Skip to content

L
linux
Commit fb2a27e7 authored by Thiemo Seufer's avatar Thiemo Seufer Committed by Ralf Baechle
Browse files
Options

[MIPS] Reimplement clear_page/copy_page 

Fold the SB-1 specific implementation of clear_page/copy_page in the
generic version, and rewrite that one in tlbex style. The immediate
benefits:
  - It converts the compile-time workaround for SB-1 pass 1 prefetches
    to a more efficient run-time check.
  - It allows adjustment of loop unfolling, which helps to reduce the
    number of redundant cdex cache ops.
  - It fixes some esoteric cornercases (the cache line length calculations
    can go wrong, and support for 64k pages without prefetch instructions
    will overflow the addiu immediate).
  - Somewhat better guesses of "good" prefetch values.
Signed-off-by: default avatarThiemo Seufer <ths@networkno.de>
Signed-off-by: default avatarRalf Baechle <ralf@linux-mips.org>
parent 064922a8
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 725 additions and 862 deletions
arch/mips/mm/Makefile
View file @ fb2a27e7
......@@ -4,30 +4,29 @@
obj-y += cache.o dma-default.o extable.o fault.o \
init.o pgtable.o tlbex.o tlbex-fault.o \
uasm.o
uasm.o page.o
obj-$(CONFIG_32BIT) += ioremap.o pgtable-32.o
obj-$(CONFIG_64BIT) += pgtable-64.o
obj-$(CONFIG_HIGHMEM) += highmem.o
obj-$(CONFIG_CPU_LOONGSON2) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_MIPS32) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_MIPS64) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_NEVADA) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_R10000) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_R3000) += c-r3k.o tlb-r3k.o pg-r4k.o
obj-$(CONFIG_CPU_R4300) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_R4X00) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_R5000) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_R5432) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_R8000) += c-r4k.o cex-gen.o pg-r4k.o tlb-r8k.o
obj-$(CONFIG_CPU_RM7000) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_RM9000) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_SB1) += c-r4k.o cerr-sb1.o cex-sb1.o pg-sb1.o \
tlb-r4k.o
obj-$(CONFIG_CPU_TX39XX) += c-tx39.o pg-r4k.o tlb-r3k.o
obj-$(CONFIG_CPU_TX49XX) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_VR41XX) += c-r4k.o cex-gen.o pg-r4k.o tlb-r4k.o
obj-$(CONFIG_CPU_LOONGSON2) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_MIPS32) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_MIPS64) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_NEVADA) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_R10000) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_R3000) += c-r3k.o tlb-r3k.o
obj-$(CONFIG_CPU_R4300) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_R4X00) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_R5000) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_R5432) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_R8000) += c-r4k.o cex-gen.o tlb-r8k.o
obj-$(CONFIG_CPU_RM7000) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_RM9000) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_SB1) += c-r4k.o cerr-sb1.o cex-sb1.o tlb-r4k.o
obj-$(CONFIG_CPU_TX39XX) += c-tx39.o tlb-r3k.o
obj-$(CONFIG_CPU_TX49XX) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_CPU_VR41XX) += c-r4k.o cex-gen.o tlb-r4k.o
obj-$(CONFIG_IP22_CPU_SCACHE) += sc-ip22.o
obj-$(CONFIG_R5000_CPU_SCACHE) += sc-r5k.o
......
arch/mips/mm/page.c 0 → 100644
View file @ fb2a27e7
This diff is collapsed.
arch/mips/mm/pg-r4k.c deleted 100644 → 0
View file @ 064922a8
This diff is collapsed.
arch/mips/mm/pg-sb1.c deleted 100644 → 0
View file @ 064922a8
/*
* Copyright (C) 1996 David S. Miller (dm@engr.sgi.com)
* Copyright (C) 1997, 2001 Ralf Baechle (ralf@gnu.org)
* Copyright (C) 2000 SiByte, Inc.
* Copyright (C) 2005 Thiemo Seufer
*
* Written by Justin Carlson of SiByte, Inc.
* and Kip Walker of Broadcom Corp.
