/* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License, Version 1.0 only * (the "License"). You may not use this file except in compliance * with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2005 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ /* * VM - Hardware Address Translation management. * * This file describes the contents of the sun-reference-mmu(sfmmu)- * specific hat data structures and the sfmmu-specific hat procedures. * The machine-independent interface is described in . */ #ifndef _VM_HAT_SFMMU_H #define _VM_HAT_SFMMU_H #pragma ident "%Z%%M% %I% %E% SMI" #ifdef __cplusplus extern "C" { #endif #ifndef _ASM #include #endif /* _ASM */ #ifdef _KERNEL #include #include #include /* * Don't alter these without considering changes to ism_map_t. */ #define DEFAULT_ISM_PAGESIZE MMU_PAGESIZE4M #define ISM_PG_SIZE(ism_vbshift) (1 << ism_vbshift) #define ISM_SZ_MASK(ism_vbshift) (ISM_PG_SIZE(ism_vbshift) - 1) #define ISM_MAP_SLOTS 8 /* Change this carefully. */ #ifndef _ASM #include #include #include #include #include #include #include #include typedef struct hat sfmmu_t; /* * SFMMU attributes for hat_memload/hat_devload */ #define SFMMU_UNCACHEPTTE 0x01000000 /* unencache in physical $ */ #define SFMMU_UNCACHEVTTE 0x02000000 /* unencache in virtual $ */ #define SFMMU_SIDEFFECT 0x04000000 /* set side effect bit */ #define SFMMU_LOAD_ALLATTR (HAT_PROT_MASK | HAT_ORDER_MASK | \ HAT_ENDIAN_MASK | HAT_NOFAULT | HAT_NOSYNC | \ SFMMU_UNCACHEPTTE | SFMMU_UNCACHEVTTE | SFMMU_SIDEFFECT) /* * sfmmu flags for hat_memload/hat_devload */ #define SFMMU_NO_TSBLOAD 0x08000000 /* do not preload tsb */ #define SFMMU_LOAD_ALLFLAG (HAT_LOAD | HAT_LOAD_LOCK | \ HAT_LOAD_ADV | HAT_LOAD_CONTIG | HAT_LOAD_NOCONSIST | \ HAT_LOAD_SHARE | HAT_LOAD_REMAP | SFMMU_NO_TSBLOAD | \ HAT_RELOAD_SHARE | HAT_NO_KALLOC | HAT_LOAD_TEXT) /* * sfmmu internal flag to hat_pageunload that spares locked mappings */ #define SFMMU_KERNEL_RELOC 0x8000 /* * mode for sfmmu_chgattr */ #define SFMMU_SETATTR 0x0 #define SFMMU_CLRATTR 0x1 #define SFMMU_CHGATTR 0x2 /* * sfmmu specific flags for page_t */ #define P_PNC 0x8 /* non-caching is permanent bit */ #define P_TNC 0x10 /* non-caching is temporary bit */ #define P_KPMS 0x20 /* kpm mapped small (vac alias prevention) */ #define P_KPMC 0x40 /* kpm conflict page (vac alias prevention) */ #define PP_GENERIC_ATTR(pp) ((pp)->p_nrm & (P_MOD | P_REF | P_RO)) #define PP_ISMOD(pp) ((pp)->p_nrm & P_MOD) #define PP_ISREF(pp) ((pp)->p_nrm & P_REF) #define PP_ISRO(pp) ((pp)->p_nrm & P_RO) #define PP_ISNC(pp) ((pp)->p_nrm & (P_PNC|P_TNC)) #define PP_ISPNC(pp) ((pp)->p_nrm & P_PNC) #define PP_ISTNC(pp) ((pp)->p_nrm & P_TNC) #define PP_ISKPMS(pp) ((pp)->p_nrm & P_KPMS) #define PP_ISKPMC(pp) ((pp)->p_nrm & P_KPMC) #define PP_SETMOD(pp) ((pp)->p_nrm |= P_MOD) #define PP_SETREF(pp) ((pp)->p_nrm |= P_REF) #define PP_SETREFMOD(pp) ((pp)->p_nrm |= (P_REF|P_MOD)) #define PP_SETRO(pp) ((pp)->p_nrm |= P_RO) #define PP_SETREFRO(pp) ((pp)->p_nrm |= (P_REF|P_RO)) #define PP_SETPNC(pp) ((pp)->p_nrm |= P_PNC) #define PP_SETTNC(pp) ((pp)->p_nrm |= P_TNC) #define PP_SETKPMS(pp) ((pp)->p_nrm |= P_KPMS) #define PP_SETKPMC(pp) ((pp)->p_nrm |= P_KPMC) #define PP_CLRMOD(pp) ((pp)->p_nrm &= ~P_MOD) #define PP_CLRREF(pp) ((pp)->p_nrm &= ~P_REF) #define PP_CLRREFMOD(pp) ((pp)->p_nrm &= ~(P_REF|P_MOD)) #define PP_CLRRO(pp) ((pp)->p_nrm &= ~P_RO) #define PP_CLRPNC(pp) ((pp)->p_nrm &= ~P_PNC) #define PP_CLRTNC(pp) ((pp)->p_nrm &= ~P_TNC) #define PP_CLRKPMS(pp) ((pp)->p_nrm &= ~P_KPMS) #define PP_CLRKPMC(pp) ((pp)->p_nrm &= ~P_KPMC) /* * All shared memory segments attached with the SHM_SHARE_MMU flag (ISM) * will be constrained to a 4M, 32M or 256M alignment. Also since every newly- * created ISM segment is created out of a new address space at base va * of 0 we don't need to store it. */ #define ISM_ALIGN(shift) (1 << shift) /* base va aligned to M */ #define ISM_ALIGNED(shift, va) (((uintptr_t)va & (ISM_ALIGN(shift) - 1)) == 0) #define ISM_SHIFT(shift, x) ((uintptr_t)x >> (shift)) /* * Pad locks out to cache sub-block boundaries to prevent * false sharing, so several processes don't contend for * the same line if they aren't using the same lock. Since * this is a typedef we also have a bit of freedom in * changing lock implementations later if we decide it * is necessary. */ typedef struct hat_lock { kmutex_t hl_mutex; uchar_t hl_pad[64 - sizeof (kmutex_t)]; } hatlock_t; #define HATLOCK_MUTEXP(hatlockp) (&((hatlockp)->hl_mutex)) /* * All segments mapped with ISM are guaranteed to be 4M, 32M or 256M aligned. * Also size is guaranteed to be in 4M, 32M or 256M chunks. * ism_seg consists of the following members: * [XX..22] base address of ism segment. XX is 63 or 31 depending whether * caddr_t is 64 bits or 32 bits. * [21..0] size of segment. * * NOTE: Don't alter this structure without changing defines above and * the tsb_miss and protection handlers. */ typedef struct ism_map { uintptr_t imap_seg; /* base va + sz of ISM segment */ ushort_t imap_vb_shift; /* mmu_pageshift for ism page size */ ushort_t imap_hatflags; /* primary ism page size */ uint_t imap_sz_mask; /* mmu_pagemask for ism page size */ sfmmu_t *imap_ismhat; /* hat id of dummy ISM as */ struct ism_ment *imap_ment; /* pointer to mapping list entry */ } ism_map_t; #define ism_start(map) ((caddr_t)((map).imap_seg & \ ~ISM_SZ_MASK((map).imap_vb_shift))) #define ism_size(map) ((map).imap_seg & ISM_SZ_MASK((map).imap_vb_shift)) #define ism_end(map) ((caddr_t)(ism_start(map) + (ism_size(map) * \ ISM_PG_SIZE((map).imap_vb_shift)))) /* * ISM mapping entry. Used to link all hat's sharing a ism_hat. * Same function as the p_mapping list for a page. */ typedef struct ism_ment { sfmmu_t *iment_hat; /* back pointer to hat_share() hat */ caddr_t iment_base_va; /* hat's va base for this ism seg */ struct ism_ment *iment_next; /* next ism map entry */ struct ism_ment *iment_prev; /* prev ism map entry */ } ism_ment_t; /* * ISM segment block. One will be hung off the sfmmu structure if a * a process uses ISM. More will be linked using ismblk_next if more * than ISM_MAP_SLOTS segments are attached to this proc. * * All modifications to fields in this structure will be protected * by the hat mutex. In order to avoid grabbing this lock in low level * routines (tsb miss/protection handlers and vatopfn) while not * introducing any race conditions with hat_unshare, we will set * CTX_ISM_BUSY bit in the ctx struct. Any mmu traps that occur * for this ctx while this bit is set will be handled in sfmmu_tsb_excption * where it will synchronize behind the hat mutex. */ typedef struct ism_blk { ism_map_t iblk_maps[ISM_MAP_SLOTS]; struct ism_blk *iblk_next; uint64_t iblk_nextpa; } ism_blk_t; /* * TSB access information. All fields are protected by the process's * hat lock. */ struct tsb_info { caddr_t tsb_va; /* tsb base virtual address */ uint64_t tsb_pa; /* tsb base physical address */ struct tsb_info *tsb_next; /* next tsb used by this process */ uint16_t