Skip to contentSkip to conversation

powerpc: consolidate feature fixup code

PATCH In review Benjamin Herrenschmidt 3 participants 7 messages Oct 12, 2006 PowerPC

lore
Share & tools
lore
#0 lore raw

PATCH 1/4 powerpc: consolidate feature fixup code

There are currently two versions of the functions for applying the feature fixups, one for CPU features and one for firmware features. In addition, they are both in assembly.

This patch replaces them with a single C function. The call site is slightly moved on ppc64 as well to be called from C instead of from assembly, though it's a very small move, and thus shouldn't cause any problem (called at the start of setup_system() instead of just before calling it).

arch/powerpc/kernel/head_64.S +0 -7
Index: linux-cell/arch/powerpc/kernel/head_64.S
===================================================================
@@ -2000,13 +2000,6 @@ _STATIC(start_here_common)
 	li	r0,0
 	stdu	r0,-STACK_FRAME_OVERHEAD(r1)
 
-	/* Apply the CPUs-specific fixups (nop out sections not relevant
-	 * to this CPU
-	 */
-	li	r3,0
-	bl	.do_cpu_ftr_fixups
-	bl	.do_fw_ftr_fixups
-
 	/* ptr to current */
 	LOAD_REG_IMMEDIATE(r4, init_task)
 	std	r4,PACACURRENT(r13)
arch/powerpc/kernel/misc_64.S +0 -93

99 lines not on this page · raw message

arch/powerpc/kernel/setup-common.c +29 -0
Index: linux-cell/arch/powerpc/kernel/setup-common.c
===================================================================
@@ -520,3 +520,32 @@ void __init setup_panic(void)
 {
 	atomic_notifier_chain_register(&panic_notifier_list, &ppc_panic_block);
 }
+
+void do_feature_fixups(unsigned long offset, unsigned long value,
+		       void *fixup_start, void *fixup_end)
+{
+	struct fixup_entry {
+		unsigned long	mask;
+		unsigned long	value;
+		unsigned int	*start;
+		unsigned int	*end;
+	} *fcur, *fend;
+
+	fcur = fixup_start;
+	fend = fixup_end;
+
+	for (; fcur < fend; fcur++) {
+		unsigned int *pcur, *pend;
+
+		if ((value & fcur->mask) == fcur->value)
+			continue;
+
+		pcur = fcur->start - offset;
+		pend = fcur->end - offset;
+		for (; pcur < pend; pcur++) {
+			*pcur = 0x60000000u;
+			asm ("dcbst 0, %0; sync; icbi 0,%0; sync; isync"
+			     : : "r" (pcur));
+		}
+	}
+}
arch/powerpc/kernel/setup.h +6 -3
Index: linux-cell/arch/powerpc/kernel/setup.h
===================================================================
@@ -1,9 +1,12 @@
 #ifndef _POWERPC_KERNEL_SETUP_H
 #define _POWERPC_KERNEL_SETUP_H
 
-void check_for_initrd(void);
-void do_init_bootmem(void);
-void setup_panic(void);
+extern void check_for_initrd(void);
+extern void do_init_bootmem(void);
+extern void setup_panic(void);
+extern void do_feature_fixups(unsigned long offset, unsigned long value,
+			      void *fixup_start, void *fixup_end);
+
 extern int do_early_xmon;
 
 #endif /* _POWERPC_KERNEL_SETUP_H */
arch/powerpc/kernel/setup_32.c +4 -1
Index: linux-cell/arch/powerpc/kernel/setup_32.c
===================================================================
@@ -90,6 +90,7 @@ int ucache_bsize;
  */
 unsigned long __init early_init(unsigned long dt_ptr)
 {
+	extern unsigned int __start___ftr_fixup, __stop___ftr_fixup;
 	unsigned long offset = reloc_offset();
 
 	/* First zero the BSS -- use memset_io, some platforms don't have
@@ -101,7 +102,9 @@ unsigned long __init early_init(unsigned
 	 * that depend on which cpu we have.
 	 */
 	identify_cpu(offset, 0);
-	do_cpu_ftr_fixups(offset);
+
+	do_feature_fixups(offset, cur_cpu_spec->cpu_features,
+			  &__start___ftr_fixup, &__stop___ftr_fixup);
 
 	return KERNELBASE + offset;
 }
arch/powerpc/kernel/setup_64.c +11 -0
Index: linux-cell/arch/powerpc/kernel/setup_64.c
===================================================================
@@ -346,8 +346,19 @@ static void __init initialize_cache_info
  */
 void __init setup_system(void)
 {
+	extern unsigned int __start___ftr_fixup, __stop___ftr_fixup;
+	extern unsigned int __start___fw_ftr_fixup, __stop___fw_ftr_fixup;
+
 	DBG(" -> setup_system()\n");
 
