11 +538,11 @@
  */
 {
 	if (clustered_apic_mode) {
-		logical_apicid_2_cpu[apicid] = cpu;	
-		cpu_2_logical_apicid[cpu] = apicid;
+		logical_apicid_to_cpu[apicid] = cpu;	
+		cpu_to_logical_apicid[cpu] = apicid;
 	} else {
-		physical_apicid_2_cpu[apicid] = cpu;	
-		cpu_2_physical_apicid[cpu] = apicid;
+		physical_apicid_to_cpu[apicid] = cpu;	
+		cpu_to_physical_apicid[cpu] = apicid;
 	}
 }
 
@@ -553,11 +553,11 @@
  */
 {
 	if (clustered_apic_mode) {
-		logical_apicid_2_cpu[apicid] = -1;	
-		cpu_2_logical_apicid[cpu] = -1;
+		logical_apicid_to_cpu[apicid] = -1;	
+		cpu_to_logical_apicid[cpu] = -1;
 	} else {
-		physical_apicid_2_cpu[apicid] = -1;	
-		cpu_2_physical_apicid[cpu] = -1;
+		physical_apicid_to_cpu[apicid] = -1;	
+		cpu_to_physical_apicid[cpu] = -1;
 	}
 }
 
diff -urN kernels/2.4/v2.4.13-ac8/arch/i386/lib/getuser.S v2.4.13-ac8+tr.5/arch/i386/lib/getuser.S
--- kernels/2.4/v2.4.13-ac8/arch/i386/lib/getuser.S	Tue Nov  6 20:43:22 2001
+++ v2.4.13-ac8+tr.5/arch/i386/lib/getuser.S	Wed Nov  7 22:33:07 2001
@@ -8,6 +8,7 @@
  * return an error value in addition to the "real"
  * return value.
  */
+#include <asm/current_asm.h>
 
 /*
  * __get_user_X
@@ -27,7 +28,6 @@
 .align 4
 .globl __get_user_1
 __get_user_1:
-	movl %cr2,%edx
 	cmpl addr_limit(%edx),%eax
 	jae bad_get_user
 1:	movzbl (%eax),%edx
@@ -39,7 +39,6 @@
 __get_user_2:
 	addl $1,%eax
 	jc bad_get_user
-	movl %cr2,%edx
 	cmpl addr_limit(%edx),%eax
 	jae bad_get_user
 2:	movzwl -1(%eax),%edx
@@ -51,7 +50,6 @@
 __get_user_4:
 	addl $3,%eax
 	jc bad_get_user
-	movl %cr2,%edx
 	cmpl addr_limit(%edx),%eax
 	jae bad_get_user
 3:	movl -3(%eax),%edx
diff -urN kernels/2.4/v2.4.13-ac8/arch/i386/mm/fault.c v2.4.13-ac8+tr.5/arch/i386/mm/fault.c
--- kernels/2.4/v2.4.13-ac8/arch/i386/mm/fault.c	Tue Nov  6 20:43:22 2001
+++ v2.4.13-ac8+tr.5/arch/i386/mm/fault.c	Sat Nov 10 14:03:13 2001
@@ -25,6 +25,7 @@
 #include <asm/uaccess.h>
 #include <asm/pgalloc.h>
 #include <asm/hardirq.h>
+#include <asm/desc.h>
 
 extern void die(const char *,struct pt_regs *,long);
 
@@ -148,7 +149,6 @@
 }
 
 asmlinkage void do_invalid_op(struct pt_regs *, unsigned long);
-extern unsigned long idt;
 
 /*
  * This routine handles page faults.  It determines the address,
@@ -173,9 +173,11 @@
 
 	/* get the address */
 	__asm__("movl %%cr2,%0":"=r" (address));
+
 	/* and restore current */
-	tsk = hard_get_current();
-	set_current(tsk);
+	set_current(hard_get_current());
+	tsk = current;
+
 
 	/* It's safe to allow irq's after cr2 has been saved */
 	if (regs->eflags & X86_EFLAGS_IF)
@@ -309,7 +311,7 @@
 	if (boot_cpu_data.f00f_bug) {
 		unsigned long nr;
 		
