X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=arch%2Fsparc64%2Fkernel%2Fhead.S;fp=arch%2Fsparc64%2Fkernel%2Fhead.S;h=31c5892f5acc49669c4326bf5162a42ebceb36ea;hb=43bc926fffd92024b46cafaf7350d669ba9ca884;hp=8104a56ca2d8e779b7999cf14d62db03155f8129;hpb=cee37fe97739d85991964371c1f3a745c00dd236;p=linux-2.6.git diff --git a/arch/sparc64/kernel/head.S b/arch/sparc64/kernel/head.S index 8104a56ca..31c5892f5 100644 --- a/arch/sparc64/kernel/head.S +++ b/arch/sparc64/kernel/head.S @@ -10,6 +10,7 @@ #include #include #include +#include #include #include #include @@ -26,21 +27,17 @@ #include #include #include +#include /* This section from from _start to sparc64_boot_end should fit into - * 0x0000.0000.0040.4000 to 0x0000.0000.0040.8000 and will be sharing space - * with bootup_user_stack, which is from 0x0000.0000.0040.4000 to - * 0x0000.0000.0040.6000 and empty_bad_page, which is from - * 0x0000.0000.0040.6000 to 0x0000.0000.0040.8000. + * 0x0000000000404000 to 0x0000000000408000. */ - .text .globl start, _start, stext, _stext _start: start: _stext: stext: -bootup_user_stack: ! 0x0000000000404000 b sparc64_boot flushw /* Flush register file. */ @@ -80,15 +77,247 @@ sparc_ramdisk_image64: .xword 0 .word _end - /* We must be careful, 32-bit OpenBOOT will get confused if it - * tries to save away a register window to a 64-bit kernel - * stack address. Flush all windows, disable interrupts, - * remap if necessary, jump onto kernel trap table, then kernel - * stack, or else we die. + /* PROM cif handler code address is in %o4. */ +sparc64_boot: +1: rd %pc, %g7 + set 1b, %g1 + cmp %g1, %g7 + be,pn %xcc, sparc64_boot_after_remap + mov %o4, %l7 + + /* We need to remap the kernel. Use position independant + * code to remap us to KERNBASE. * - * PROM entry point is on %o4 + * SILO can invoke us with 32-bit address masking enabled, + * so make sure that's clear. */ -sparc64_boot: + rdpr %pstate, %g1 + andn %g1, PSTATE_AM, %g1 + wrpr %g1, 0x0, %pstate + ba,a,pt %xcc, 1f + + .globl prom_finddev_name, prom_chosen_path, prom_root_node + .globl prom_getprop_name, prom_mmu_name, prom_peer_name + .globl prom_callmethod_name, prom_translate_name, prom_root_compatible + .globl prom_map_name, prom_unmap_name, prom_mmu_ihandle_cache + .globl prom_boot_mapped_pc, prom_boot_mapping_mode + .globl prom_boot_mapping_phys_high, prom_boot_mapping_phys_low + .globl is_sun4v +prom_peer_name: + .asciz "peer" +prom_compatible_name: + .asciz "compatible" +prom_finddev_name: + .asciz "finddevice" +prom_chosen_path: + .asciz "/chosen" +prom_getprop_name: + .asciz "getprop" +prom_mmu_name: + .asciz "mmu" +prom_callmethod_name: + .asciz "call-method" +prom_translate_name: + .asciz "translate" +prom_map_name: + .asciz "map" +prom_unmap_name: + .asciz "unmap" +prom_sun4v_name: + .asciz "sun4v" + .align 4 +prom_root_compatible: + .skip 64 +prom_root_node: + .word 0 +prom_mmu_ihandle_cache: + .word 0 +prom_boot_mapped_pc: + .word 0 +prom_boot_mapping_mode: + .word 0 + .align 8 +prom_boot_mapping_phys_high: + .xword 0 +prom_boot_mapping_phys_low: + .xword 0 +is_sun4v: + .word 0 +1: + rd %pc, %l0 + + mov (1b - prom_peer_name), %l1 + sub %l0, %l1, %l1 + mov 0, %l2 + + /* prom_root_node = prom_peer(0) */ + stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "peer" + mov 1, %l3 + stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 1 + stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1 + stx %l2, [%sp + 2047 + 128 + 0x18] ! arg1, 0 + stx %g0, [%sp + 2047 + 128 + 0x20] ! ret1 + call %l7 + add %sp, (2047 + 128), %o0 ! argument array + + ldx [%sp + 2047 + 128 + 0x20], %l4 ! prom root node + mov (1b - prom_root_node), %l1 + sub %l0, %l1, %l1 + stw %l4, [%l1] + + mov (1b - prom_getprop_name), %l1 + mov (1b - prom_compatible_name), %l2 + mov (1b - prom_root_compatible), %l5 + sub %l0, %l1, %l1 + sub %l0, %l2, %l2 + sub %l0, %l5, %l5 + + /* prom_getproperty(prom_root_node, "compatible", + * &prom_root_compatible, 64) + */ + stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "getprop" + mov 4, %l3 + stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 4 + mov 1, %l3 + stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1 + stx %l4, [%sp + 2047 + 128 + 0x18] ! arg1, prom_root_node + stx %l2, [%sp + 2047 + 128 + 0x20] ! arg2, "compatible" + stx %l5, [%sp + 2047 + 128 + 0x28] ! arg3, &prom_root_compatible + mov 64, %l3 + stx %l3, [%sp + 2047 + 128 + 0x30] ! arg4, size + stx %g0, [%sp + 2047 + 128 + 0x38] ! ret1 + call %l7 + add %sp, (2047 + 128), %o0 ! argument array + + mov (1b - prom_finddev_name), %l1 + mov (1b - prom_chosen_path), %l2 + mov (1b - prom_boot_mapped_pc), %l3 + sub %l0, %l1, %l1 + sub %l0, %l2, %l2 + sub %l0, %l3, %l3 + stw %l0, [%l3] + sub %sp, (192 + 128), %sp + + /* chosen_node = prom_finddevice("/chosen") */ + stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "finddevice" + mov 1, %l3 + stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 1 + stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1 + stx %l2, [%sp + 2047 + 128 + 0x18] ! arg1, "/chosen" + stx %g0, [%sp + 2047 + 128 + 0x20] ! ret1 + call %l7 + add %sp, (2047 + 128), %o0 ! argument array + + ldx [%sp + 2047 + 128 + 0x20], %l4 ! chosen device node + + mov (1b - prom_getprop_name), %l1 + mov (1b - prom_mmu_name), %l2 + mov (1b - prom_mmu_ihandle_cache), %l5 + sub %l0, %l1, %l1 + sub %l0, %l2, %l2 + sub %l0, %l5, %l5 + + /* prom_mmu_ihandle_cache = prom_getint(chosen_node, "mmu") */ + stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "getprop" + mov 4, %l3 + stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 4 + mov 1, %l3 + stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1 + stx %l4, [%sp + 2047 + 128 + 0x18] ! arg1, chosen_node + stx %l2, [%sp + 2047 + 128 + 0x20] ! arg2, "mmu" + stx %l5, [%sp + 2047 + 128 + 0x28] ! arg3, &prom_mmu_ihandle_cache + mov 4, %l3 + stx %l3, [%sp + 2047 + 128 + 0x30] ! arg4, sizeof(arg3) + stx %g0, [%sp + 2047 + 128 + 0x38] ! ret1 + call %l7 + add %sp, (2047 + 128), %o0 ! argument array + + mov (1b - prom_callmethod_name), %l1 + mov (1b - prom_translate_name), %l2 + sub %l0, %l1, %l1 + sub %l0, %l2, %l2 + lduw [%l5], %l5 ! prom_mmu_ihandle_cache + + stx %l1, [%sp + 2047 + 128 + 0x00] ! service, "call-method" + mov 3, %l3 + stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 3 + mov 5, %l3 + stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 5 + stx %l2, [%sp + 2047 + 128 + 0x18] ! arg1: "translate" + stx %l5, [%sp + 2047 + 128 + 0x20] ! arg2: prom_mmu_ihandle_cache + /* PAGE align */ + srlx %l0, 13, %l3 + sllx %l3, 13, %l3 + stx %l3, [%sp + 2047 + 128 + 0x28] ! arg3: vaddr, our PC + stx %g0, [%sp + 2047 + 128 + 0x30] ! res1 + stx %g0, [%sp + 2047 + 128 + 0x38] ! res2 + stx %g0, [%sp + 2047 + 128 + 0x40] ! res3 + stx %g0, [%sp + 2047 + 128 + 0x48] ! res4 + stx %g0, [%sp + 2047 + 128 + 0x50] ! res5 + call %l7 + add %sp, (2047 + 128), %o0 ! argument array + + ldx [%sp + 2047 + 128 + 0x40], %l1 ! translation mode + mov (1b - prom_boot_mapping_mode), %l4 + sub %l0, %l4, %l4 + stw %l1, [%l4] + mov (1b - prom_boot_mapping_phys_high), %l4 + sub %l0, %l4, %l4 + ldx [%sp + 2047 + 128 + 0x48], %l2 ! physaddr high + stx %l2, [%l4 + 0x0] + ldx [%sp + 2047 + 128 + 0x50], %l3 ! physaddr low + /* 4MB align */ + srlx %l3, 22, %l3 + sllx %l3, 22, %l3 + stx %l3, [%l4 + 0x8] + + /* Leave service as-is, "call-method" */ + mov 7, %l3 + stx %l3, [%sp + 2047 + 128 + 0x08] ! num_args, 7 + mov 1, %l3 + stx %l3, [%sp + 2047 + 128 + 0x10] ! num_rets, 1 + mov (1b - prom_map_name), %l3 + sub %l0, %l3, %l3 + stx %l3, [%sp + 2047 + 128 + 0x18] ! arg1: "map" + /* Leave arg2 as-is, prom_mmu_ihandle_cache */ + mov -1, %l3 + stx %l3, [%sp + 2047 + 128 + 0x28] ! arg3: mode (-1 default) + sethi %hi(8 * 1024 * 1024), %l3 + stx %l3, [%sp + 2047 + 128 + 0x30] ! arg4: size (8MB) + sethi %hi(KERNBASE), %l3 + stx %l3, [%sp + 2047 + 128 + 0x38] ! arg5: vaddr (KERNBASE) + stx %g0, [%sp + 2047 + 128 + 0x40] ! arg6: empty + mov (1b - prom_boot_mapping_phys_low), %l3 + sub %l0, %l3, %l3 + ldx [%l3], %l3 + stx %l3, [%sp + 2047 + 128 + 0x48] ! arg7: phys addr + call %l7 + add %sp, (2047 + 128), %o0 ! argument array + + add %sp, (192 + 128), %sp + +sparc64_boot_after_remap: + sethi %hi(prom_root_compatible), %g1 + or %g1, %lo(prom_root_compatible), %g1 + sethi %hi(prom_sun4v_name), %g7 + or %g7, %lo(prom_sun4v_name), %g7 + mov 5, %g3 +1: ldub [%g7], %g2 + ldub [%g1], %g4 + cmp %g2, %g4 + bne,pn %icc, 2f + add %g7, 1, %g7 + subcc %g3, 1, %g3 + bne,pt %xcc, 1b + add %g1, 1, %g1 + + sethi %hi(is_sun4v), %g1 + or %g1, %lo(is_sun4v), %g1 + mov 1, %g7 + stw %g7, [%g1] + +2: + BRANCH_IF_SUN4V(g1, jump_to_sun4u_init) BRANCH_IF_CHEETAH_BASE(g1,g7,cheetah_boot) BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,cheetah_plus_boot) ba,pt %xcc, spitfire_boot @@ -125,185 +354,7 @@ cheetah_generic_boot: stxa %g0, [%g3] ASI_IMMU membar #Sync - wrpr %g0, (PSTATE_PRIV|PSTATE_PEF|PSTATE_IE), %pstate - wr %g0, 0, %fprs - - /* Just like for Spitfire, we probe itlb-2 for a mapping which - * matches our current %pc. We take the physical address in - * that mapping and use it to make our own. - */ - - /* %g5 holds the tlb data */ - sethi %uhi(_PAGE_VALID | _PAGE_SZ4MB), %g5 - sllx %g5, 32, %g5 - or %g5, (_PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_L | _PAGE_W | _PAGE_G), %g5 - - /* Put PADDR tlb data mask into %g3. */ - sethi %uhi(_PAGE_PADDR), %g3 - or %g3, %ulo(_PAGE_PADDR), %g3 - sllx %g3, 32, %g3 - sethi %hi(_PAGE_PADDR), %g7 - or %g7, %lo(_PAGE_PADDR), %g7 - or %g3, %g7, %g3 - - set 2 << 16, %l0 /* TLB entry walker. */ - set 0x1fff, %l2 /* Page mask. */ - rd %pc, %l3 - andn %l3, %l2, %g2 /* vaddr comparator */ - -1: ldxa [%l0] ASI_ITLB_TAG_READ, %g1 - membar #Sync - andn %g1, %l2, %g1 - cmp %g1, %g2 - be,pn %xcc, cheetah_got_tlbentry - nop - and %l0, (127 << 3), %g1 - cmp %g1, (127 << 3) - blu,pt %xcc, 1b - add %l0, (1 << 3), %l0 - - /* Search the small TLB. OBP never maps us like that but - * newer SILO can. - */ - clr %l0 - -1: ldxa [%l0] ASI_ITLB_TAG_READ, %g1 - membar #Sync - andn %g1, %l2, %g1 - cmp %g1, %g2 - be,pn %xcc, cheetah_got_tlbentry - nop - cmp %l0, (15 << 3) - blu,pt %xcc, 1b - add %l0, (1 << 3), %l0 - - /* BUG() if we get here... */ - ta 0x5 - -cheetah_got_tlbentry: - ldxa [%l0] ASI_ITLB_DATA_ACCESS, %g0 - ldxa [%l0] ASI_ITLB_DATA_ACCESS, %g1 - membar #Sync - and %g1, %g3, %g1 - set 0x5fff, %l0 - andn %g1, %l0, %g1 - or %g5, %g1, %g5 - - /* Clear out any KERNBASE area entries. */ - set 2 << 16, %l0 - sethi %hi(KERNBASE), %g3 - sethi %hi(KERNBASE<<1), %g7 - mov TLB_TAG_ACCESS, %l7 - - /* First, check ITLB */ -1: ldxa [%l0] ASI_ITLB_TAG_READ, %g1 - membar #Sync - andn %g1, %l2, %g1 - cmp %g1, %g3 - blu,pn %xcc, 2f - cmp %g1, %g7 - bgeu,pn %xcc, 2f - nop - stxa %g0, [%l7] ASI_IMMU - membar #Sync - stxa %g0, [%l0] ASI_ITLB_DATA_ACCESS - membar #Sync - -2: and %l0, (127 << 3), %g1 - cmp %g1, (127 << 3) - blu,pt %xcc, 1b - add %l0, (1 << 3), %l0 - - /* Next, check DTLB */ - set 2 << 16, %l0 -1: ldxa [%l0] ASI_DTLB_TAG_READ, %g1 - membar #Sync - andn %g1, %l2, %g1 - cmp %g1, %g3 - blu,pn %xcc, 2f - cmp %g1, %g7 - bgeu,pn %xcc, 2f - nop - stxa %g0, [%l7] ASI_DMMU - membar #Sync - stxa %g0, [%l0] ASI_DTLB_DATA_ACCESS - membar #Sync - -2: and %l0, (511 << 3), %g1 - cmp %g1, (511 << 3) - blu,pt %xcc, 1b - add %l0, (1 << 3), %l0 - - /* On Cheetah+, have to check second DTLB. */ - BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,l0,2f) - ba,pt %xcc, 9f - nop - -2: set 3 << 16, %l0 -1: ldxa [%l0] ASI_DTLB_TAG_READ, %g1 - membar #Sync - andn %g1, %l2, %g1 - cmp %g1, %g3 - blu,pn %xcc, 2f - cmp %g1, %g7 - bgeu,pn %xcc, 2f - nop - stxa %g0, [%l7] ASI_DMMU - membar #Sync - stxa %g0, [%l0] ASI_DTLB_DATA_ACCESS - membar #Sync - -2: and %l0, (511 << 3), %g1 - cmp %g1, (511 << 3) - blu,pt %xcc, 1b - add %l0, (1 << 3), %l0 - -9: - - /* Now lock the TTE we created into ITLB-0 and DTLB-0, - * entry 15 (and maybe 14 too). - */ - sethi %hi(KERNBASE), %g3 - set (0 << 16) | (15 << 3), %g7 - stxa %g3, [%l7] ASI_DMMU - membar #Sync - stxa %g5, [%g7] ASI_DTLB_DATA_ACCESS - membar #Sync - stxa %g3, [%l7] ASI_IMMU - membar #Sync - stxa %g5, [%g7] ASI_ITLB_DATA_ACCESS - membar #Sync - flush %g3 - membar #Sync - sethi %hi(_end), %g3 /* Check for bigkernel case */ - or %g3, %lo(_end), %g3 - srl %g3, 23, %g3 /* Check if _end > 8M */ - brz,pt %g3, 1f - sethi %hi(KERNBASE), %g3 /* Restore for fixup code below */ - sethi %hi(0x400000), %g3 - or %g3, %lo(0x400000), %g3 - add %g5, %g3, %g5 /* New tte data */ - andn %g5, (_PAGE_G), %g5 - sethi %hi(KERNBASE+0x400000), %g3 - or %g3, %lo(KERNBASE+0x400000), %g3 - set (0 << 16) | (14 << 3), %g7 - stxa %g3, [%l7] ASI_DMMU - membar #Sync - stxa %g5, [%g7] ASI_DTLB_DATA_ACCESS - membar #Sync - stxa %g3, [%l7] ASI_IMMU - membar #Sync - stxa %g5, [%g7] ASI_ITLB_DATA_ACCESS - membar #Sync - flush %g3 - membar #Sync - sethi %hi(KERNBASE), %g3 /* Restore for fixup code below */ - ba,pt %xcc, 1f - nop - -1: set sun4u_init, %g2 - jmpl %g2 + %g0, %g0 - nop + ba,a,pt %xcc, jump_to_sun4u_init spitfire_boot: /* Typically PROM has already enabled both MMU's and both on-chip @@ -313,6 +364,7 @@ spitfire_boot: stxa %g1, [%g0] ASI_LSU_CONTROL membar #Sync +jump_to_sun4u_init: /* * Make sure we are in privileged mode, have address masking, * using the ordinary globals and have enabled floating @@ -324,197 +376,64 @@ spitfire_boot: wrpr %g0, (PSTATE_PRIV|PSTATE_PEF|PSTATE_IE), %pstate wr %g0, 0, %fprs -spitfire_create_mappings: - /* %g5 holds the tlb data */ - sethi %uhi(_PAGE_VALID | _PAGE_SZ4MB), %g5 - sllx %g5, 32, %g5 - or %g5, (_PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_L | _PAGE_W | _PAGE_G), %g5 - - /* Base of physical memory cannot reliably be assumed to be - * at 0x0! Figure out where it happens to be. -DaveM - */ - - /* Put PADDR tlb data mask into %g3. */ - sethi %uhi(_PAGE_PADDR_SF), %g3 - or %g3, %ulo(_PAGE_PADDR_SF), %g3 - sllx %g3, 32, %g3 - sethi %hi(_PAGE_PADDR_SF), %g7 - or %g7, %lo(_PAGE_PADDR_SF), %g7 - or %g3, %g7, %g3 + set sun4u_init, %g2 + jmpl %g2 + %g0, %g0 + nop - /* Walk through entire ITLB, looking