+	/* Apply the CPUs-specific and firmware specific fixups to kernel
+	 * text (nop out sections not relevant to this CPU or this firmware)
+	 */
+	do_feature_fixups(0, cur_cpu_spec->cpu_features,
+			  &__start___ftr_fixup, &__stop___ftr_fixup);
+	do_feature_fixups(0, powerpc_firmware_features,
+			  &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
+
 	/*
 	 * Unflatten the device-tree passed by prom_init or kexec
 	 */
#1 lore raw
Quoted 9 lines · Benjamin Herrenschmidt

There are currently two versions of the functions for applying the feature fixups, one for CPU features and one for firmware features. In addition, they are both in assembly.

This patch replaces them with a single C function. The call site is slightly moved on ppc64 as well to be called from C instead of from assembly, though it's a very small move, and thus shouldn't cause any problem (called at the start of setup_system() instead of just before calling it).

Nice to see it written in C :)

Quoted 35 lines

Index: linux-cell/arch/powerpc/kernel/setup.h
===================================================================

arch/powerpc/kernel/setup.h +6 -3
@@ -1,9 +1,12 @@
 #ifndef _POWERPC_KERNEL_SETUP_H
 #define _POWERPC_KERNEL_SETUP_H
 
-void check_for_initrd(void);
-void do_init_bootmem(void);
-void setup_panic(void);
+extern void check_for_initrd(void);
+extern void do_init_bootmem(void);
+extern void setup_panic(void);
+extern void do_feature_fixups(unsigned long offset, unsigned long value,
+			      void *fixup_start, void *fixup_end);
+
 extern int do_early_xmon;
 
 #endif /* _POWERPC_KERNEL_SETUP_H */

Index: linux-cell/arch/powerpc/kernel/setup_32.c
===================================================================

arch/powerpc/kernel/setup_32.c +1 -0
@@ -90,6 +90,7 @@ int ucache_bsize;
  */
 unsigned long __init early_init(unsigned long dt_ptr)
 {
+	extern unsigned int __start___ftr_fixup, __stop___ftr_fixup;
 

Index: linux-cell/arch/powerpc/kernel/setup_64.c
===================================================================

arch/powerpc/kernel/setup_64.c static void __init initialize_cache_info · line 350

Quoted 5 lines
arch/powerpc/kernel/setup_64.c +2 -0
@@ -346,8 +346,19 @@ static void __init initialize_cache_info
  */
 void __init setup_system(void)
 {
+	extern unsigned int __start___ftr_fixup, __stop___ftr_fixup;
+	extern unsigned int __start___fw_ftr_fixup, __stop___fw_ftr_fixup;

Every extern in a C file ... god kills a kitten! :)

cheers

Signature · 8 lines
Michael Ellerman
OzLabs, IBM Australia Development Lab

wwweb: http://michael.ellerman.id.au
phone: +61 2 6212 1183 (tie line 70 21183)

We do not inherit the earth from our ancestors,
we borrow it from our children. - S.M.A.R.T Person
#2 lore raw
Quoted 8 lines

arch/powerpc/kernel/setup_64.c static void __init initialize_cache_info · line 350

Quoted 5 lines

void __init setup_system(void)
{

arch/powerpc/kernel/setup_64.c +2 -0
+	extern unsigned int __start___ftr_fixup, __stop___ftr_fixup;
+	extern unsigned int __start___fw_ftr_fixup, __stop___fw_ftr_fixup;

Every extern in a C file ... god kills a kitten! :)

Well... it's a bit special in this case.. just a reference to a section, I'd rather keep it close to the only call site...

Ben.

#3 lore raw
In reply to Benjamin Herrenschmidt
Quoted 143 lines · On Thu, 12 Oct 2006 18:29:53 +1000 Benjamin Herrenschmidt…

There are currently two versions of the functions for applying the feature fixups, one for CPU features and one for firmware features. In addition, they are both in assembly.

This patch replaces them with a single C function. The call site is slightly moved on ppc64 as well to be called from C instead of from assembly, though it's a very small move, and thus shouldn't cause any problem (called at the start of setup_system() instead of just before calling it).