tsb_szc; /* tsb size code */ uint16_t tsb_flags; /* flags for this tsb; see below */ uint_t tsb_ttesz_mask; /* page size masks; see below */ tte_t tsb_tte; /* tte to lock into DTLB */ sfmmu_t *tsb_sfmmu; /* sfmmu */ kmem_cache_t *tsb_cache; /* cache from which mem allocated */ vmem_t *tsb_vmp; /* vmem arena from which mem alloc'd */ }; /* * Values for "tsb_ttesz_mask" bitmask. */ #define TSB8K (1 << TTE8K) #define TSB64K (1 << TTE64K) #define TSB512K (1 << TTE512K) #define TSB4M (1 << TTE4M) #define TSB32M (1 << TTE32M) #define TSB256M (1 << TTE256M) /* * Values for "tsb_flags" field. */ #define TSB_RELOC_FLAG 0x1 #define TSB_FLUSH_NEEDED 0x2 #define TSB_SWAPPED 0x4 /* * The platform dependent hat structure. * tte counts should be protected by cas. * cpuset is protected by cas. * * Note that sfmmu_xhat_provider MUST be the first element. */ struct hat { void *sfmmu_xhat_provider; /* NULL for CPU hat */ cpuset_t sfmmu_cpusran; /* cpu bit mask for efficient xcalls */ struct as *sfmmu_as; /* as this hat provides mapping for */ ulong_t sfmmu_ttecnt[MMU_PAGE_SIZES]; /* per sz tte counts */ ulong_t sfmmu_ismttecnt[MMU_PAGE_SIZES]; /* est. ism ttes */ union _h_un { ism_blk_t *sfmmu_iblkp; /* maps to ismhat(s) */ ism_ment_t *sfmmu_imentp; /* ism hat's mapping list */ } h_un; uint_t sfmmu_free:1; /* hat to be freed - set on as_free */ uint_t sfmmu_ismhat:1; /* hat is dummy ism hatid */ uint_t sfmmu_ctxflushed:1; /* ctx has been flushed */ uchar_t sfmmu_rmstat; /* refmod stats refcnt */ uchar_t sfmmu_clrstart; /* start color bin for page coloring */ ushort_t sfmmu_clrbin; /* per as phys page coloring bin */ short sfmmu_cnum; /* context number */ ushort_t sfmmu_flags; /* flags */ struct tsb_info *sfmmu_tsb; /* list of per as tsbs */ uint64_t sfmmu_ismblkpa; /* pa of sfmmu_iblkp, or -1 */ kcondvar_t sfmmu_tsb_cv; /* signals TSB swapin or relocation */ uchar_t sfmmu_cext; /* context page size encoding */ uint8_t sfmmu_pgsz[MMU_PAGE_SIZES]; /* ranking for MMU */ #ifdef sun4v struct hv_tsb_block sfmmu_hvblock; #endif }; #define sfmmu_iblk h_un.sfmmu_iblkp #define sfmmu_iment h_un.sfmmu_imentp /* * bit mask for managing vac conflicts on large pages. * bit 1 is for uncache flag. * bits 2 through min(num of cache colors + 1,31) are * for cache colors that have already been flushed. */ #define CACHE_UNCACHE 1 #define CACHE_NUM_COLOR (shm_alignment >> MMU_PAGESHIFT) #define CACHE_VCOLOR_MASK(vcolor) (2 << (vcolor & (CACHE_NUM_COLOR - 1))) #define CacheColor_IsFlushed(flag, vcolor) \ ((flag) & CACHE_VCOLOR_MASK(vcolor)) #define CacheColor_SetFlushed(flag, vcolor) \ ((flag) |= CACHE_VCOLOR_MASK(vcolor)) /* * Flags passed to sfmmu_page_cache to flush page from vac or not. */ #define CACHE_FLUSH 0 #define CACHE_NO_FLUSH 1 /* * Flags passed to sfmmu_tlbcache_demap */ #define FLUSH_NECESSARY_CPUS 0 #define FLUSH_ALL_CPUS 1 /* * Software context structure. The size of this structure is currently * hardwired into the tsb miss handlers in assembly code through the * CTX_SZ_SHIFT define. Since this define is used in a shift we should keep * this structure a power of two. * * ctx_flags: * Bit 0 : Free flag. */ struct ctx { union _ctx_un { sfmmu_t *ctx_sfmmup; /* back pointer to hat id */ struct ctx *ctx_freep; /* next ctx in freelist */ } ctx_un; krwlock_t ctx_rwlock; /* protect context from stealer */ uint32_t ctx_flags; /* flags */ uint8_t pad[12]; }; #define ctx_sfmmu ctx_un.ctx_sfmmup #define ctx_free ctx_un.ctx_freep #ifdef DEBUG /* * For debugging purpose only. Maybe