--UHN/qo2QbUvPLonB
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline

On Thu, Mar 22, 2007 at 06:32:14 +0300, Valeriy E. Ushakov wrote:

> I've finally had some time for sh3 hacking and decided to look into
> implementing fast TLB MISS handling.

I now have it working on sh3 and sh4.  It gives nice performance boost
under load.  E.g on my usl-5p running off nfs root compiling
pkgtools/digest with -pipe -Os -freorder-blocks (-O2 is too -falign-*
zealous in the generated code) compilation time goes down from

    real    11m10.377s
    user    8m53.265s
    sys     2m8.534s
to
    real    8m17.268s
    user    6m10.767s
    sys     1m58.874s

(I actually had to re-check, just to exlude the possibility that my
original measurements were done with some extra debugging options in
the kernel, but no, it's really that good :)

Another run using a kernel with fast handler and evcnt instrumentation
charges that test with

    fast user va miss    35175030
    fast kernel va miss   1378232
    slow user va miss      793411
    slow kernel va miss      7395

pkgsrc is make/shell happy and gcc4 is a memory hog so the test puts
quite a bit of pressure on 64 entry sh4 TLB - and nearly %98 of tlb
misses can be served via fast path.

Attached is a tiny patch to pmap.c and new exception_vector.S (patch
is just a tad smaller and harder to read).

I'm going to stare at it a bit more just in case, to make sure I'm not
breaking some assumptions inadvertedly and will commit in a few days.
If anyone is feeling like reviewing this with fresh eyes - it will be
appreciated.

SY, Uwe
-- 
uwe@ptc.spbu.ru                         |       Zu Grunde kommen
http://snark.ptc.spbu.ru/~uwe/          |       Ist zu Grunde gehen

--UHN/qo2QbUvPLonB
Content-Type: text/plain; charset=us-ascii
Content-Disposition: attachment; filename="exception_vector.S"

/*	$NetBSD: exception_vector.S,v 1.28 2007/03/18 20:18:36 uwe Exp $	*/

/*-
 * Copyright (c) 2002 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *        This product includes software developed by the NetBSD
 *        Foundation, Inc. and its contributors.
 * 4. Neither the name of The NetBSD Foundation nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

#include "opt_cputype.h"
#include "opt_ddb.h"
#include "assym.h"

#include <sh3/param.h>
#include <sh3/locore.h>
#include <sh3/exception.h>
#include <sh3/ubcreg.h>
#include <sh3/pte.h>
#include <sh3/mmu_sh3.h>
#include <sh3/mmu_sh4.h>

/* 
 * Align vectors more strictly here (where we don't really care) so
 * that .align 5 (i.e. 32B cache line) before data block does the
 * right thing w.r.t. final destinations after vectors are copied.
 */
#define _ALIGN_TEXT	.align 5
#include <sh3/asm.h>

__KERNEL_RCSID(0, "$NetBSD: exception_vector.S,v 1.28 2007/03/18 20:18:36 uwe Exp $")

#undef TLBMISS_EVCNT		/* XXX */

/*
 * Exception vectors.
 * The following routines are copied to vector addresses.
 *	sh_vector_generic:	VBR + 0x100
 *	sh_vector_tlbmiss:	VBR + 0x400
 *	sh_vector_interrupt:	VBR + 0x600
 */

#define VECTOR_END_MARKER(sym)			\
		.globl	_C_LABEL(sym);		\
	_C_LABEL(sym):

#define __INC64(cnt, one_lo, one_hi, var_lo, var_hi, mem_lo, mem_hi) \
	mov	#1, one_lo		; \
	mov.l	@cnt, mem_lo		; \
	clrt				; \
	mov.l	@(4, cnt), mem_hi	; \
	mov	#0, one_hi		; \
	addc	one_lo, var_lo		; \
	addc	one_hi, var_hi		; \
	mov.l	mem_lo, @cnt		; \
	mov.l	mem_hi, @(4, cnt)

#if _BYTE_ORDER == _LITTLE_ENDIAN
#define INC64(cnt, one_lo, one_hi, var_lo, var_hi) \
	__INC64(cnt, one_lo, one_hi, var_lo, var_hi, var_lo, var_hi)
#else
#define INC64(cnt, one_lo, one_hi, var_lo, var_hi) \
	__INC64(cnt, one_lo, one_hi, var_lo, var_hi, var_hi, var_lo)
#endif