Index: linux-cell/arch/powerpc/kernel/head_64.S
===================================================================

arch/powerpc/kernel/head_64.S +0 -7
@@ -2000,13 +2000,6 @@ _STATIC(start_here_common)
 	li	r0,0
 	stdu	r0,-STACK_FRAME_OVERHEAD(r1)
 
-	/* Apply the CPUs-specific fixups (nop out sections not relevant
-	 * to this CPU
-	 */
-	li	r3,0
-	bl	.do_cpu_ftr_fixups
-	bl	.do_fw_ftr_fixups
-
 	/* ptr to current */
 	LOAD_REG_IMMEDIATE(r4, init_task)
 	std	r4,PACACURRENT(r13)

Index: linux-cell/arch/powerpc/kernel/misc_64.S
===================================================================

arch/powerpc/kernel/misc_64.S +0 -93
@@ -277,99 +277,6 @@ _GLOBAL(identify_cpu)
 	mr	r3,r5
 	bctr
 
-/*
- * do_cpu_ftr_fixups - goes through the list of CPU feature fixups
- * and writes nop's over sections of code that don't apply for this cpu.
- * r3 = data offset (not changed)
- */
-_GLOBAL(do_cpu_ftr_fixups)
-	/* Get CPU 0 features */
-	LOAD_REG_IMMEDIATE(r6,cur_cpu_spec)
-	sub	r6,r6,r3
-	ld	r4,0(r6)
-	sub	r4,r4,r3
-	ld	r4,CPU_SPEC_FEATURES(r4)
-	/* Get the fixup table */
-	LOAD_REG_IMMEDIATE(r6,__start___ftr_fixup)
-	sub	r6,r6,r3
-	LOAD_REG_IMMEDIATE(r7,__stop___ftr_fixup)
-	sub	r7,r7,r3
-	/* Do the fixup */
-1:	cmpld	r6,r7
-	bgelr
-	addi	r6,r6,32
-	ld	r8,-32(r6)	/* mask */
-	and	r8,r8,r4
-	ld	r9,-24(r6)	/* value */
-	cmpld	r8,r9
-	beq	1b
-	ld	r8,-16(r6)	/* section begin */
-	ld	r9,-8(r6)	/* section end */
-	subf.	r9,r8,r9
-	beq	1b
-	/* write nops over the section of code */
-	/* todo: if large section, add a branch at the start of it */
-	srwi	r9,r9,2
-	mtctr	r9
-	sub	r8,r8,r3
-	lis	r0,0x60000000@h	/* nop */
-3:	stw	r0,0(r8)
-	andi.	r10,r4,CPU_FTR_SPLIT_ID_CACHE@l
-	beq	2f
-	dcbst	0,r8		/* suboptimal, but simpler */
-	sync
-	icbi	0,r8
-2:	addi	r8,r8,4
-	bdnz	3b
-	sync			/* additional sync needed on g4 */
-	isync
-	b	1b
-
-/*
- * do_fw_ftr_fixups - goes through the list of firmware feature fixups
- * and writes nop's over sections of code that don't apply for this firmware.
- * r3 = data offset (not changed)
- */
-_GLOBAL(do_fw_ftr_fixups)
-	/* Get firmware features */
-	LOAD_REG_IMMEDIATE(r6,powerpc_firmware_features)
-	sub	r6,r6,r3
-	ld	r4,0(r6)
-	/* Get the fixup table */
-	LOAD_REG_IMMEDIATE(r6,__start___fw_ftr_fixup)
-	sub	r6,r6,r3
-	LOAD_REG_IMMEDIATE(r7,__stop___fw_ftr_fixup)
-	sub	r7,r7,r3
-	/* Do the fixup */
-1:	cmpld	r6,r7
-	bgelr
-	addi	r6,r6,32
-	ld	r8,-32(r6)	/* mask */
-	and	r8,r8,r4
-	ld	r9,-24(r6)	/* value */
-	cmpld	r8,r9
-	beq	1b
-	ld	r8,-16(r6)	/* section begin */
-	ld	r9,-8(r6)	/* section end */
-	subf.	r9,r8,r9
-	beq	1b
-	/* write nops over the section of code */
-	/* todo: if large section, add a branch at the start of it */
-	srwi	r9,r9,2
-	mtctr	r9
-	sub	r8,r8,r3
-	lis	r0,0x60000000@h	/* nop */
-3:	stw	r0,0(r8)
-BEGIN_FTR_SECTION
-	dcbst	0,r8		/* suboptimal, but simpler */
-	sync
-	icbi	0,r8
-END_FTR_SECTION_IFSET(CPU_FTR_SPLIT_ID_CACHE)
-	addi	r8,r8,4
-	bdnz	3b
-	sync			/* additional sync needed on g4 */
-	isync
-	b	1b
 