*
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
*/
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/smp.h>
#include <asm/io.h>
#include <asm/sibyte/sb1250.h>
#include <asm/sibyte/sb1250_regs.h>
#include <asm/sibyte/sb1250_dma.h>
#ifdef CONFIG_SB1_PASS_1_WORKAROUNDS
#define SB1_PREF_LOAD_STREAMED_HINT "0"
#define SB1_PREF_STORE_STREAMED_HINT "1"
#else
#define SB1_PREF_LOAD_STREAMED_HINT "4"
#define SB1_PREF_STORE_STREAMED_HINT "5"
#endif
static inline void clear_page_cpu(void *page)
{
unsigned char *addr = (unsigned char *) page;
unsigned char *end = addr + PAGE_SIZE;
/*
* JDCXXX - This should be bottlenecked by the write buffer, but these
* things tend to be mildly unpredictable...should check this on the
* performance model
*
* We prefetch 4 lines ahead. We're also "cheating" slightly here...
* since we know we're on an SB1, we force the assembler to take
* 64-bit operands to speed things up
*/
__asm__ __volatile__(
" .set push \n"
" .set mips4 \n"
" .set noreorder \n"
#ifdef CONFIG_CPU_HAS_PREFETCH
" daddiu %0, %0, 128 \n"
" pref " SB1_PREF_STORE_STREAMED_HINT ", -128(%0) \n"
/* Prefetch the first 4 lines */
" pref " SB1_PREF_STORE_STREAMED_HINT ", -96(%0) \n"
" pref " SB1_PREF_STORE_STREAMED_HINT ", -64(%0) \n"
" pref " SB1_PREF_STORE_STREAMED_HINT ", -32(%0) \n"
"1: sd $0, -128(%0) \n" /* Throw out a cacheline of 0's */
" sd $0, -120(%0) \n"
" sd $0, -112(%0) \n"
" sd $0, -104(%0) \n"
" daddiu %0, %0, 32 \n"
" bnel %0, %1, 1b \n"
" pref " SB1_PREF_STORE_STREAMED_HINT ", -32(%0) \n"
" daddiu %0, %0, -128 \n"
#endif
" sd $0, 0(%0) \n" /* Throw out a cacheline of 0's */
"1: sd $0, 8(%0) \n"
" sd $0, 16(%0) \n"
" sd $0, 24(%0) \n"
" daddiu %0, %0, 32 \n"
" bnel %0, %1, 1b \n"
" sd $0, 0(%0) \n"
" .set pop \n"
: "+r" (addr)
: "r" (end)
: "memory");
}
static inline void copy_page_cpu(void *to, void *from)
{
unsigned char *src = (unsigned char *)from;
unsigned char *dst = (unsigned char *)to;
unsigned char *end = src + PAGE_SIZE;
/*
* The pref's used here are using "streaming" hints, which cause the
* copied data to be kicked out of the cache sooner. A page copy often
* ends up copying a lot more data than is commonly used, so this seems
* to make sense in terms of reducing cache pollution, but I've no real
* performance data to back this up
*/
__asm__ __volatile__(
" .set push \n"
" .set mips4 \n"
" .set noreorder \n"
#ifdef CONFIG_CPU_HAS_PREFETCH
" daddiu %0, %0, 128 \n"
" daddiu %1, %1, 128 \n"
" pref " SB1_PREF_LOAD_STREAMED_HINT ", -128(%0)\n"
/* Prefetch the first 4 lines */
" pref " SB1_PREF_STORE_STREAMED_HINT ", -128(%1)\n"
" pref " SB1_PREF_LOAD_STREAMED_HINT ", -96(%0)\n"
" pref " SB1_PREF_STORE_STREAMED_HINT ", -96(%1)\n"
" pref " SB1_PREF_LOAD_STREAMED_HINT ", -64(%0)\n"
" pref " SB1_PREF_STORE_STREAMED_HINT ", -64(%1)\n"
" pref " SB1_PREF_LOAD_STREAMED_HINT ", -32(%0)\n"
"1: pref " SB1_PREF_STORE_STREAMED_HINT ", -32(%1)\n"
# ifdef CONFIG_64BIT
" ld $8, -128(%0) \n" /* Block copy a cacheline */