-		nr = (address - idt) >> 3;
+		nr = (address - (unsigned long)idt) >> 3;
 
 		if (nr == 6) {
 			do_invalid_op(regs, 0);
diff -urN kernels/2.4/v2.4.13-ac8/include/asm-i386/current.h v2.4.13-ac8+tr.5/include/asm-i386/current.h
--- kernels/2.4/v2.4.13-ac8/include/asm-i386/current.h	Tue Nov  6 20:43:27 2001
+++ v2.4.13-ac8+tr.5/include/asm-i386/current.h	Sun Nov 11 19:28:29 2001
@@ -1,28 +1,25 @@
 #ifndef _I386_CURRENT_H
 #define _I386_CURRENT_H
 
-struct task_struct;
+#include <linux/per_cpu.h>
 
-static inline struct task_struct * get_current(void)
+static inline struct task_struct *get_current(void) __attribute__((const));
+static inline struct task_struct *get_current(void)
 {
-	struct task_struct *tsk;
-	__asm__("movl %%cr2,%0;": "=r" (tsk));
-	return tsk;
+	return per_data(smp_processor_id())->curr;
 }
 
 /* for within NMI, do_page_fault, cpu_init */
 static inline struct task_struct * hard_get_current(void)
 {
 	struct task_struct **ptsk;
-	__asm__("andl %%esp,%0; ":"=r" (ptsk) : "0" (~8191UL));
+	__asm__ __volatile__("andl %%esp,%0; ":"=r" (ptsk) : "0" (~8191UL));
 	return *ptsk;
 }
 
 static inline void set_current(struct task_struct *tsk)
 {
-	__asm__("movl %0,%%cr2;"
-			: /* no output */
-			:"r" (tsk));
+	per_data(smp_processor_id())->curr = tsk;
 }
    
 /* Note: the implementation is hardcoded into arch/i386/lib/getuser.S */
diff -urN kernels/2.4/v2.4.13-ac8/include/asm-i386/current_asm.h v2.4.13-ac8+tr.5/include/asm-i386/current_asm.h
--- kernels/2.4/v2.4.13-ac8/include/asm-i386/current_asm.h	Wed Dec 31 19:00:00 1969
+++ v2.4.13-ac8+tr.5/include/asm-i386/current_asm.h	Sat Nov 10 12:48:47 2001
@@ -0,0 +1,30 @@
+/* asm/current_asm.h
+ */
+#ifndef __ASM__CURRENT_ASM_H
+#define __ASM__CURRENT_ASM_H
+
+#include <linux/config.h>
+#include <linux/per_cpu.h>
+#include <asm/desc.h>
+
+#if 1 /*def CONFIG_SMP*/
+/* Pass in the long and short versions of the register.
+ * eg GET_CURRENT(%ebx,%bx)
+ * All of this braindamage comes to us c/o a bug in gas: the
+ * opcode we want should actually be generated by strl, but 
+ * unfortunately gas doesn't realize that the operand size 
+ * prefix applies to str.  Please take a wet noodle and thread 
+ * it into my eye as that will be less painful than dealing 
+ * with this mess.  -ben
+ */
+#define GET_CURRENT(reg,regw)				\
+	str regw					\
+	; shll $LOG2_PER_CPU_SIZE-2,reg			\
+	; aligned_data_adjusted = aligned_data-(__FIRST_TSS_ENTRY << (3 + LOG2_PER_CPU_SIZE - 2))	\
+	; movl aligned_data_adjusted(reg),reg
+
+#else
+#define GET_CURRENT(reg,regw)	movl (aligned_data),reg
+#endif
+
+#endif /* __ASM__CURRENT_ASM_H */
diff -urN kernels/2.4/v2.4.13-ac8/include/asm-i386/desc.h v2.4.13-ac8+tr.5/include/asm-i386/desc.h
--- kernels/2.4/v2.4.13-ac8/include/asm-i386/desc.h	Tue Nov  6 20:43:27 2001
+++ v2.4.13-ac8+tr.5/include/asm-i386/desc.h	Tue Nov  6 21:21:32 2001
@@ -68,40 +68,6 @@
 
 #define __load_LDT(n) __asm__ __volatile__("lldt %%ax"::"a" (__LDT(n)<<3))
 
