/* $Id: ultra.S,v 1.36 1999/12/15 15:45:18 davem Exp $ * ultra.S: Don't expand these all over the place... * * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu) */ #include #include #include /* This file is meant to be read efficiently by the CPU, not humans. * Staraj sie tego nikomu nie pierdolnac... */ .text .align 32 .globl __flush_tlb_page, __flush_tlb_mm, __flush_tlb_range __flush_tlb_page: /* %o0=(ctx & 0x3ff), %o1=page&PAGE_MASK, %o2=SECONDARY_CONTEXT */ /*IC1*/ ldxa [%o2] ASI_DMMU, %g2 cmp %g2, %o0 bne,pn %icc, __flush_tlb_page_slow or %o1, 0x10, %g3 stxa %g0, [%g3] ASI_DMMU_DEMAP stxa %g0, [%g3] ASI_IMMU_DEMAP retl flush %g6 __flush_tlb_mm: /* %o0=(ctx & 0x3ff), %o1=SECONDARY_CONTEXT */ /*IC2*/ ldxa [%o1] ASI_DMMU, %g2 cmp %g2, %o0 bne,pn %icc, __flush_tlb_mm_slow mov 0x50, %g3 stxa %g0, [%g3] ASI_DMMU_DEMAP stxa %g0, [%g3] ASI_IMMU_DEMAP retl flush %g6 __flush_tlb_range: /* %o0=(ctx&0x3ff), %o1=start&PAGE_MASK, %o2=SECONDARY_CONTEXT, * %o3=end&PAGE_MASK, %o4=PAGE_SIZE, %o5=(end - start) */ #define TLB_MAGIC 206 /* Students, do you know how I calculated this? -DaveM */ /*IC3*/ cmp %o5, %o4 be,pt %xcc, __flush_tlb_page srlx %o5, 13, %g5 cmp %g5, TLB_MAGIC bgeu,pn %icc, __flush_tlb_range_constant_time or %o1, 0x10, %g5 ldxa [%o2] ASI_DMMU, %g2 cmp %g2, %o0 __flush_tlb_range_page_by_page: /*IC4*/ bne,pn %icc, __flush_tlb_range_pbp_slow sub %o5, %o4, %o5 1: stxa %g0, [%g5 + %o5] ASI_DMMU_DEMAP stxa %g0, [%g5 + %o5] ASI_IMMU_DEMAP brnz,pt %o5, 1b sub %o5, %o4, %o5 retl flush %g6 __flush_tlb_range_constant_time: /* %o0=ctx, %o1=start, %o3=end */ /*IC5*/ rdpr %pstate, %g1 wrpr %g1, PSTATE_IE, %pstate mov TLB_TAG_ACCESS, %g3 mov (62 << 3), %g2 1: ldxa [%g2] ASI_ITLB_TAG_READ, %o4 and %o4, 0x3ff, %o5 cmp %o5, %o0 bne,pt %icc, 2f /*IC6*/ andn %o4, 0x3ff, %o4 cmp %o4, %o1 blu,pt %xcc, 2f cmp %o4, %o3 blu,pn %xcc, 4f 2: ldxa [%g2] ASI_DTLB_TAG_READ, %o4 and %o4, 0x3ff, %o5 cmp %o5, %o0 /*IC7*/ andn %o4, 0x3ff, %o4 bne,pt %icc, 3f cmp %o4, %o1 blu,pt %xcc, 3f cmp %o4, %o3 blu,pn %xcc, 5f nop 3: brnz,pt %g2, 1b /*IC8*/ sub %g2, (1 << 3), %g2 retl wrpr %g1, 0x0, %pstate 4: stxa %g0, [%g3] ASI_IMMU stxa %g0, [%g2] ASI_ITLB_DATA_ACCESS ba,pt %xcc, 2b flush %g6 5: stxa %g0, [%g3] ASI_DMMU /*IC9*/ stxa %g0, [%g2] ASI_DTLB_DATA_ACCESS ba,pt %xcc, 3b flush %g6 .align 32 __flush_tlb_mm_slow: /*IC10*/rdpr %pstate, %g1 wrpr %g1, PSTATE_IE, %pstate