" ld $9, -120(%0) \n"
" ld $10, -112(%0) \n"
" ld $11, -104(%0) \n"
" sd $8, -128(%1) \n"
" sd $9, -120(%1) \n"
" sd $10, -112(%1) \n"
" sd $11, -104(%1) \n"
# else
" lw $2, -128(%0) \n" /* Block copy a cacheline */
" lw $3, -124(%0) \n"
" lw $6, -120(%0) \n"
" lw $7, -116(%0) \n"
" lw $8, -112(%0) \n"
" lw $9, -108(%0) \n"
" lw $10, -104(%0) \n"
" lw $11, -100(%0) \n"
" sw $2, -128(%1) \n"
" sw $3, -124(%1) \n"
" sw $6, -120(%1) \n"
" sw $7, -116(%1) \n"
" sw $8, -112(%1) \n"
" sw $9, -108(%1) \n"
" sw $10, -104(%1) \n"
" sw $11, -100(%1) \n"
# endif
" daddiu %0, %0, 32 \n"
" daddiu %1, %1, 32 \n"
" bnel %0, %2, 1b \n"
" pref " SB1_PREF_LOAD_STREAMED_HINT ", -32(%0)\n"
" daddiu %0, %0, -128 \n"
" daddiu %1, %1, -128 \n"
#endif
#ifdef CONFIG_64BIT
" ld $8, 0(%0) \n" /* Block copy a cacheline */
"1: ld $9, 8(%0) \n"
" ld $10, 16(%0) \n"
" ld $11, 24(%0) \n"
" sd $8, 0(%1) \n"
" sd $9, 8(%1) \n"
" sd $10, 16(%1) \n"
" sd $11, 24(%1) \n"
#else
" lw $2, 0(%0) \n" /* Block copy a cacheline */
"1: lw $3, 4(%0) \n"
" lw $6, 8(%0) \n"
" lw $7, 12(%0) \n"
" lw $8, 16(%0) \n"
" lw $9, 20(%0) \n"
" lw $10, 24(%0) \n"
" lw $11, 28(%0) \n"
" sw $2, 0(%1) \n"
" sw $3, 4(%1) \n"
" sw $6, 8(%1) \n"
" sw $7, 12(%1) \n"
" sw $8, 16(%1) \n"
" sw $9, 20(%1) \n"
" sw $10, 24(%1) \n"
" sw $11, 28(%1) \n"
#endif
" daddiu %0, %0, 32 \n"
" daddiu %1, %1, 32 \n"
" bnel %0, %2, 1b \n"
#ifdef CONFIG_64BIT
" ld $8, 0(%0) \n"
#else
" lw $2, 0(%0) \n"
#endif
" .set pop \n"
: "+r" (src), "+r" (dst)
: "r" (end)
#ifdef CONFIG_64BIT
: "$8", "$9", "$10", "$11", "memory");
#else
: "$2", "$3", "$6", "$7", "$8", "$9", "$10", "$11", "memory");
#endif
}
#ifdef CONFIG_SIBYTE_DMA_PAGEOPS
/*
* Pad descriptors to cacheline, since each is exclusively owned by a
* particular CPU.
*/
typedef struct dmadscr_s {
u64 dscr_a;
u64 dscr_b;
u64 pad_a;
u64 pad_b;
} dmadscr_t;
static dmadscr_t page_descr[DM_NUM_CHANNELS]
__attribute__((aligned(SMP_CACHE_BYTES)));
void sb1_dma_init(void)
{
int i;
for (i = 0; i < DM_NUM_CHANNELS; i++) {
const u64 base_val = CPHYSADDR((unsigned long)&page_descr[i]) |
V_DM_DSCR_BASE_RINGSZ(1);
void *base_reg = IOADDR(A_DM_REGISTER(i, R_DM_DSCR_BASE));
__raw_writeq(base_val, base_reg);
__raw_writeq(base_val | M_DM_DSCR_BASE_RESET, base_reg);
__raw_writeq(base_val | M_DM_DSCR_BASE_ENABL, base_reg);
}
}
void clear_page(void *page)
{
u64 to_phys = CPHYSADDR((unsigned long)page);
unsigned int cpu = smp_processor_id();
/* if the page is not in KSEG0, use old way */
if ((long)KSEGX((unsigned long)page) != (long)CKSEG0)
return clear_page_cpu(page);
page_descr[cpu].dscr_a = to_phys | M_DM_DSCRA_ZERO_MEM |
M_DM_DSCRA_L2C_DEST | M_DM_DSCRA_INTERRUPT;
page_descr[cpu].dscr_b = V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE);
__raw_writeq(1, IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT)));
/*
* Don't really want to do it this way, but there's no
* reliable way to delay completion detection.