-/*
- * This is the ldt that every process will get unless we need
- * something other than this.
- */
-extern struct desc_struct default_ldt[];
-extern void set_intr_gate(unsigned int irq, void * addr);
-extern void set_ldt_desc(unsigned int n, void *addr, unsigned int size);
-extern void set_tss_desc(unsigned int n, void *addr);
-
-static inline void clear_LDT(void)
-{
-	int cpu = smp_processor_id();
-	set_ldt_desc(cpu, &default_ldt[0], 5);
-	__load_LDT(cpu);
-}
-
-/*
- * load one particular LDT into the current CPU
- */
-static inline void load_LDT (struct mm_struct *mm)
-{
-	int cpu = smp_processor_id();
-	void *segments = mm->context.segments;
-	int count = LDT_ENTRIES;
-
-	if (!segments) {
-		segments = &default_ldt[0];
-		count = 5;
-	}
-		
-	set_ldt_desc(cpu, segments, count);
-	__load_LDT(cpu);
-}
-
 #endif /* !__ASSEMBLY__ */
 
 #endif
diff -urN kernels/2.4/v2.4.13-ac8/include/asm-i386/descfn.h v2.4.13-ac8+tr.5/include/asm-i386/descfn.h
--- kernels/2.4/v2.4.13-ac8/include/asm-i386/descfn.h	Wed Dec 31 19:00:00 1969
+++ v2.4.13-ac8+tr.5/include/asm-i386/descfn.h	Tue Nov  6 21:23:59 2001
@@ -0,0 +1,42 @@
+#ifndef __ARCH_DESCFN_H
+#define __ARCH_DESCFN_H
+
+#ifndef __ARCH_DESC_H
+#include <asm/desc.h>
+#endif
+
+/*
+ * This is the ldt that every process will get unless we need
+ * something other than this.
+ */
+extern struct desc_struct default_ldt[];
+extern void set_intr_gate(unsigned int irq, void * addr);
+extern void set_ldt_desc(unsigned int n, void *addr, unsigned int size);
+extern void set_tss_desc(unsigned int n, void *addr);
+
+static inline void clear_LDT(void)
+{
+	int cpu = smp_processor_id();
+	set_ldt_desc(cpu, &default_ldt[0], 5);
+	__load_LDT(cpu);
+}
+
+/*
+ * load one particular LDT into the current CPU
+ */
+static inline void load_LDT (struct mm_struct *mm)
+{
+	int cpu = smp_processor_id();
+	void *segments = mm->context.segments;
+	int count = LDT_ENTRIES;
+
+	if (!segments) {
+		segments = &default_ldt[0];
+		count = 5;
+	}
+		
+	set_ldt_desc(cpu, segments, count);
+	__load_LDT(cpu);
+}
+
+#endif /* __ARCH_DESCFN_H */
diff -urN kernels/2.4/v2.4.13-ac8/include/asm-i386/mmu_context.h v2.4.13-ac8+tr.5/include/asm-i386/mmu_context.h
--- kernels/2.4/v2.4.13-ac8/include/asm-i386/mmu_context.h	Tue Nov  6 21:23:24 2001
+++ v2.4.13-ac8+tr.5/include/asm-i386/mmu_context.h	Sun Nov 11 19:37:12 2001
@@ -5,6 +5,7 @@
 #include <asm/desc.h>
 #include <asm/atomic.h>
 #include <asm/pgalloc.h>
+#include <asm/descfn.h>
 
 /*
  * possibly do the LDT unload here?
diff -urN kernels/2.4/v2.4.13-ac8/include/asm-i386/smp.h v2.4.13-ac8+tr.5/include/asm-i386/smp.h
--- kernels/2.4/v2.4.13-ac8/include/asm-i386/smp.h	Tue Nov  6 21:00:35 2001
+++ v2.4.13-ac8+tr.5/include/asm-i386/smp.h	Sun Nov 11 19:37:06 2001
@@ -8,6 +8,7 @@
 #include <linux/config.h>
 #include <linux/threads.h>
 #include <linux/ptrace.h>
+#include <asm/desc.h>
 #endif
 