/*
 * LINTSTUB: Var: char sh_vector_generic[1];
 *
 * void sh_vector_generic(void);
 *	Copied to VBR+0x100.  This code should be position independent
 *	and maximum 786 bytes long (== 0x400 - 0x100).
 */
NENTRY(sh_vector_generic)
	__EXCEPTION_ENTRY
	__INTR_MASK(r0, r1)
	/* Identify exception cause */
	MOV	(EXPEVT, r0)
	mov.l	@r0, r0
	mov.l	r0, @(TF_EXPEVT, r14)	/* tf->tf_expevt = EXPEVT */
	/* Get curlwp */
	mov.l	.Lg_curlwp, r1
	mov.l	@r1, r4			/* 1st arg */
	/* Get TEA */
	MOV	(TEA, r1)
	mov.l	@r1, r6			/* 3rd arg */
	/* Check TLB exception or not */
	mov.l	.Lg_TLB_PROT_ST, r1
	cmp/hi	r1, r0
	bt	1f

	/* tlb_exception(curlwp, tf, TEA); */
	__EXCEPTION_UNBLOCK(r0, r1)
	mov.l	.Lg_tlb_exception, r0
	jsr	@r0
	 mov	r14, r5			/* 2nd arg */
	bra	2f
	 nop

	/* general_exception(curlwp, tf, TEA); */
1:	mov	r4, r8
#ifdef DDB
	mov	#0, r2
	MOV	(BBRA, r1)
	mov.w	r2, @r1			/* disable UBC */
	mov.l	r2, @(TF_UBC, r14)	/* clear tf->tf_ubc */
#endif /* DDB */
	__EXCEPTION_UNBLOCK(r0, r1)
	mov.l	.Lg_general_exception, r0
	jsr	@r0
	 mov	r14, r5			/* 2nd arg */

	/* Check for ASTs on exit to user mode. */
	mov	r8, r4
	mov.l	.Lg_ast, r0
	jsr	@r0
	 mov	r14, r5
#ifdef DDB	/* BBRA = tf->tf_ubc */
	__EXCEPTION_BLOCK(r0, r1)
	mov.l	@(TF_UBC, r14), r0
	MOV	(BBRA, r1)
	mov.w	r0, @r1
#endif /* DDB */
2:	__EXCEPTION_RETURN

	.align	5
.Lg_curlwp:		.long	_C_LABEL(curlwp)
REG_SYMBOL(EXPEVT)
REG_SYMBOL(BBRA)
REG_SYMBOL(TEA)
.Lg_tlb_exception:	.long	_C_LABEL(tlb_exception)
.Lg_general_exception:	.long	_C_LABEL(general_exception)
.Lg_ast:		.long	_C_LABEL(ast)
.Lg_TLB_PROT_ST:	.long	EXPEVT_TLB_PROT_ST

/* LINTSTUB: Var: char sh_vector_generic_end[1]; */
VECTOR_END_MARKER(sh_vector_generic_end)
	SET_ENTRY_SIZE(sh_vector_generic)


#ifdef SH3
/*
 * LINTSTUB: Var: char sh3_vector_tlbmiss[1];
 *
 * TLB miss vector.  We run through the fast path first, checking if
 * there's a valid mapping in curlwp or kernel pmap.  We do fast path
 * with exceptions disabled, so no P3 addresses please (including no
 * kernel stack, as we cannot wire TLB entries on sh3).  We can only
 * use BANK1 registers, and of those r6 and r7 are already taken.
 *
 * If we don't find a valid mapping in the fast path, we do context
 * save and call tlb exception handler.
 *
 * Copied to VBR+0x400.  This code should be position independent
 * and maximum 512 bytes long (== 0x600 - 0x400).
 */
NENTRY(sh3_vector_tlbmiss)
	mov	#(SH3_PTEH & 0xff), r4
	mov.l	.L3_VPN_cleanup, r0
	mov.l	@r4, r5
	and	r0, r5		! trim vpn to 4K page boundary
	!! For the duration of fast path we keep
	!! r4: SH3_PTEH - other PTE regs are addressable as @(offset, r4)
	!! r5: { VPN, ASID } that caused the miss

	cmp/pz	r5		! user space address?
	bt/s	.L3_user_va
	 mov	r5, r2		! copy of vpn to compute indices into ptd/ptp