*/
while (!(__raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG)))
& M_DM_DSCR_BASE_INTERRUPT))
;
__raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE)));
}
void copy_page(void *to, void *from)
{
u64 from_phys = CPHYSADDR((unsigned long)from);
u64 to_phys = CPHYSADDR((unsigned long)to);
unsigned int cpu = smp_processor_id();
/* if any page is not in KSEG0, use old way */
if ((long)KSEGX((unsigned long)to) != (long)CKSEG0
|| (long)KSEGX((unsigned long)from) != (long)CKSEG0)
return copy_page_cpu(to, from);
page_descr[cpu].dscr_a = to_phys | M_DM_DSCRA_L2C_DEST |
M_DM_DSCRA_INTERRUPT;
page_descr[cpu].dscr_b = from_phys | V_DM_DSCRB_SRC_LENGTH(PAGE_SIZE);
__raw_writeq(1, IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_COUNT)));
/*
* Don't really want to do it this way, but there's no
* reliable way to delay completion detection.
*/
while (!(__raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE_DEBUG)))
& M_DM_DSCR_BASE_INTERRUPT))
;
__raw_readq(IOADDR(A_DM_REGISTER(cpu, R_DM_DSCR_BASE)));
}
#else /* !CONFIG_SIBYTE_DMA_PAGEOPS */
void clear_page(void *page)
{
return clear_page_cpu(page);
}
void copy_page(void *to, void *from)
{
return copy_page_cpu(to, from);
}
#endif /* !CONFIG_SIBYTE_DMA_PAGEOPS */
EXPORT_SYMBOL(clear_page);
EXPORT_SYMBOL(copy_page);
void __cpuinit build_clear_page(void)
{
}
void __cpuinit build_copy_page(void)
{
}
arch/mips/mm/uasm.c
View file @ fb2a27e7
......@@ -58,13 +58,13 @@ enum opcode {
insn_invalid,
insn_addu, insn_addiu, insn_and, insn_andi, insn_beq,
insn_beql, insn_bgez, insn_bgezl, insn_bltz, insn_bltzl,
insn_bne, insn_daddu, insn_daddiu, insn_dmfc0, insn_dmtc0,
insn_dsll, insn_dsll32, insn_dsra, insn_dsrl, insn_dsrl32,
insn_dsubu, insn_eret, insn_j, insn_jal, insn_jr, insn_ld,
insn_ll, insn_lld, insn_lui, insn_lw, insn_mfc0, insn_mtc0,
insn_ori, insn_rfe, insn_sc, insn_scd, insn_sd, insn_sll,
insn_sra, insn_srl, insn_subu, insn_sw, insn_tlbp, insn_tlbwi,
insn_tlbwr, insn_xor, insn_xori
insn_bne, insn_cache, insn_daddu, insn_daddiu, insn_dmfc0,
insn_dmtc0, insn_dsll, insn_dsll32, insn_dsra, insn_dsrl,
insn_dsrl32, insn_dsubu, insn_eret, insn_j, insn_jal, insn_jr,
insn_ld, insn_ll, insn_lld, insn_lui, insn_lw, insn_mfc0,
insn_mtc0, insn_ori, insn_pref, insn_rfe, insn_sc, insn_scd,
insn_sd, insn_sll, insn_sra, insn_srl, insn_subu, insn_sw,
insn_tlbp, insn_tlbwi, insn_tlbwr, insn_xor, insn_xori
};
struct insn {
......@@ -94,6 +94,7 @@ static struct insn insn_table[] __cpuinitdata = {
{ insn_bltz, M(bcond_op, 0, bltz_op, 0, 0, 0), RS | BIMM },
