diff options
Diffstat (limited to 'arch/sparc/mm/hypersparc.S')
| -rw-r--r-- | arch/sparc/mm/hypersparc.S | 414 |
1 files changed, 414 insertions, 0 deletions
diff --git a/arch/sparc/mm/hypersparc.S b/arch/sparc/mm/hypersparc.S new file mode 100644 index 000000000..6c2521e85 --- /dev/null +++ b/arch/sparc/mm/hypersparc.S @@ -0,0 +1,414 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* + * hypersparc.S: High speed Hypersparc mmu/cache operations. + * + * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu) + */ + +#include <asm/ptrace.h> +#include <asm/psr.h> +#include <asm/asm-offsets.h> +#include <asm/asi.h> +#include <asm/page.h> +#include <asm/pgtable.h> +#include <asm/pgtsrmmu.h> +#include <linux/init.h> + + .text + .align 4 + + .globl hypersparc_flush_cache_all, hypersparc_flush_cache_mm + .globl hypersparc_flush_cache_range, hypersparc_flush_cache_page + .globl hypersparc_flush_page_to_ram + .globl hypersparc_flush_page_for_dma, hypersparc_flush_sig_insns + .globl hypersparc_flush_tlb_all, hypersparc_flush_tlb_mm + .globl hypersparc_flush_tlb_range, hypersparc_flush_tlb_page + +hypersparc_flush_cache_all: + WINDOW_FLUSH(%g4, %g5) + sethi %hi(vac_cache_size), %g4 + ld [%g4 + %lo(vac_cache_size)], %g5 + sethi %hi(vac_line_size), %g1 + ld [%g1 + %lo(vac_line_size)], %g2 +1: + subcc %g5, %g2, %g5 ! hyper_flush_unconditional_combined + bne 1b + sta %g0, [%g5] ASI_M_FLUSH_CTX + retl + sta %g0, [%g0] ASI_M_FLUSH_IWHOLE ! hyper_flush_whole_icache + + /* We expand the window flush to get maximum performance. */ +hypersparc_flush_cache_mm: +#ifndef CONFIG_SMP + ld [%o0 + AOFF_mm_context], %g1 + cmp %g1, -1 + be hypersparc_flush_cache_mm_out +#endif + WINDOW_FLUSH(%g4, %g5) + + sethi %hi(vac_line_size), %g1 + ld [%g1 + %lo(vac_line_size)], %o1 + sethi %hi(vac_cache_size), %g2 + ld [%g2 + %lo(vac_cache_size)], %o0 + add %o1, %o1, %g1 + add %o1, %g1, %g2 + add %o1, %g2, %g3 + add %o1, %g3, %g4 + add %o1, %g4, %g5 + add %o1, %g5, %o4 + add %o1, %o4, %o5 + + /* BLAMMO! */ +1: + subcc %o0, %o5, %o0 ! hyper_flush_cache_user + sta %g0, [%o0 + %g0] ASI_M_FLUSH_USER + sta %g0, [%o0 + %o1] ASI_M_FLUSH_USER + sta %g0, [%o0 + %g1] ASI_M_FLUSH_USER + sta %g0, [%o0 + %g2] ASI_M_FLUSH_USER + sta %g0, [%o0 + %g3] ASI_M_FLUSH_USER + sta %g0, [%o0 + %g4] ASI_M_FLUSH_USER + sta %g0, [%o0 + %g5] ASI_M_FLUSH_USER + bne 1b + sta %g0, [%o0 + %o4] ASI_M_FLUSH_USER +hypersparc_flush_cache_mm_out: + retl + nop + + /* The things we do for performance... */ +hypersparc_flush_cache_range: + ld [%o0 + VMA_VM_MM], %o0 +#ifndef CONFIG_SMP + ld [%o0 + AOFF_mm_context], %g1 + cmp %g1, -1 + be