	!! kernel space address, use pmap_kernel(), adjust vpn for indexing
	!! see __pmap_kpte_lookup
.L3_kernel_va:
	mov.l	.L3_VM_MIN_KERNEL_ADDRESS, r0
	mov.l	.L3_kernptd,  r1 ! pmap_kernel()->pm_ptp
	bra	.L3_fetch_pte
	 sub	r0, r2		! vpn -= VM_MIN_KERNEL_ADDRESS

	!! user space address, use curlwp's pmap
.L3_user_va:
	mov.l	.L3_curptd,  r1	! curlwp->...->pm_ptp

	!! see __pmap_pte_lookup
.L3_fetch_pte:
	mov.l	@r1, r3		! fetch ptd

	!! r2: vpn, prepared for indexing into ptd
	!! r3: pt_entry_t **ptd => pt_entry_t *ptp => pt_entry_t pte

#ifdef DEBUG
	tst	r3, r3		! ptd == NULL  - cannot happen
	bt/s	.L3_call_tlb_exception
#endif
	 mov	#-22, r1	! __PMAP_PTP_SHIFT

	!! __PMAP_PTP_INDEX(vpn)
	mov	r2, r0
	shld	r1, r0		! vpn >> __PMAP_PTP_SHIFT
	mov.l	.L3_ptp_index_mask, r1
	and	r1, r0		! ... & (__PMAP_PTP_N - 1)
	shll2	r0		! array index -> array offset
	mov.l	@(r0, r3), r3	! ptp = ptd[idx]
	tst	r3, r3		! if (ptp == NULL)
	bt/s	.L3_call_tlb_exception
	 mov	#-(PGSHIFT - 2), r1

	!! __PMAP_PTP_OFSET(vpn) - except we pre-shift 2 bits left to
	!! get the array offset directly, as we know bits 10 and 11
	!! are zero (we cleaned them in r5 to get 4K aligned VPN)
	shld	r1, r2		! vpn >> (PGSHIFT - 2)
	mov.l	.L3_ptp_offset_mask, r0
	and	r2, r0		! ... & ((__PMAP_PTP_PG_N - 1) << 2)
	mov.l	@(r0, r3), r3	! pte = ptp[idx]

	!! r3: pte
	!! r4: SH3_PTEH
	!! r5: { VPN, ASID }
.L3_load:
	mov.l	.L3_PG_V, r0
	tst	r0, r3		! if ((pte & PG_V) == 0)
	bt/s	.L3_call_tlb_exception
	 nop
	mov.l	.L3_PG_HW_BITS, r1
	cmp/pz	r5		! user space address?
	and	r1, r3		! pte &= PG_HW_BITS
	bf/s	.L3_load_kernel
	 mov.l	r3, @(0x04, r4)	! *SH3_PTEL = pte

	!! load mapping for a user space page
	!! we reload PTEH to enter VPN aligned to 4K page boundary
.L3_load_user:
	mov.l	r5, @r4		! *SH3_PTEH = { VPN, ASID }
	ldtlb			! needs 2 insns padding before RTE
#ifdef TLBMISS_EVCNT
	mov.l	.L3_ev_umiss, r1
	bra	.L3_update_evcnt
	 nop
#else
	nop
	nop
	rte
	 nop
#endif

	!! load mapping for a kernel space page
	!! we need to temporary set ASID to 0
.L3_load_kernel:
	mov.l	.L3_clear_ASID, r1
	and	r5, r1		! *SH3_PTEH & ~SH3_PTEH_ASID_MASK
	mov.l	r1, @r4		! *SH3_PTEH = { VPN, ASID = 0 }
	ldtlb
	mov.l	r5, @r4		! restore ASID