{ insn_bltzl, M(bcond_op, 0, bltzl_op, 0, 0, 0), RS | BIMM },
{ insn_bne, M(bne_op, 0, 0, 0, 0, 0), RS | RT | BIMM },
{ insn_cache, M(cache_op, 0, 0, 0, 0, 0), RS | RT | SIMM },
{ insn_daddiu, M(daddiu_op, 0, 0, 0, 0, 0), RS | RT | SIMM },
{ insn_daddu, M(spec_op, 0, 0, 0, 0, daddu_op), RS | RT | RD },
{ insn_dmfc0, M(cop0_op, dmfc_op, 0, 0, 0, 0), RT | RD | SET},
......@@ -116,6 +117,7 @@ static struct insn insn_table[] __cpuinitdata = {
{ insn_mfc0, M(cop0_op, mfc_op, 0, 0, 0, 0), RT | RD | SET},
{ insn_mtc0, M(cop0_op, mtc_op, 0, 0, 0, 0), RT | RD | SET},
{ insn_ori, M(ori_op, 0, 0, 0, 0, 0), RS | RT | UIMM },
{ insn_pref, M(pref_op, 0, 0, 0, 0, 0), RS | RT | SIMM },
{ insn_rfe, M(cop0_op, cop_op, 0, 0, 0, rfe_op), 0 },
{ insn_sc, M(sc_op, 0, 0, 0, 0, 0), RS | RT | SIMM },
{ insn_scd, M(scd_op, 0, 0, 0, 0, 0), RS | RT | SIMM },
......@@ -337,6 +339,7 @@ I_u1s2(_bgezl)
I_u1s2(_bltz)
I_u1s2(_bltzl)
I_u1u2s3(_bne)
I_u2s3u1(_cache)
I_u1u2u3(_dmfc0)
I_u1u2u3(_dmtc0)
I_u2u1s3(_daddiu)
......@@ -359,6 +362,7 @@ I_u2s3u1(_lw)
I_u1u2u3(_mfc0)
I_u1u2u3(_mtc0)
I_u2u1u3(_ori)
I_u2s3u1(_pref)
I_0(_rfe)
I_u2s3u1(_sc)
I_u2s3u1(_scd)
......@@ -554,6 +558,14 @@ uasm_il_beqzl(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid)
uasm_i_beqzl(p, reg, 0);
}
void __cpuinit
uasm_il_bne(u32 **p, struct uasm_reloc **r, unsigned int reg1,
unsigned int reg2, int lid)
{
uasm_r_mips_pc16(r, *p, lid);
uasm_i_bne(p, reg1, reg2, 0);
}
void __cpuinit
uasm_il_bnez(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid)
{
......
arch/mips/mm/uasm.h
View file @ fb2a27e7
......@@ -55,6 +55,7 @@ Ip_u1s2(_bgezl);
Ip_u1s2(_bltz);
Ip_u1s2(_bltzl);
Ip_u1u2s3(_bne);
Ip_u2s3u1(_cache);
Ip_u1u2u3(_dmfc0);
Ip_u1u2u3(_dmtc0);
Ip_u2u1s3(_daddiu);
......@@ -77,6 +78,7 @@ Ip_u2s3u1(_lw);
Ip_u1u2u3(_mfc0);
Ip_u1u2u3(_mtc0);
Ip_u2u1u3(_ori);
Ip_u2s3u1(_pref);
Ip_0(_rfe);
Ip_u2s3u1(_sc);
Ip_u2s3u1(_scd);
......@@ -177,6 +179,8 @@ void uasm_il_bltz(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid);
void uasm_il_b(u32 **p, struct uasm_reloc **r, int lid);
void uasm_il_beqz(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid);
void uasm_il_beqzl(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid);
void uasm_il_bne(u32 **p, struct uasm_reloc **r, unsigned int reg1,
unsigned int reg2, int lid);
void uasm_il_bnez(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid);
void uasm_il_bgezl(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid);
void uasm_il_bgez(u32 **p, struct uasm_reloc **r, unsigned int reg, int lid);
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7