hypersparc_flush_cache_range_out +#endif + WINDOW_FLUSH(%g4, %g5) + + sethi %hi(vac_line_size), %g1 + ld [%g1 + %lo(vac_line_size)], %o4 + sethi %hi(vac_cache_size), %g2 + ld [%g2 + %lo(vac_cache_size)], %o3 + + /* Here comes the fun part... */ + add %o2, (PAGE_SIZE - 1), %o2 + andn %o1, (PAGE_SIZE - 1), %o1 + add %o4, %o4, %o5 + andn %o2, (PAGE_SIZE - 1), %o2 + add %o4, %o5, %g1 + sub %o2, %o1, %g4 + add %o4, %g1, %g2 + sll %o3, 2, %g5 + add %o4, %g2, %g3 + cmp %g4, %g5 + add %o4, %g3, %g4 + blu 0f + add %o4, %g4, %g5 + add %o4, %g5, %g7 + + /* Flush entire user space, believe it or not this is quicker + * than page at a time flushings for range > (cache_size<<2). + */ +1: + subcc %o3, %g7, %o3 + sta %g0, [%o3 + %g0] ASI_M_FLUSH_USER + sta %g0, [%o3 + %o4] ASI_M_FLUSH_USER + sta %g0, [%o3 + %o5] ASI_M_FLUSH_USER + sta %g0, [%o3 + %g1] ASI_M_FLUSH_USER + sta %g0, [%o3 + %g2] ASI_M_FLUSH_USER + sta %g0, [%o3 + %g3] ASI_M_FLUSH_USER + sta %g0, [%o3 + %g4] ASI_M_FLUSH_USER + bne 1b + sta %g0, [%o3 + %g5] ASI_M_FLUSH_USER + retl + nop + + /* Below our threshold, flush one page at a time. */ +0: + ld [%o0 + AOFF_mm_context], %o0 + mov SRMMU_CTX_REG, %g7 + lda [%g7] ASI_M_MMUREGS, %o3 + sta %o0, [%g7] ASI_M_MMUREGS + add %o2, -PAGE_SIZE, %o0 +1: + or %o0, 0x400, %g7 + lda [%g7] ASI_M_FLUSH_PROBE, %g7 + orcc %g7, 0, %g0 + be,a 3f + mov %o0, %o2 + add %o4, %g5, %g7 +2: + sub %o2, %g7, %o2 + sta %g0, [%o2 + %g0] ASI_M_FLUSH_PAGE + sta %g0, [%o2 + %o4] ASI_M_FLUSH_PAGE + sta %g0, [%o2 + %o5] ASI_M_FLUSH_PAGE + sta %g0, [%o2 + %g1] ASI_M_FLUSH_PAGE + sta %g0, [%o2 + %g2] ASI_M_FLUSH_PAGE + sta %g0, [%o2 + %g3] ASI_M_FLUSH_PAGE + andcc %o2, 0xffc, %g0 + sta %g0, [%o2 + %g4] ASI_M_FLUSH_PAGE + bne 2b + sta %g0, [%o2 + %g5] ASI_M_FLUSH_PAGE +3: + cmp %o2, %o1 + bne 1b + add %o2, -PAGE_SIZE, %o0 + mov SRMMU_FAULT_STATUS, %g5 + lda [%g5] ASI_M_MMUREGS, %g0 + mov SRMMU_CTX_REG, %g7 + sta %o3, [%g7] ASI_M_MMUREGS +hypersparc_flush_cache_range_out: + retl + nop + + /* HyperSparc requires a valid mapping where we are about to flush + * in order to check for a physical tag match during the flush. + */ + /* Verified, my ass... */ +hypersparc_flush_cache_page: + ld [%o0 + VMA_VM_MM], %o0 + ld [%o0 + AOFF_mm_context], %g2 +#ifndef CONFIG_SMP + cmp %g2, -1 + be hypersparc_flush_cache_page_out +#endif + WINDOW_FLUSH(%g4, %g5) + + sethi %hi(vac_line_size), %g1 + ld [%g1 + %lo(vac_line_size)], %o4 + mov SRMMU_CTX_REG, %o3 + andn %o1, (PAGE_SIZE - 1), %o1 + lda [%o3] ASI_M_MMUREGS, %o2 + sta %g2, [%o3] ASI_M_MMUREGS + or %o1, 0x400, %o5 + lda [%o5] ASI_M_FLUSH_PROBE, %g1 + orcc %g0, %g1, %g0 + be 2f + add %o4, %o4, %o5 + sub %o1, -PAGE_SIZE, %o1 + add %o4, %o5, %g1 + add %o4, %g1, %g2 + add %o4, %g2, %g3 + add %o4, %g3, %g4 + add %o4, %g4, %g5 + add %o4, %g5, %g7 + + /* BLAMMO! */ +1: + sub %o1, %g7, %o1 + sta %g0, [%o1 + %g0] ASI_M_FLUSH_PAGE + sta %g0, [%o1 + %o4] ASI_M_FLUSH_PAGE + sta %g0, [%o1 + %o5] ASI_M_FLUSH_PAGE + sta %g0, [%o1 + %g1] ASI_M_FLUSH_PAGE + sta %g0, [%o1 + %g2] ASI_M_FLUSH_PAGE + sta %g0, [%o1 + %g3] ASI_M_FLUSH_PAGE + andcc %o1, 0xffc, %g0 + sta %g0, [%o1 + %g4] ASI_M_FLUSH_PAGE + bne 1b + sta %g0, [%o1 + %g5] ASI_M_FLUSH_PAGE +2: + mov SRMMU_FAULT_STATUS, %g7 + mov SRMMU_CTX_REG, %g4 + lda [%g7] ASI_M_MMUREGS, %g0 + sta %o2, [%g4] ASI_M_MMUREGS +hypersparc_flush_cache_page_out: + retl + nop + +hypersparc_flush_sig_insns: + flush %o1 + retl + flush %o1 + 4 + + /* HyperSparc is copy-back. */ +hypersparc_flush_page_to_ram: + sethi %hi(vac_line_size), %g1 + ld [%g1 + %lo(vac_line_size)], %o4 + andn %o0, (PAGE_SIZE - 1), %o0 + add %o4, %o4, %o5 + or %o0, 0x400, %g7 + lda [%g7] ASI_M_FLUSH_PROBE, %g5 + add %o4, %o5, %g1 + orcc %g5, 0, %g0 + be 2f + add %o4, %g1, %g2 + add %o4, %g2, %g3 + sub %o0, -PAGE_SIZE, %o0 + add %o4, %g3, %g4 + add %o4, %g4, %g5 + add %o4, %g5, %g7 + + /* BLAMMO! */ +1: + sub %o0, %g7, %o0 + sta %g0, [%o0 + %g0] ASI_M_FLUSH_PAGE + sta %g0, [%o0 + %o4] ASI_M_FLUSH_PAGE + sta %g0, [%o0 + %o5] ASI_M_FLUSH_PAGE + sta %g0, [%o0 + %g1] ASI_M_FLUSH_PAGE + sta %g0, [%o0 + %g2] ASI_M_FLUSH_PAGE + sta %g0, [%o0 + %g3] ASI_M_FLUSH_PAGE + andcc %o0, 0xffc, %g0 + sta %g0, [%o0 + %g4] ASI_M_FLUSH_PAGE + bne 1b + sta %g0, [%o0 + %g5] ASI_M_FLUSH_PAGE +2: + mov SRMMU_FAULT_STATUS, %g1 + retl + lda [%g1] ASI_M_MMUREGS, %g0 + + /* HyperSparc is IO cache coherent. */ +hypersparc_flush_page_for_dma: + retl + nop + + /* It was noted that at boot time a TLB flush all in a delay slot + * can deliver an illegal instruction to the processor if the timing + * is just right... + */ +hypersparc_flush_tlb_all: + mov 0x400, %g1 + sta %g0, [%g1] ASI_M_FLUSH_PROBE + retl + nop + +hypersparc_flush_tlb_mm: + mov SRMMU_CTX_REG, %g1 + ld [%o0 + AOFF_mm_context], %o1 + lda [%g1] ASI_M_MMUREGS, %g5 +#ifndef CONFIG_SMP + cmp %o1, -1 + be hypersparc_flush_tlb_mm_out +#endif + mov 0x300, %g2 + sta %o1, [%g1] ASI_M_MMUREGS + sta %g0, [%g2] ASI_M_FLUSH_PROBE +hypersparc_flush_tlb_mm_out: + retl + sta %g5, [%g1] ASI_M_MMUREGS + +hypersparc_flush_tlb_range: + ld [%o0 + VMA_VM_MM], %o0 + mov SRMMU_CTX_REG, %g1 + ld [%o0 + AOFF_mm_context], %o3 + lda [%g1] ASI_M_MMUREGS, %g5 +#ifndef CONFIG_SMP + cmp %o3, -1 + be hypersparc_flush_tlb_range_out +#endif + sethi %hi(~((1 << PGDIR_SHIFT) - 