#ifdef TLBMISS_EVCNT
	mov.l	.L3_ev_kmiss, r1

.L3_update_evcnt:
	INC64(r1, r2, r3, r4, r5)
#else
	nop
#endif
	rte
	 nop


	!! if we haven't found a valid mapping in the fast path
	!!     tlb_exception(curlwp, trapframe, tea)
.L3_call_tlb_exception:
	__EXCEPTION_ENTRY
	mov.l	.L3_SH3_EXPEVT, r2
	mov.l	.L3_curlwp, r1
	mov	#(SH3_TEA & 0xff), r0
	mov.l	@r2, r2			! *SH3_EXPEVT
	mov.l	@r0, r6			! arg3: va = *SH3_TEA
	mov.l	@r1, r4			! arg1: curlwp
	__INTR_MASK(r0, r1)
	__EXCEPTION_UNBLOCK(r0, r1)
	mov.l	.L3_tlb_exception, r0
	mov.l	r2, @(TF_EXPEVT, r14)	! tf->tf_expevt = EXPEVT
	jsr	@r0
	 mov	r14, r5			! arg2: trapframe
	__EXCEPTION_RETURN

	.align	4
.L3_VPN_cleanup:		.long	~0x00000c00
.L3_curptd:			.long	_C_LABEL(curptd)
.L3_kernptd:			.long	_C_LABEL(__pmap_kernel)
.L3_VM_MIN_KERNEL_ADDRESS:	.long	VM_MIN_KERNEL_ADDRESS
.L3_ptp_index_mask:		.long	0x1ff
.L3_ptp_offset_mask:		.long	0x3ff << 2
.L3_PG_HW_BITS:			.long	PG_HW_BITS
.L3_PG_V:			.long	PG_V
.L3_clear_ASID:			.long	~SH3_PTEH_ASID_MASK
.L3_SH3_EXPEVT:			.long	SH3_EXPEVT
.L3_curlwp:			.long	_C_LABEL(curlwp)
.L3_tlb_exception:		.long	_C_LABEL(tlb_exception)
#ifdef TLBMISS_EVCNT
.L3_ev_umiss:			.long	_C_LABEL(ev_tlbmiss_ufast)
.L3_ev_kmiss:			.long	_C_LABEL(ev_tlbmiss_kfast)
#endif

/* LINTSTUB: Var: char sh3_vector_tlbmiss_end[1]; */
VECTOR_END_MARKER(sh3_vector_tlbmiss_end)
	SET_ENTRY_SIZE(sh3_vector_tlbmiss)

#endif /* SH3 */


#ifdef SH4
/*
 * LINTSTUB: Var: char sh4_vector_tlbmiss[1];
 *
 * TLB miss vector.  We run through the fast path first, checking if
 * there's a valid mapping in curlwp or kernel pmap.  We do fast path
 * with exceptions disabled, so no P3 addresses please (though we can
 * use kernel stack if need be, as its TLB entries are wired).  We can
 * only use BANK1 registers, and of those r6 and r7 are already taken.
 *
 * If we don't find a valid mapping in the fast path, we do context
 * save and call tlb exception handler.
 *
 * Copied to VBR+0x400.  This code should be relocatable
 * and maximum 512 bytes long (== 0x600 - 0x400).
 */
NENTRY(sh4_vector_tlbmiss)
	mov.l	.L4_SH4_PTEH, r4
	mov.l	.L4_VPN_cleanup, r0
	mov.l	@r4, r5
	and	r0, r5		! trim vpn to 4K page boundary
	!! For the duration of fast path we keep
	!! r4: SH4_PTEH - other PTE regs are addressable as @(offset, r4)
	!! r5: { VPN, ASID } that caused the miss

	cmp/pz	r5		! user space address?
	bt/s	.L4_user_va
	 mov	r5, r2		! copy of vpn to compute indices into ptd/ptp

	!! kernel space address, use pmap_kernel(), adjust vpn for indexing
	!! see __pmap_kpte_lookup
.L4_kernel_va:
	mov.l	.L4_VM_MIN_KERNEL_ADDRESS, r0
	mov.l	.L4_kernptd,  r1 ! pmap_kernel()->pm_ptp
	bra	.L4_fetch_pte
	 sub	r0, r2		! vpn -= VM_MIN_KERNEL_ADDRESS

	!! user space address, use curlwp's pmap
.L4_user_va:
	mov.l	.L4_curptd,  r1	! curlwp->...->pm_ptp

	!! see __pmap_pte_lookup
.L4_fetch_pte:
	mov.l	@r1, r3		! fetch ptd

	!! r2: vpn, prepared for indexing into ptd
	!! r3: pt_entry_t **ptd => pt_entry_t *ptp => pt_entry_t pte