 #ifdef CONFIG_X86_LOCAL_APIC
@@ -83,10 +84,10 @@
  * the real APIC ID <-> CPU # mapping.
  */
 #define MAX_APICID 256
-extern volatile int cpu_to_physical_apicid[NR_CPUS];
-extern volatile int physical_apicid_to_cpu[MAX_APICID];
-extern volatile int cpu_to_logical_apicid[NR_CPUS];
-extern volatile int logical_apicid_to_cpu[MAX_APICID];
+extern volatile int physical_apicid_to_cpu[];
+extern volatile int cpu_to_physical_apicid[];
+extern volatile int cpu_to_logical_apicid[];
+extern volatile int logical_apicid_to_cpu[];
 
 /*
  * General functions that each host system must provide.
@@ -101,7 +102,24 @@
  * so this is correct in the x86 case.
  */
 
-#define smp_processor_id() (current->processor)
+static inline unsigned get_TR(void) __attribute__ ((pure));
+static inline unsigned get_TR(void)
+{
+	extern int dummy_cpu_id;
+	unsigned tr;
+	/* The PAIN!  The HORROR!
+	 * Technically this is wrong, wrong, wrong, but 
+	 * gas doesn't know about strl.  *sigh*  Please 
+	 * flog them with a wet noodle repeatedly.
+	 * The extra parameter is a dummy value to prevent
+	 * gcc from assuming that the value is const across
+	 * function calls.  Fun!  -ben
+	 */
+	__asm__ ("str %w0" : "=r" (tr) : "m" (dummy_cpu_id));
+	return tr;
+}
+
+#define smp_processor_id()	( ((get_TR() >> 3) - __FIRST_TSS_ENTRY) >> 2 )
 
 static __inline int hard_smp_processor_id(void)
 {
diff -urN kernels/2.4/v2.4.13-ac8/include/asm-i386/smpboot.h v2.4.13-ac8+tr.5/include/asm-i386/smpboot.h
--- kernels/2.4/v2.4.13-ac8/include/asm-i386/smpboot.h	Fri Nov  9 23:55:07 2001
+++ v2.4.13-ac8+tr.5/include/asm-i386/smpboot.h	Sat Nov 10 15:57:58 2001
@@ -36,21 +36,21 @@
  * Mappings between logical cpu number and logical / physical apicid
  * The first four macros are trivial, but it keeps the abstraction consistent
  */
-extern volatile int logical_apicid_2_cpu[];
-extern volatile int cpu_2_logical_apicid[];
-extern volatile int physical_apicid_2_cpu[];
-extern volatile int cpu_2_physical_apicid[];
+extern volatile int logical_apicid_to_cpu[];
+extern volatile int cpu_to_logical_apicid[];
+extern volatile int physical_apicid_to_cpu[];
+extern volatile int cpu_to_physical_apicid[];
 
-#define logical_apicid_to_cpu(apicid) logical_apicid_2_cpu[apicid]
-#define cpu_to_logical_apicid(cpu) cpu_2_logical_apicid[cpu]
-#define physical_apicid_to_cpu(apicid) physical_apicid_2_cpu[apicid]
-#define cpu_to_physical_apicid(cpu) cpu_2_physical_apicid[cpu]
+#define logical_apicid_to_cpu(apicid) logical_apicid_to_cpu[apicid]
+#define cpu_to_logical_apicid(cpu) cpu_to_logical_apicid[cpu]
+#define physical_apicid_to_cpu(apicid) physical_apicid_to_cpu[apicid]
+#define cpu_to_physical_apicid(cpu) cpu_to_physical_apicid[cpu]
 #ifdef CONFIG_MULTIQUAD			/* use logical IDs to bootstrap */
-#define boot_apicid_to_cpu(apicid) logical_apicid_2_cpu[apicid]
-#define cpu_to_boot_apicid(cpu) cpu_2_logical_apicid[cpu]
+#define boot_apicid_to_cpu(apicid) logical_apicid_to_cpu[apicid]
+#define cpu_to_boot_apicid(cpu) cpu_to_logical_apicid[cpu]
 #else /* !CONFIG_MULTIQUAD */		/* use physical IDs to bootstrap */
-#define boot_apicid_to_cpu(apicid) physical_apicid_2_cpu[apicid]
-#define cpu_to_boot_apicid(cpu) cpu_2_physical_apicid[cpu]
+#define boot_apicid_to_cpu(apicid) physical_apicid_to_cpu[apicid]
+#define cpu_to_boot_apicid(cpu) cpu_to_physical_apicid[cpu]
 #endif /* CONFIG_MULTIQUAD */
 