1)), %o4 + sta %o3, [%g1] ASI_M_MMUREGS + and %o1, %o4, %o1 + add %o1, 0x200, %o1 + sta %g0, [%o1] ASI_M_FLUSH_PROBE +1: + sub %o1, %o4, %o1 + cmp %o1, %o2 + blu,a 1b + sta %g0, [%o1] ASI_M_FLUSH_PROBE +hypersparc_flush_tlb_range_out: + retl + sta %g5, [%g1] ASI_M_MMUREGS + +hypersparc_flush_tlb_page: + ld [%o0 + VMA_VM_MM], %o0 + mov SRMMU_CTX_REG, %g1 + ld [%o0 + AOFF_mm_context], %o3 + andn %o1, (PAGE_SIZE - 1), %o1 +#ifndef CONFIG_SMP + cmp %o3, -1 + be hypersparc_flush_tlb_page_out +#endif + lda [%g1] ASI_M_MMUREGS, %g5 + sta %o3, [%g1] ASI_M_MMUREGS + sta %g0, [%o1] ASI_M_FLUSH_PROBE +hypersparc_flush_tlb_page_out: + retl + sta %g5, [%g1] ASI_M_MMUREGS + + __INIT + + /* High speed page clear/copy. */ +hypersparc_bzero_1page: +/* NOTE: This routine has to be shorter than 40insns --jj */ + clr %g1 + mov 32, %g2 + mov 64, %g3 + mov 96, %g4 + mov 128, %g5 + mov 160, %g7 + mov 192, %o2 + mov 224, %o3 + mov 16, %o1 +1: + stda %g0, [%o0 + %g0] ASI_M_BFILL + stda %g0, [%o0 + %g2] ASI_M_BFILL + stda %g0, [%o0 + %g3] ASI_M_BFILL + stda %g0, [%o0 + %g4] ASI_M_BFILL + stda %g0, [%o0 + %g5] ASI_M_BFILL + stda %g0, [%o0 + %g7] ASI_M_BFILL + stda %g0, [%o0 + %o2] ASI_M_BFILL + stda %g0, [%o0 + %o3] ASI_M_BFILL + subcc %o1, 1, %o1 + bne 1b + add %o0, 256, %o0 + + retl + nop + +hypersparc_copy_1page: +/* NOTE: This routine has to be shorter than 70insns --jj */ + sub %o1, %o0, %o2 ! difference + mov 16, %g1 +1: + sta %o0, [%o0 + %o2] ASI_M_BCOPY + add %o0, 32, %o0 + sta %o0, [%o0 + %o2] ASI_M_BCOPY + add %o0, 32, %o0 + sta %o0, [%o0 + %o2] ASI_M_BCOPY + add %o0, 32, %o0 + sta %o0, [%o0 + %o2] ASI_M_BCOPY + add %o0, 32, %o0 + sta %o0, [%o0 + %o2] ASI_M_BCOPY + add %o0, 32, %o0 + sta %o0, [%o0 + %o2] ASI_M_BCOPY + add %o0, 32, %o0 + sta %o0, [%o0 + %o2] ASI_M_BCOPY + add %o0, 32, %o0 + sta %o0, [%o0 + %o2] ASI_M_BCOPY + subcc %g1, 1, %g1 + bne 1b + add %o0, 32, %o0 + + retl + nop + + .globl hypersparc_setup_blockops +hypersparc_setup_blockops: + sethi %hi(bzero_1page), %o0 + or %o0, %lo(bzero_1page), %o0 + sethi %hi(hypersparc_bzero_1page), %o1 + or %o1, %lo(hypersparc_bzero_1page), %o1 + sethi %hi(hypersparc_copy_1page), %o2 + or %o2, %lo(hypersparc_copy_1page), %o2 + ld [%o1], %o4 +1: + add %o1, 4, %o1 + st %o4, [%o0] + add %o0, 4, %o0 + cmp %o1, %o2 + bne 1b + ld [%o1], %o4 + sethi %hi(__copy_1page), %o0 + or %o0, %lo(__copy_1page), %o0 + sethi %hi(hypersparc_setup_blockops), %o2 + or %o2, %lo(hypersparc_setup_blockops), %o2 + ld [%o1], %o4 +1: + add %o1, 4, %o1 + st %o4, [%o0] + add %o0, 4, %o0 + cmp %o1, %o2 + bne 1b + ld [%o1], %o4 + sta %g0, [%g0] ASI_M_FLUSH_IWHOLE + retl + nop |