stxa %o0, [%o1] ASI_DMMU stxa %g0, [%g3] ASI_DMMU_DEMAP stxa %g0, [%g3] ASI_IMMU_DEMAP flush %g6 stxa %g2, [%o1] ASI_DMMU flush %g6 /*IC11*/retl wrpr %g1, 0, %pstate .align 32 __flush_tlb_page_slow: /*IC12*/rdpr %pstate, %g1 wrpr %g1, PSTATE_IE, %pstate stxa %o0, [%o2] ASI_DMMU stxa %g0, [%g3] ASI_DMMU_DEMAP stxa %g0, [%g3] ASI_IMMU_DEMAP flush %g6 stxa %g2, [%o2] ASI_DMMU flush %g6 /*IC13*/retl wrpr %g1, 0, %pstate .align 32 __flush_tlb_range_pbp_slow: /*IC13*/rdpr %pstate, %g1 wrpr %g1, PSTATE_IE, %pstate stxa %o0, [%o2] ASI_DMMU 2: stxa %g0, [%g5 + %o5] ASI_DMMU_DEMAP stxa %g0, [%g5 + %o5] ASI_IMMU_DEMAP brnz,pt %o5, 2b sub %o5, %o4, %o5 flush %g6 /*IC14*/stxa %g2, [%o2] ASI_DMMU flush %g6 retl wrpr %g1, 0x0, %pstate .align 32 .globl flush_icache_page flush_icache_page: /* %o0 = phys_page */ sethi %hi(1 << 13), %o2 ! IC_set bit mov 1, %g1 srlx %o0, 5, %o0 clr %o1 ! IC_addr sllx %g1, 36, %g1 sub %g1, 1, %g2 or %o0, %g1, %o0 ! VALID+phys-addr comparitor sllx %g2, 1, %g2 andn %g2, 0xfe, %g2 ! IC_tag mask 1: ldda [%o1] ASI_IC_TAG, %o4 and %o5, %g2, %o5 cmp %o5, %o0 be,pn %xcc, iflush1 add %o1, 0x20, %g3 2: ldda [%o1 + %o2] ASI_IC_TAG, %o4 and %o5, %g2, %o5 cmp %o5, %o0 be,pn %xcc, iflush2 nop 3: cmp %g3, %o2 bne,pt %xcc, 1b mov %g3, %o1 retl nop iflush1:stxa %g0, [%o1] ASI_IC_TAG flush %g6 ba,a,pt %xcc, 2b iflush2:stxa %g0, [%o1 + %o2] ASI_IC_TAG flush %g6 ba,a,pt %xcc, 3b #ifdef __SMP__ /* These are all called by the slaves of a cross call, at * trap level 1, with interrupts fully disabled. * * Register usage: * %g5 mm->context (all tlb flushes) * %g1 address arg 1 (tlb page and range flushes) * %g7 address arg 2 (tlb range flush only) * * %g6 ivector table, don't touch * %g2 scratch 1 * %g3 scratch 2 * %g4 scratch 3 * * TODO: Make xcall TLB range flushes use the tricks above... -DaveM */ .align 32 .globl xcall_flush_tlb_page, xcall_flush_tlb_mm, xcall_flush_tlb_range xcall_flush_tlb_page: mov SECONDARY_CONTEXT, %g2 or %g1, 0x10, %g4 ldxa [%g2] ASI_DMMU, %g3 stxa %g5, [%g2] ASI_DMMU stxa %g0, [%g4] ASI_DMMU_DEMAP stxa %g0, [%g4] ASI_IMMU_DEMAP stxa %g3, [%g2] ASI_DMMU retry xcall_flush_tlb_mm: mov SECONDARY_CONTEXT, %g2 mov 0x50, %g4 ldxa [%g2] ASI_DMMU, %g3 stxa %g5, [%g2] ASI_DMMU stxa %g0, [%g4] ASI_DMMU_DEMAP stxa %g0, [%g4] ASI_IMMU_DEMAP stxa %g3, [%g2] ASI_DMMU retry xcall_flush_tlb_range: sethi %hi(8192 - 1), %g2 or %g2, %lo(8192 - 1), %g2 andn %g1, %g2, %g1 andn %g7, %g2, %g7 sub %g7, %g1, %g3 add %g2, 1, %g2 orcc %g1, 0x10, %g1 srlx %g3, 13, %g4 cmp %g4, 96 bgu,pn %icc, xcall_flush_tlb_mm mov SECONDARY_CONTEXT, %g4 ldxa [%g4] ASI_DMMU, %g7 sub %g3, %g2, %g3 stxa %g5, [%g4] ASI_DMMU nop nop 1: stxa %g0, [%g1 + %g3] ASI_DMMU_DEMAP stxa %g0, [%g1 + %g3] ASI_IMMU_DEMAP brnz,pt %g3, 1b sub %g3, %g2, %g3 stxa %g7, [%g4] ASI_DMMU retry nop nop .globl xcall_report_regs xcall_report_regs: rdpr %pstate, %g2 wrpr %g2, PSTATE_IG | PSTATE_AG, %pstate rdpr %pil, %g2 wrpr %g0, 15, %pil sethi %hi(109f), %g7 b,pt %xcc, etrap_irq 109: or %g7, %lo(109b), %g7 call __show_regs add %sp, STACK_BIAS + REGWIN_SZ, %o0 b,pt %xcc, rtrap clr %l6 .globl xcall_capture xcall_capture: rdpr %pstate, %g2 wrpr %g2, PSTATE_IG | PSTATE_AG, %pstate rdpr %pil, %g2 wrpr %g0, 15, %pil sethi %hi(109f), %g7 b,pt %xcc, etrap_irq 109: or %g7, %lo(109b), %g7 call smp_penguin_jailcell nop b,pt %xcc, rtrap clr %l6 .globl xcall_promstop xcall_promstop: rdpr %pstate, %g2 wrpr %g2, PSTATE_IG | PSTATE_AG, %pstate rdpr %pil, %g2 wrpr %g0, 15, %pil sethi %hi(109f), %g7 b,pt %xcc, etrap_irq 109: or %g7, %lo(109b), %g7 flushw call prom_stopself nop /* We should not return, just spin if we do... */ 1: b,a,pt %xcc, 1b nop .globl xcall_receive_signal xcall_receive_signal: rdpr %pstate, %g2 wrpr %g2, PSTATE_IG | PSTATE_AG, %pstate rdpr %tstate, %g1 andcc %g1, TSTATE_PRIV, %g0 /* If we did not trap from user space, just ignore. */ bne,pn %xcc, 99f sethi %hi(109f), %g7 b,pt %xcc, etrap 109: or %g7, %lo(109b), %g7 b,pt %xcc, rtrap clr %l6 99: retry /* These two are not performance critical... */ .globl xcall_flush_tlb_all xcall_flush_tlb_all: clr %g2 clr %g3 1: ldxa [%g3] ASI_DTLB_DATA_ACCESS, %g4 and %g4, _PAGE_L, %g5 brnz,pn %g5, 2f mov TLB_TAG_ACCESS, %g7 stxa %g0, [%g7] ASI_DMMU membar #Sync stxa %g0, [%g3] ASI_DTLB_DATA_ACCESS membar #Sync 2: ldxa [%g3] ASI_ITLB_DATA_ACCESS, %g4 and %g4, _PAGE_L, %g5 brnz,pn %g5, 2f mov TLB_TAG_ACCESS, %g7 stxa %g0, [%g7] ASI_IMMU membar #Sync stxa %g0, [%g3] ASI_ITLB_DATA_ACCESS 2: add %g2, 1, %g2 cmp %g2, 63 ble,pt %icc, 1b sll %g2, 3, %g3 flush %g6 retry .globl xcall_flush_cache_all xcall_flush_cache_all: sethi %hi(16383), %g2 or %g2, %lo(16383), %g2 clr %g3 1: stxa %g0, [%g3] ASI_IC_TAG add %g3, 32, %g3 cmp %g3, %g2 bleu,pt %xcc, 1b nop flush %g6 retry #endif /* __SMP__ */