 
diff -urN kernels/2.4/v2.4.13-ac8/include/asm-i386/uaccess.h v2.4.13-ac8+tr.5/include/asm-i386/uaccess.h
--- kernels/2.4/v2.4.13-ac8/include/asm-i386/uaccess.h	Wed Nov  7 18:09:12 2001
+++ v2.4.13-ac8+tr.5/include/asm-i386/uaccess.h	Sun Nov 11 21:46:52 2001
@@ -109,7 +109,7 @@
 #define __get_user_x(size,ret,x,ptr) \
 	__asm__ __volatile__("call __get_user_" #size \
 		:"=a" (ret),"=d" (x) \
-		:"0" (ptr))
+		:"0" (ptr), "1" (current))
 
 /* Careful: we have to cast the result to the type of the pointer for sign reasons */
 #define get_user(x,ptr)							\
diff -urN kernels/2.4/v2.4.13-ac8/include/linux/per_cpu.h v2.4.13-ac8+tr.5/include/linux/per_cpu.h
--- kernels/2.4/v2.4.13-ac8/include/linux/per_cpu.h	Wed Dec 31 19:00:00 1969
+++ v2.4.13-ac8+tr.5/include/linux/per_cpu.h	Thu Nov  8 22:15:26 2001
@@ -0,0 +1,32 @@
+#ifndef __LINUX__PER_CPU__H
+#define __LINUX__PER_CPU__H
+
+#define LOG2_PER_CPU_SIZE	8
+#define PER_CPU_SIZE		(1 << LOG2_PER_CPU_SIZE)
+
+#ifndef __ASSEMBLY__
+struct task_struct;
+
+struct per_cpu_data {
+	/* Assembly code relies on curr being the first member of this 
+	 * structure.  Please change it if this gets rearranged.
+	 */
+	struct task_struct	*curr;
+	cycles_t		last_schedule;
+};
+
+union aligned_data {
+	struct per_cpu_data	data;
+	char __pad [PER_CPU_SIZE];
+
+	/* Make sure the padding is large enough by forcing an error 
+	 * if it isn't.  -ben
+	 */
+	char __pad2 [PER_CPU_SIZE - sizeof(struct per_cpu_data)];
+};
+
+extern union aligned_data aligned_data[];
+
+#define per_data(nr)	(&aligned_data[nr].data)
+#endif
+#endif
diff -urN kernels/2.4/v2.4.13-ac8/init/main.c v2.4.13-ac8+tr.5/init/main.c
--- kernels/2.4/v2.4.13-ac8/init/main.c	Tue Nov  6 20:43:28 2001
+++ v2.4.13-ac8+tr.5/init/main.c	Sat Nov 10 13:11:06 2001
@@ -635,7 +635,6 @@
  * Interrupts are still disabled. Do necessary setups, then
  * enable them
  */
-	lock_kernel();
 	printk(linux_banner);
 	setup_arch(&command_line);
 	printk("Kernel command line: %s\n", saved_command_line);
@@ -646,6 +645,13 @@
 	softirq_init();
 	time_init();
 
+	/* At the very least, this has to come after trap_init as x86
+	 * needs to perform CPU setup before current is valid.  This 
+	 * should be okay as we're still running with interrupts disabled 
+	 * and no other CPUs are up yet.  -ben
+	 */
+	lock_kernel();
+
 	/*
 	 * HACK ALERT! This is early. We're enabling the console before
 	 * we've done PCI setups etc, and console_init() must be aware of
diff -urN kernels/2.4/v2.4.13-ac8/kernel/ksyms.c v2.4.13-ac8+tr.5/kernel/ksyms.c
--- kernels/2.4/v2.4.13-ac8/kernel/ksyms.c	Tue Nov  6 20:43:28 2001
+++ v2.4.13-ac8+tr.5/kernel/ksyms.c	Sat Nov 10 16:57:06 2001
@@ -447,6 +447,7 @@
 #endif
 EXPORT_SYMBOL(kstat);
 EXPORT_SYMBOL(nr_running);
+EXPORT_SYMBOL(aligned_data);
 