 #if defined(CONFIG_PPC_PMAC) || defined(CONFIG_PPC_MAPLE)
 /*

Index: linux-cell/arch/powerpc/kernel/setup-common.c
===================================================================

arch/powerpc/kernel/setup-common.c +0 -0
@@ -520,3 +520,32 @@ void __init setup_panic(void)
 {
 	atomic_notifier_chain_register(&panic_notifier_list, &ppc_panic_block);
 }

arch/powerpc/kernel/setup-common.c void __init setup_panic(void) · line 532

Quoted 10 lines
arch/powerpc/kernel/setup-common.c +10 -0
+
+void do_feature_fixups(unsigned long offset, unsigned long value,
+		       void *fixup_start, void *fixup_end)
+{
+	struct fixup_entry {
+		unsigned long	mask;
+		unsigned long	value;
+		unsigned int	*start;
+		unsigned int	*end;
+	} *fcur, *fend;

Shouldn't there be a better place to keep this struct definition than in the function it's used? Some header file?

arch/powerpc/kernel/setup-common.c void __init setup_panic(void) · line 537

Quoted 5 lines
arch/powerpc/kernel/setup-common.c +5 -0
+
+	fcur = fixup_start;
+	fend = fixup_end;
+
+	for (; fcur < fend; fcur++) {
	for (fcur = fixup_start; fcur < fend; fcur++) {

arch/powerpc/kernel/setup-common.c void __init setup_panic(void) · line 549

Quoted 12 lines
arch/powerpc/kernel/setup-common.c +12 -0
+		unsigned int *pcur, *pend;
+
+		if ((value & fcur->mask) == fcur->value)
+			continue;
+
+		pcur = fcur->start - offset;
+		pend = fcur->end - offset;
+		for (; pcur < pend; pcur++) {
+			*pcur = 0x60000000u;
+			asm ("dcbst 0, %0; sync; icbi 0,%0; sync; isync"
+			     : : "r" (pcur));
+		}

Ok, now it's in C, no reason to do a dcbst/icbi for every word. We did it in asm for simplicity's sake. Split it in two loops, one to nop, second to step per line and do the flushes. :-)

Quoted 75 lines
arch/powerpc/kernel/setup-common.c +2 -0
+	}
+}

Index: linux-cell/arch/powerpc/kernel/setup.h
===================================================================

arch/powerpc/kernel/setup.h +6 -3
@@ -1,9 +1,12 @@
 #ifndef _POWERPC_KERNEL_SETUP_H
 #define _POWERPC_KERNEL_SETUP_H
 
-void check_for_initrd(void);
-void do_init_bootmem(void);
-void setup_panic(void);
+extern void check_for_initrd(void);
+extern void do_init_bootmem(void);
+extern void setup_panic(void);
+extern void do_feature_fixups(unsigned long offset, unsigned long value,
+			      void *fixup_start, void *fixup_end);
+
 extern int do_early_xmon;
 
 #endif /* _POWERPC_KERNEL_SETUP_H */

Index: linux-cell/arch/powerpc/kernel/setup_32.c
===================================================================

arch/powerpc/kernel/setup_32.c +4 -1
@@ -90,6 +90,7 @@ int ucache_bsize;
  */
 unsigned long __init early_init(unsigned long dt_ptr)
 {
+	extern unsigned int __start___ftr_fixup, __stop___ftr_fixup;
 	unsigned long offset = reloc_offset();
 
 	/* First zero the BSS -- use memset_io, some platforms don't have
@@ -101,7 +102,9 @@ unsigned long __init early_init(unsigned
 	 * that depend on which cpu we have.
 	 */
 	identify_cpu(offset, 0);
-	do_cpu_ftr_fixups(offset);
+
+	do_feature_fixups(offset, cur_cpu_spec->cpu_features,
+			  &__start___ftr_fixup, &__stop___ftr_fixup);
 
 	return KERNELBASE + offset;
 }

Index: linux-cell/arch/powerpc/kernel/setup_64.c
===================================================================

arch/powerpc/kernel/setup_64.c +11 -0
@@ -346,8 +346,19 @@ static void __init initialize_cache_info
  */
 void __init setup_system(void)
 {
+	extern unsigned int __start___ftr_fixup, __stop___ftr_fixup;
+	extern unsigned int __start___fw_ftr_fixup, __stop___fw_ftr_fixup;
+
 	DBG(" -> setup_system()\n");
 