#ifdef DEBUG
	tst	r3, r3		! ptd == NULL  - cannot happen
	bt/s	.L4_call_tlb_exception
#endif
	 mov	#-22, r1	! __PMAP_PTP_SHIFT

	!! __PMAP_PTP_INDEX(vpn)
	mov	r2, r0
	shld	r1, r0		! vpn >> __PMAP_PTP_SHIFT
	mov.l	.L4_ptp_index_mask, r1
	and	r1, r0		! ... & (__PMAP_PTP_N - 1)
	shll2	r0		! array index -> array offset
	mov.l	@(r0, r3), r3	! ptp = ptd[idx]
	tst	r3, r3		! if (ptp == NULL)
	bt/s	.L4_call_tlb_exception
	 mov	#-(PGSHIFT - 2), r1

	!! __PMAP_PTP_OFSET(vpn) - except we pre-shift 2 bits left to
	!! get the array offset directly, as we know bits 10 and 11
	!! are zero (we cleaned them in r5 to get 4K aligned VPN)
	shld	r1, r2		! vpn >> (PGSHIFT - 2)
	mov.l	.L4_ptp_offset_mask, r0
	and	r2, r0		! ... & ((__PMAP_PTP_PG_N - 1) << 2)
	mov.l	@(r0, r3), r3	! pte = ptp[idx]

	!! r3: pte
	!! r4: SH4_PTEH
	!! r5: { VPN, ASID }
.L4_load:
	mov.l	.L4_PG_V, r0
	tst	r0, r3		! if ((pte & PG_V) == 0)
	bt/s	.L4_call_tlb_exception
	 mov	r3, r0		! prepare PCMCIA SA bits for SH4_PTEA
	mov.l	.L4_PG_HW_BITS, r1
	shlr8	r0
	and	r1, r3		! pte &= PG_HW_BITS
	shlr	r0		! pte >> _PG_PCMCIA_SHIFT
	cmp/pz	r5		! user space address?
	and	#SH4_PTEA_SA_MASK, r0
	mov.l	r3, @(0x04, r4)	! *SH4_PTEL = pte
	bf/s	.L4_load_kernel
	 mov.l	r0, @(0x34, r4)	! *SH4_PTEA = PCMCIA space attrs

	!! load mapping for a user space page
	!! we reload PTEH to enter VPN aligned to 4K page boundary
.L4_load_user:
	mov.l	r5, @r4		! *SH4_PTEH = { VPN, ASID }
	ldtlb			! needs 1 insn padding before RTE
#ifdef TLBMISS_EVCNT
	mov.l	.L4_ev_umiss, r1
	bra	.L4_update_evcnt
	 nop
#else
	nop
	rte
	 nop
#endif

	!! load mapping for a kernel space page
	!! we need to temporary set ASID to 0
.L4_load_kernel:
	mov.l	.L4_clear_ASID, r1
	and	r5, r1		! *SH4_PTEH & ~SH4_PTEH_ASID_MASK
	mov.l	r1, @r4		! *SH4_PTEH = { VPN, ASID = 0 }
	ldtlb
	mov.l	r5, @r4		! restore ASID

#ifdef TLBMISS_EVCNT
	mov.l	.L4_ev_kmiss, r1

.L4_update_evcnt:
	INC64(r1, r2, r3, r4, r5)
#endif
	rte
	 nop


	!! if we haven't found a valid mapping in the fast path
	!!     tlb_exception(curlwp, trapframe, tea)
.L4_call_tlb_exception:
	__EXCEPTION_ENTRY
	mov.l	.L4_SH4_PTEH, r0
	mov.l	.L4_curlwp, r1
	mov.l	@(0x24, r0), r2		! *SH4_EXPEVT
	mov.l	@(0x0c, r0), r6		! arg3: va = *SH4_TEA
	mov.l	@r1, r4			! arg1: curlwp
	__INTR_MASK(r0, r1)
	__EXCEPTION_UNBLOCK(r0, r1)
	mov.l	.L4_tlb_exception, r0
	mov.l	r2, @(TF_EXPEVT, r14)	! tf->tf_expevt = EXPEVT
	jsr	@r0
	 mov	r14, r5			! arg2: trapframe
	__EXCEPTION_RETURN