for entry which maps - * our %pc currently, stick PADDR from there into %g5 tlb data. - */ - clr %l0 /* TLB entry walker. */ - set 0x1fff, %l2 /* Page mask. */ - rd %pc, %l3 - andn %l3, %l2, %g2 /* vaddr comparator */ -1: - /* Yes, the nops seem to be necessary for now, don't ask me why. -DaveM */ - ldxa [%l0] ASI_ITLB_TAG_READ, %g1 - nop - nop - nop - andn %g1, %l2, %g1 /* Get vaddr */ - cmp %g1, %g2 - be,a,pn %xcc, spitfire_got_tlbentry - ldxa [%l0] ASI_ITLB_DATA_ACCESS, %g1 - cmp %l0, (63 << 3) - blu,pt %xcc, 1b - add %l0, (1 << 3), %l0 - - /* BUG() if we get here... */ - ta 0x5 - -spitfire_got_tlbentry: - /* Nops here again, perhaps Cheetah/Blackbird are better behaved... */ - nop - nop - nop - and %g1, %g3, %g1 /* Mask to just get paddr bits. */ - set 0x5fff, %l3 /* Mask offset to get phys base. */ - andn %g1, %l3, %g1 - - /* NOTE: We hold on to %g1 paddr base as we need it below to lock - * NOTE: the PROM cif code into the TLB. - */ +sun4u_init: + BRANCH_IF_SUN4V(g1, sun4v_init) - or %g5, %g1, %g5 /* Or it into TAG being built. */ + /* Set ctx 0 */ + mov PRIMARY_CONTEXT, %g7 + stxa %g0, [%g7] ASI_DMMU + membar #Sync - clr %l0 /* TLB entry walker. */ - sethi %hi(KERNBASE), %g3 /* 4M lower limit */ - sethi %hi(KERNBASE<<1), %g7 /* 8M upper limit */ - mov TLB_TAG_ACCESS, %l7 -1: - /* Yes, the nops seem to be necessary for now, don't ask me why. -DaveM */ - ldxa [%l0] ASI_ITLB_TAG_READ, %g1 - nop - nop - nop - andn %g1, %l2, %g1 /* Get vaddr */ - cmp %g1, %g3 - blu,pn %xcc, 2f - cmp %g1, %g7 - bgeu,pn %xcc, 2f - nop - stxa %g0, [%l7] ASI_IMMU - stxa %g0, [%l0] ASI_ITLB_DATA_ACCESS + mov SECONDARY_CONTEXT, %g7 + stxa %g0, [%g7] ASI_DMMU membar #Sync -2: - cmp %l0, (63 << 3) - blu,pt %xcc, 1b - add %l0, (1 << 3), %l0 - nop; nop; nop + ba,pt %xcc, sun4u_continue + nop - clr %l0 /* TLB entry walker. */ -1: - /* Yes, the nops seem to be necessary for now, don't ask me why. -DaveM */ - ldxa [%l0] ASI_DTLB_TAG_READ, %g1 - nop - nop - nop - andn %g1, %l2, %g1 /* Get vaddr */ - cmp %g1, %g3 - blu,pn %xcc, 2f - cmp %g1, %g7 - bgeu,pn %xcc, 2f +sun4v_init: + /* Set ctx 0 */ + mov PRIMARY_CONTEXT, %g7 + stxa %g0, [%g7] ASI_MMU + membar #Sync + + mov SECONDARY_CONTEXT, %g7 + stxa %g0, [%g7] ASI_MMU + membar #Sync + ba,pt %xcc, niagara_tlb_fixup nop - stxa %g0, [%l7] ASI_DMMU - stxa %g0, [%l0] ASI_DTLB_DATA_ACCESS - membar #Sync -2: - cmp %l0, (63 << 3) - blu,pt %xcc, 1b - add %l0, (1 << 3), %l0 - nop; nop; nop +sun4u_continue: + BRANCH_IF_ANY_CHEETAH(g1, g7, cheetah_tlb_fixup) + ba,pt %xcc, spitfire_tlb_fixup + nop - /* PROM never puts any TLB entries into the MMU with the lock bit - * set. So we gladly use tlb entry 63 for KERNBASE. And maybe 62 too. - */ +niagara_tlb_fixup: + mov 3, %g2 /* Set TLB type to hypervisor. */ + sethi %hi(tlb_type), %g1 + stw %g2, [%g1 + %lo(tlb_type)] - sethi %hi(KERNBASE), %g3 - mov (63 << 3), %g7 - stxa %g3, [%l7] ASI_DMMU /* KERNBASE into TLB TAG */ - stxa %g5, [%g7] ASI_DTLB_DATA_ACCESS /* TTE into TLB DATA */ - membar #Sync - stxa %g3, [%l7] ASI_IMMU /* KERNBASE into TLB TAG */ - stxa %g5, [%g7] ASI_ITLB_DATA_ACCESS /* TTE into TLB DATA */ - membar #Sync - flush %g3 - membar #Sync - sethi %hi(_end), %g3 /* Check for bigkernel case */ - or %g3, %lo(_end), %g3 - srl %g3, 23, %g3 /* Check if _end > 8M */ - brz,pt %g3, 2f - sethi %hi(KERNBASE), %g3 /* Restore for fixup code below */ - sethi %hi(0x400000), %g3 - or %g3, %lo(0x400000), %g3 - add %g5, %g3, %g5 /* New tte data */ - andn %g5, (_PAGE_G), %g5 - sethi %hi(KERNBASE+0x400000), %g3 - or %g3, %lo(KERNBASE+0x400000), %g3 - mov (62 << 3), %g7 - stxa %g3, [%l7] ASI_DMMU - stxa %g5, [%g7] ASI_DTLB_DATA_ACCESS - membar #Sync - stxa %g3, [%l7] ASI_IMMU - stxa %g5, [%g7] ASI_ITLB_DATA_ACCESS - membar #Sync - flush %g3 - membar #Sync - sethi %hi(KERNBASE), %g3 /* Restore for fixup code below */ -2: ba,pt %xcc, 1f + /* Patch copy/clear ops. */ + call niagara_patch_copyops nop -1: - set sun4u_init, %g2 - jmpl %g2 + %g0, %g0 + call niagara_patch_bzero + nop + call niagara_patch_pageops nop -sun4u_init: - /* Set ctx 0 */ - mov PRIMARY_CONTEXT, %g7 - stxa %g0, [%g7] ASI_DMMU - membar #Sync - - mov SECONDARY_CONTEXT, %g7 - stxa %g0, [%g7] ASI_DMMU - membar #Sync - - /* We are now safely (we hope) in Nucleus context (0), rewrite - * the KERNBASE TTE's so they no longer have the global bit set. - * Don't forget to setup TAG_ACCESS first 8-) - */ - mov TLB_TAG_ACCESS, %g2 - stxa %g3, [%g2] ASI_IMMU - stxa %g3, [%g2] ASI_DMMU - membar #Sync - - BRANCH_IF_ANY_CHEETAH(g1,g7,cheetah_tlb_fixup) + /* Patch TLB/cache ops. */ + call hypervisor_patch_cachetlbops + nop - ba,pt %xcc, spitfire_tlb_fixup + ba,pt %xcc, tlb_fixup_done nop cheetah_tlb_fixup: - set (0 << 16) | (15 << 3), %g7 - ldxa [%g7] ASI_ITLB_DATA_ACCESS, %g0 - ldxa [%g7] ASI_ITLB_DATA_ACCESS, %g1 - andn %g1, (_PAGE_G), %g1 - stxa %g1, [%g7] ASI_ITLB_DATA_ACCESS - membar #Sync - - ldxa [%g7] ASI_DTLB_DATA_ACCESS, %g0 - ldxa [%g7] ASI_DTLB_DATA_ACCESS, %g1 - andn %g1, (_PAGE_G), %g1 - stxa %g1, [%g7] ASI_DTLB_DATA_ACCESS - membar #Sync - - /* Kill instruction prefetch queues. */ - flush %g3 - membar #Sync - mov 2, %g2 /* Set TLB type to cheetah+. */ BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,1f) @@ -523,26 +442,11 @@ cheetah_tlb_fixup: 1: sethi %hi(tlb_type), %g1 stw %g2, [%g1 + %lo(tlb_type)] - BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,1f) - ba,pt %xcc, 2f - nop - -1: /* Patch context register writes to support nucleus page - * size correctly. - */ - call cheetah_plus_patch_etrap - nop - call cheetah_plus_patch_rtrap - nop - call cheetah_plus_patch_fpdis - nop - call cheetah_plus_patch_winfixup - nop - - -2: /* Patch copy/page operations to cheetah optimized versions. */ + /* Patch copy/page operations to cheetah optimized versions. */ call cheetah_patch_copyops nop + call cheetah_patch_copy_page + nop call cheetah_patch_cachetlbops nop @@ -550,21 +454,6 @@ cheetah_tlb_fixup: nop spitfire_tlb_fixup: - mov (63 << 3), %g7 - ldxa [%g7] ASI_ITLB_DATA_ACCESS, %g1 - andn %g1, (_PAGE_G), %g1 - stxa %g1, [%g7] ASI_ITLB_DATA_ACCESS - membar #Sync - - ldxa [%g7] ASI_DTLB_DATA_ACCESS, %g1 - andn %g1, (_PAGE_G), %g1 - stxa %g1, [%g7] ASI_DTLB_DATA_ACCESS - membar #Sync - - /* Kill instruction prefetch queues. */ - flush %g3 - membar #Sync - /* Set TLB type to spitfire. */ mov 0, %g2 sethi %hi(tlb_type), %g1 @@ -577,24 +466,6 @@ tlb_fixup_done: mov %sp, %l6 mov %o4, %l7 -#if 0 /* We don't do it like this anymore, but for historical hack value - * I leave this snippet here to show how crazy we can be sometimes. 