+	/* Apply the CPUs-specific and firmware specific fixups to kernel
+	 * text (nop out sections not relevant to this CPU or this firmware)
+	 */
+	do_feature_fixups(0, cur_cpu_spec->cpu_features,
+			  &__start___ftr_fixup, &__stop___ftr_fixup);
+	do_feature_fixups(0, powerpc_firmware_features,
+			  &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
+
 	/*
 	 * Unflatten the device-tree passed by prom_init or kexec
 	 */

_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://ozlabs.org/mailman/listinfo/linuxppc-dev

#4 lore raw
Quoted 13 lines

arch/powerpc/kernel/setup-common.c void __init setup_panic(void) · line 532

Quoted 10 lines
arch/powerpc/kernel/setup-common.c +9 -0
+void do_feature_fixups(unsigned long offset, unsigned long value,
+		       void *fixup_start, void *fixup_end)
+{
+	struct fixup_entry {
+		unsigned long	mask;
+		unsigned long	value;
+		unsigned int	*start;
+		unsigned int	*end;
+	} *fcur, *fend;

Shouldn't there be a better place to keep this struct definition than in the function it's used? Some header file?

It's not used anywhere else.... which header would you put it in ?

Quoted 8 lines

arch/powerpc/kernel/setup-common.c void __init setup_panic(void) · line 537

Quoted 5 lines
arch/powerpc/kernel/setup-common.c +5 -0
+
+	fcur = fixup_start;
+	fend = fixup_end;
+
+	for (; fcur < fend; fcur++) {
	for (fcur = fixup_start; fcur < fend; fcur++) {
Quoted 16 lines

arch/powerpc/kernel/setup-common.c void __init setup_panic(void) · line 549

Quoted 12 lines
arch/powerpc/kernel/setup-common.c +12 -0
+		unsigned int *pcur, *pend;
+
+		if ((value & fcur->mask) == fcur->value)
+			continue;
+
+		pcur = fcur->start - offset;
+		pend = fcur->end - offset;
+		for (; pcur < pend; pcur++) {
+			*pcur = 0x60000000u;
+			asm ("dcbst 0, %0; sync; icbi 0,%0; sync; isync"
+			     : : "r" (pcur));
+		}

Ok, now it's in C, no reason to do a dcbst/icbi for every word. We did it in asm for simplicity's sake. Split it in two loops, one to nop, second to step per line and do the flushes. :-)

Do we care ? :) But yah, I suppose I can do that.

Thanks,
Ben.

#5 lore raw
Quoted 14 lines · On Fri, 13 Oct 2006 00:47:29 +1000 Benjamin Herrenschmidt…
Quoted 13 lines

arch/powerpc/kernel/setup-common.c void __init setup_panic(void) · line 532

Quoted 10 lines
arch/powerpc/kernel/setup-common.c +9 -0
+void do_feature_fixups(unsigned long offset, unsigned long value,
+		       void *fixup_start, void *fixup_end)
+{
+	struct fixup_entry {
+		unsigned long	mask;
+		unsigned long	value;
+		unsigned int	*start;
+		unsigned int	*end;
+	} *fcur, *fend;

Shouldn't there be a better place to keep this struct definition than in the function it's used? Some header file?

Quoted 1 line

It's not used anywhere else.... which header would you put it in ?

Near where the asm-corresponding ones are located (cputable.h).

-Olof

Quoted 5 lines
Quoted 3 lines

Ok, now it's in C, no reason to do a dcbst/icbi for every word. We did it in asm for simplicity's sake. Split it in two loops, one to nop, second to step per line and do the flushes. :-)

Do we care ? :) But yah, I suppose I can do that.

Close call. :-) It's just easier to do now, might as well do it. But yeah, no big deal.

It will also be easier to graft in a branch forward instead of a large amount of nops for large feature sections. I guess it will take some benchmarking to see what the breakeven point is.

-Olof

#6 lore raw
Quoted 4 lines

It will also be easier to graft in a branch forward instead of a large amount of nops for large feature sections. I guess it will take some benchmarking to see what the breakeven point is.

Yeah, I've been thinking about that one for some time (in fact since I did the first implementation of feature fixup). Thing is, back when it was all asm, it was hard, and we never really had big feature sections. Now that we have all that stuff with with PURR etc... in the exception entry/return path, it might be worth having another look. I suspect that the breakeven point will be very CPU dependant tho.

Ben.