	.align	5
.L4_SH4_PTEH:			.long	SH4_PTEH
.L4_VPN_cleanup:		.long	~0x00000c00
.L4_curptd:			.long	_C_LABEL(curptd)
.L4_kernptd:			.long	_C_LABEL(__pmap_kernel)
.L4_VM_MIN_KERNEL_ADDRESS:	.long	VM_MIN_KERNEL_ADDRESS
.L4_ptp_index_mask:		.long	0x1ff
.L4_ptp_offset_mask:		.long	0x3ff << 2
.L4_PG_HW_BITS:			.long	PG_HW_BITS
.L4_PG_V:			.long	PG_V
.L4_clear_ASID:			.long	~SH4_PTEH_ASID_MASK
.L4_curlwp:			.long	_C_LABEL(curlwp)
.L4_tlb_exception:		.long	_C_LABEL(tlb_exception)
#ifdef TLBMISS_EVCNT
.L4_ev_umiss:			.long	_C_LABEL(ev_tlbmiss_ufast)
.L4_ev_kmiss:			.long	_C_LABEL(ev_tlbmiss_kfast)
#endif

/* LINTSTUB: Var: char sh4_vector_tlbmiss_end[1]; */
VECTOR_END_MARKER(sh4_vector_tlbmiss_end)
	SET_ENTRY_SIZE(sh4_vector_tlbmiss)

#endif /* SH4 */


/*
 * LINTSTUB: Var: char sh_vector_interrupt[1];
 *
 * void sh_vector_interrupt(void);
 *	Copied to VBR+0x600.  This code should be position independent.
 */
NENTRY(sh_vector_interrupt)
	__EXCEPTION_ENTRY
	xor	r0, r0
	mov.l	r0, @(TF_EXPEVT, r14)	/* (for debug) */
	stc	r0_bank, r6		/* ssp */
	/* Enable exceptions for P3 access */
	__INTR_MASK(r0, r1)
	__EXCEPTION_UNBLOCK(r0, r1)
	/* ++uvmexp.intrs */
	mov.l	.Li_uvmexp_intrs, r0
	mov.l	@r0, r1
	add	#1 r1
	mov.l	r1, @r0
	/* Dispatch interrupt handler */
	mov.l	.Li_intc_intr, r0
	jsr	@r0		/* intc_intr(ssr, spc, ssp) */
	 nop
	/* Check for ASTs on exit to user mode. */
	mov.l	.Li_curlwp, r0
	mov.l	@r0, r4		/* 1st arg */
	mov.l	.Li_ast, r0
	jsr	@r0
	 mov	r14, r5		/* 2nd arg */
	__EXCEPTION_RETURN

	.align	5
.Li_curlwp:		.long	_C_LABEL(curlwp)
.Li_intc_intr:		.long	_C_LABEL(intc_intr)
.Li_ast:		.long	_C_LABEL(ast)
.Li_uvmexp_intrs:	.long	_C_LABEL(uvmexp) + UVMEXP_INTRS

/* LINTSTUB: Var: char sh_vector_interrupt_end[1]; */
VECTOR_END_MARKER(sh_vector_interrupt_end)
	SET_ENTRY_SIZE(sh_vector_interrupt)

--UHN/qo2QbUvPLonB
Content-Type: text/plain; charset=us-ascii
Content-Disposition: attachment; filename="pmap.c-diff"

Index: pmap.c
===================================================================
RCS file: /cvsroot/src/sys/arch/sh3/sh3/pmap.c,v
retrieving revision 1.62
diff -u --unified -r1.62 pmap.c
--- pmap.c	12 Mar 2007 18:18:26 -0000	1.62
+++ pmap.c	27 Mar 2007 22:55:49 -0000
@@ -68,6 +68,9 @@
 paddr_t avail_start;		/* PA of first available physical page */
 paddr_t avail_end;		/* PA of last available physical page */
 
+/* For the fast tlb miss handler */
+pt_entry_t **curptd;		/* p1 va of curlwp->...->pm_ptp */
+
 /* pmap pool */
 STATIC struct pool __pmap_pmap_pool;
 
@@ -313,7 +316,9 @@
 		pmap->pm_asid = __pmap_asid_alloc();
 
 	KDASSERT(pmap->pm_asid >=0 && pmap->pm_asid < 256);
+
 	sh_tlb_set_asid(pmap->pm_asid);
+	curptd = pmap->pm_ptp;
 }
 
 void

--UHN/qo2QbUvPLonB--