 /* misc */
 EXPORT_SYMBOL(panic);
diff -urN kernels/2.4/v2.4.13-ac8/kernel/sched.c v2.4.13-ac8+tr.5/kernel/sched.c
--- kernels/2.4/v2.4.13-ac8/kernel/sched.c	Tue Nov  6 20:43:28 2001
+++ v2.4.13-ac8+tr.5/kernel/sched.c	Sat Nov 10 16:29:58 2001
@@ -28,6 +28,7 @@
 #include <linux/kernel_stat.h>
 #include <linux/completion.h>
 #include <linux/prefetch.h>
+#include <linux/per_cpu.h>
 
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
@@ -97,16 +98,10 @@
  * We align per-CPU scheduling data on cacheline boundaries,
  * to prevent cacheline ping-pong.
  */
-static union {
-	struct schedule_data {
-		struct task_struct * curr;
-		cycles_t last_schedule;
-	} schedule_data;
-	char __pad [SMP_CACHE_BYTES];
-} aligned_data [NR_CPUS] __cacheline_aligned = { {{&init_task,0}}};
+union aligned_data aligned_data[NR_CPUS] __cacheline_aligned;
 
-#define cpu_curr(cpu) aligned_data[(cpu)].schedule_data.curr
-#define last_schedule(cpu) aligned_data[(cpu)].schedule_data.last_schedule
+#define cpu_curr(cpu)		per_data(cpu)->curr
+#define last_schedule(cpu)	per_data(cpu)->last_schedule
 
 struct kernel_stat kstat;
 extern struct task_struct *child_reaper;
@@ -532,7 +527,7 @@
  */
 asmlinkage void schedule(void)
 {
-	struct schedule_data * sched_data;
+	struct per_cpu_data * sched_data;
 	struct task_struct *prev, *next, *p;
 	struct list_head *tmp;
 	int this_cpu, c;
@@ -543,7 +538,7 @@
 	if (!current->active_mm) BUG();
 need_resched_back:
 	prev = current;
-	this_cpu = prev->processor;
+	this_cpu = smp_processor_id();	/* This better than current->processor on up */
 
 	if (in_interrupt())
 		goto scheduling_in_interrupt;
@@ -554,7 +549,7 @@
 	 * 'sched_data' is protected by the fact that we can run
 	 * only one process per CPU.
 	 */
-	sched_data = & aligned_data[this_cpu].schedule_data;
+	sched_data = per_data(this_cpu);
 
 	spin_lock_irq(&runqueue_lock);
 
@@ -1057,7 +1052,7 @@
 	// Subtract non-idle processes running on other CPUs.
 	for (i = 0; i < smp_num_cpus; i++) {
 		int cpu = cpu_logical_map(i);
-		if (aligned_data[cpu].schedule_data.curr != idle_task(cpu))
+		if (per_data(cpu)->curr != idle_task(cpu))
 			nr_pending--;
 	}
 #else
@@ -1309,17 +1304,18 @@
 
 void __init init_idle(void)
 {
-	struct schedule_data * sched_data;
-	sched_data = &aligned_data[smp_processor_id()].schedule_data;
+	struct per_cpu_data * sched_data;
+	int cpu = smp_processor_id();
+	sched_data = per_data(cpu);
 
 	if (current != &init_task && task_on_runqueue(current)) {
 		printk("UGH! (%d:%d) was on the runqueue, removing.\n",
-			smp_processor_id(), current->pid);
+			cpu, current->pid);
 		del_from_runqueue(current);
 	}
 	sched_data->curr = current;
 	sched_data->last_schedule = get_cycles();
-	clear_bit(current->processor, &wait_init_idle);
+	clear_bit(cpu, &wait_init_idle);
 }
 
 extern void init_timervecs (void);
@@ -1334,6 +1330,7 @@
 	int nr;
 
 	init_task.processor = cpu;
+	set_current(&init_task);
 
 	for(nr = 0; nr < PIDHASH_SZ; nr++)
 		pidhash[nr] = NULL;