8-) - */ - - /* Setup "Linux Current Register", thanks Sun 8-) */ - wr %g0, 0x1, %pcr - - /* Blackbird errata workaround. See commentary in - * smp.c:smp_percpu_timer_interrupt() for more - * information. - */ - ba,pt %xcc, 99f - nop - .align 64 -99: wr %g6, %g0, %pic - rd %pic, %g0 -#endif - wr %g0, ASI_P, %asi mov 1, %g1 sllx %g1, THREAD_SHIFT, %g1 @@ -623,74 +494,119 @@ tlb_fixup_done: call prom_init mov %l7, %o0 ! OpenPROM cif handler + /* Initialize current_thread_info()->cpu as early as possible. + * In order to do that accurately we have to patch up the get_cpuid() + * assembler sequences. And that, in turn, requires that we know + * if we are on a Starfire box or not. While we're here, patch up + * the sun4v sequences as well. + */ + call check_if_starfire + nop + call per_cpu_patch + nop + call sun4v_patch + nop + +#ifdef CONFIG_SMP + call hard_smp_processor_id + nop + cmp %o0, NR_CPUS + blu,pt %xcc, 1f + nop + call boot_cpu_id_too_large + nop + /* Not reached... */ + +1: +#else + mov 0, %o0 +#endif + stb %o0, [%g6 + TI_CPU] + /* Off we go.... */ call start_kernel nop /* Not reached... */ -/* IMPORTANT NOTE: Whenever making changes here, check - * trampoline.S as well. -jj */ - .globl setup_tba -setup_tba: /* i0 = is_starfire */ - save %sp, -160, %sp - - rdpr %tba, %g7 - sethi %hi(prom_tba), %o1 - or %o1, %lo(prom_tba), %o1 - stx %g7, [%o1] + /* This is meant to allow the sharing of this code between + * boot processor invocation (via setup_tba() below) and + * secondary processor startup (via trampoline.S). The + * former does use this code, the latter does not yet due + * to some complexities. That should be fixed up at some + * point. + * + * There used to be enormous complexity wrt. transferring + * over from the firwmare's trap table to the Linux kernel's. + * For example, there was a chicken & egg problem wrt. building + * the OBP page tables, yet needing to be on the Linux kernel + * trap table (to translate PAGE_OFFSET addresses) in order to + * do that. + * + * We now handle OBP tlb misses differently, via linear lookups + * into the prom_trans[] array. So that specific problem no + * longer exists. Yet, unfortunately there are still some issues + * preventing trampoline.S from using this code... ho hum. + */ + .globl setup_trap_table +setup_trap_table: + save %sp, -192, %sp - /* Setup "Linux" globals 8-) */ + /* Force interrupts to be disabled. */ rdpr %pstate, %o1 - mov %g6, %o2 - wrpr %o1, (PSTATE_AG|PSTATE_IE), %pstate - sethi %hi(sparc64_ttable_tl0), %g1 - wrpr %g1, %tba - mov %o2, %g6 + andn %o1, PSTATE_IE, %o1 + wrpr %o1, 0x0, %pstate + wrpr %g0, 15, %pil + + /* Make the firmware call to jump over to the Linux trap table. */ + sethi %hi(is_sun4v), %o0 + lduw [%o0 + %lo(is_sun4v)], %o0 + brz,pt %o0, 1f + nop - /* Set up MMU globals */ - wrpr %o1, (PSTATE_MG|PSTATE_IE), %pstate + TRAP_LOAD_TRAP_BLOCK(%g2, %g3) + add %g2, TRAP_PER_CPU_FAULT_INFO, %g2 + stxa %g2, [%g0] ASI_SCRATCHPAD - /* Set fixed globals used by dTLB miss handler. */ -#define KERN_HIGHBITS ((_PAGE_VALID|_PAGE_SZ4MB)^0xfffff80000000000) -#define KERN_LOWBITS (_PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_W) + /* Compute physical address: + * + * paddr = kern_base + (mmfsa_vaddr - KERNBASE) + */ + sethi %hi(KERNBASE), %g3 + sub %g2, %g3, %g2 + sethi %hi(kern_base), %g3 + ldx [%g3 + %lo(kern_base)], %g3 + add %g2, %g3, %o1 - mov TSB_REG, %g1 - stxa %g0, [%g1] ASI_DMMU - membar #Sync - stxa %g0, [%g1] ASI_IMMU - membar #Sync - mov TLB_SFSR, %g1 - sethi %uhi(KERN_HIGHBITS), %g2 - or %g2, %ulo(KERN_HIGHBITS), %g2 - sllx %g2, 32, %g2 - or %g2, KERN_LOWBITS, %g2 - - BRANCH_IF_ANY_CHEETAH(g3,g7,cheetah_vpte_base) - ba,pt %xcc, spitfire_vpte_base + call prom_set_trap_table_sun4v + sethi %hi(sparc64_ttable_tl0), %o0 + + ba,pt %xcc, 2f nop -cheetah_vpte_base: - sethi %uhi(VPTE_BASE_CHEETAH), %g3 - or %g3, %ulo(VPTE_BASE_CHEETAH), %g3 - ba,pt %xcc, 2f - sllx %g3, 32, %g3 +1: call prom_set_trap_table + sethi %hi(sparc64_ttable_tl0), %o0 -spitfire_vpte_base: - sethi %uhi(VPTE_BASE_SPITFIRE), %g3 - or %g3, %ulo(VPTE_BASE_SPITFIRE), %g3 - sllx %g3, 32, %g3 + /* Start using proper page size encodings in ctx register. */ +2: sethi %hi(sparc64_kern_pri_context), %g3 + ldx [%g3 + %lo(sparc64_kern_pri_context)], %g2 -2: - clr %g7 -#undef KERN_HIGHBITS -#undef KERN_LOWBITS + mov PRIMARY_CONTEXT, %g1 + +661: stxa %g2, [%g1] ASI_DMMU + .section .sun4v_1insn_patch, "ax" + .word 661b + stxa %g2, [%g1] ASI_MMU + .previous + + membar #Sync /* Kill PROM timer */ sethi %hi(0x80000000), %o2 sllx %o2, 32, %o2 wr %o2, 0, %tick_cmpr - BRANCH_IF_ANY_CHEETAH(o2,o3,1f) + BRANCH_IF_SUN4V(o2, 1f) + BRANCH_IF_ANY_CHEETAH(o2, o3, 1f) ba,pt %xcc, 2f nop @@ -701,73 +617,69 @@ spitfire_vpte_base: sllx %o2, 32, %o2 wr %o2, %asr25 - /* Ok, we're done setting up all the state our trap mechanims needs, - * now get back into normal globals and let the PROM know what is up. - */ 2: wrpr %g0, %g0, %wstate - wrpr %o1, PSTATE_IE, %pstate call init_irqwork_curcpu nop - call prom_set_trap_table - sethi %hi(sparc64_ttable_tl0), %o0 - - BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g2,g3,1f) - ba,pt %xcc, 2f - nop - -1: /* Start using proper page size encodings in ctx register. */ - sethi %uhi(CTX_CHEETAH_PLUS_NUC), %g3 - mov PRIMARY_CONTEXT, %g1 - sllx %g3, 32, %g3 - sethi %hi(CTX_CHEETAH_PLUS_CTX0), %g2 - or %g3, %g2, %g3 - stxa %g3, [%g1] ASI_DMMU - membar #Sync - -2: + /* Now we can turn interrupts back on. */ rdpr %pstate, %o1 or %o1, PSTATE_IE, %o1 wrpr %o1, 0, %pstate + wrpr %g0, 0x0, %pil ret restore -/* - * The following skips make sure the trap table in ttable.S is aligned - * on a 32K boundary as required by the v9 specs for TBA register. - */ -sparc64_boot_end: - .skip 0x2000 + _start - sparc64_boot_end -bootup_user_stack_end: - .skip 0x2000 - -#ifdef CONFIG_SBUS -/* This is just a hack to fool make depend config.h discovering - strategy: As the .S files below need config.h, but - make depend does not find it for them, we include config.h - in head.S */ -#endif + .globl setup_tba +setup_tba: + save %sp, -192, %sp -! 0x0000000000408000 + /* The boot processor is the only cpu which invokes this + * routine, the other cpus set things up via trampoline.S. + * So save the OBP trap table address here. + */ + rdpr %tba, %g7 + sethi %hi(prom_tba), %o1 + or %o1, %lo(prom_tba), %o1 + stx %g7, [%o1] -#include "ttable.S" -#include "systbls.S" + call setup_trap_table + nop - .align 1024 - .globl swapper_pg_dir -swapper_pg_dir: - .word 0 + ret + restore +sparc64_boot_end: +#include "ktlb.S" +#include "tsb.S" #include "etrap.S" #include "rtrap.S" #include "winfixup.S" #include "entry.S" +#include "sun4v_tlb_miss.S" +#include "sun4v_ivec.S" - /* This is just anal retentiveness on my part... */ - .align 16384 +/* + * The following skip makes sure the trap table in ttable.S is aligned + * on a 32K boundary as required by the v9 specs for TBA register. + * + * We align to a 32K boundary, then we have the 32K kernel TSB, + * then the 32K aligned trap table. + */ +1: + .skip 0x4000 + _start - 1b + + .globl swapper_tsb +swapper_tsb: + .skip (32 * 1024) + +! 0x0000000000408000 + +#include "ttable.S" + +#include "systbls.S" .data .align 8 @@ -775,8 +687,11 @@ swapper_pg_dir: prom_tba: .xword 0 tlb_type: .word 0 /* Must NOT end up in BSS */ .section ".fixup",#alloc,#execinstr - .globl __ret_efault + + .globl __ret_efault, __retl_efault __ret_efault: ret restore %g0, -EFAULT, %o0 - +__retl_efault: + retl + mov -EFAULT, %o0