removed later. */ struct ctx_trace { sfmmu_t *sc_sfmmu_stolen; sfmmu_t *sc_sfmmu_stealing; clock_t sc_time; ushort_t sc_type; ushort_t sc_cnum; }; #define CTX_TRC_STEAL 0x1 #define CTX_TRC_FREE 0x0 #define TRSIZE 0x400 #define NEXT_CTXTR(ptr) (((ptr) >= ctx_trace_last) ? \ ctx_trace_first : ((ptr) + 1)) #define TRACE_CTXS(mutex, ptr, cnum, stolen_sfmmu, stealing_sfmmu, type) \ mutex_enter(mutex); \ (ptr)->sc_sfmmu_stolen = (stolen_sfmmu); \ (ptr)->sc_sfmmu_stealing = (stealing_sfmmu); \ (ptr)->sc_cnum = (cnum); \ (ptr)->sc_type = (type); \ (ptr)->sc_time = lbolt; \ (ptr) = NEXT_CTXTR(ptr); \ num_ctx_stolen += (type); \ mutex_exit(mutex); #else #define TRACE_CTXS(mutex, ptr, cnum, stolen_sfmmu, stealing_sfmmu, type) #endif /* DEBUG */ #endif /* !_ASM */ /* * Macros for sfmmup->sfmmu_flags access. The macros that change the flags * ASSERT() that we're holding the HAT lock before changing the flags; * however callers that read the flags may do so without acquiring the lock * in a fast path, and then recheck the flag after acquiring the lock in * a slow path. */ #define SFMMU_FLAGS_ISSET(sfmmup, flags) \ (((sfmmup)->sfmmu_flags & (flags)) == (flags)) #define SFMMU_FLAGS_CLEAR(sfmmup, flags) \ (ASSERT(sfmmu_hat_lock_held((sfmmup))), \ (sfmmup)->sfmmu_flags &= ~(flags)) #define SFMMU_FLAGS_SET(sfmmup, flags) \ (ASSERT(sfmmu_hat_lock_held((sfmmup))), \ (sfmmup)->sfmmu_flags |= (flags)) /* * sfmmu HAT flags */ #define HAT_64K_FLAG 0x01 #define HAT_512K_FLAG 0x02 #define HAT_4M_FLAG 0x04 #define HAT_32M_FLAG 0x08 #define HAT_256M_FLAG 0x10 #define HAT_4MTEXT_FLAG 0x80 #define HAT_SWAPPED 0x100 /* swapped out */ #define HAT_SWAPIN 0x200 /* swapping in */ #define HAT_BUSY 0x400 /* replacing TSB(s) */ #define HAT_ISMBUSY 0x800 /* adding/removing/traversing ISM maps */ #define HAT_LGPG_FLAGS \ (HAT_64K_FLAG | HAT_512K_FLAG | HAT_4M_FLAG | \ HAT_32M_FLAG | HAT_256M_FLAG) #define HAT_FLAGS_MASK \ (HAT_LGPG_FLAGS | HAT_4MTEXT_FLAG | HAT_SWAPPED | \ HAT_SWAPIN | HAT_BUSY | HAT_ISMBUSY) /* * Context flags */ #define CTX_FREE_FLAG 0x1 #define CTX_FLAGS_MASK 0x1 #define CTX_SET_FLAGS(ctx, flag) \ { \ uint32_t old, new; \ \ do { \ new = old = (ctx)->ctx_flags; \ new &= CTX_FLAGS_MASK; \ new |= flag; \ new = cas32(&(ctx)->ctx_flags, old, new); \ } while (new != old); \ } #define CTX_CLEAR_FLAGS(ctx, flag) \ { \ uint32_t old, new; \ \ do { \ new = old = (ctx)->ctx_flags; \ new &= CTX_FLAGS_MASK & ~(flag); \ new = cas32(&(ctx)->ctx_flags, old, new); \ } while (new != old); \ } #define ctxtoctxnum(ctx) ((ushort_t)((ctx) - ctxs)) /* * Defines needed for ctx stealing. */ #define GET_CTX_RETRY_CNT 100 /* * Starting with context 0, the first NUM_LOCKED_CTXS contexts * are locked so that sfmmu_getctx can't steal any of these * contexts. At the time this software was being developed, the * only context that needs to be locked is context 0 (the kernel * context), and context 1 (reserved for stolen context). So this constant * was originally defined to be 2. */ #define NUM_LOCKED_CTXS 2 #define INVALID_CONTEXT 1 #ifndef _ASM /* * Kernel page relocation stuff. */ struct sfmmu_callback { int key; int (*prehandler)(caddr_t, uint_t, uint_t, void *); int (*posthandler)(caddr_t, uint_t, uint_t, void *, pfn_t); int (*errhandler)(caddr_t, uint_t, uint_t, void *); int capture_cpus; }; extern int sfmmu_max_cb_id; extern struct sfmmu_callback *sfmmu_cb_table; extern int hat_kpr_enabled; struct pa_hment; /* * RFE: With multihat gone we gain back an int. We could use this to * keep ref bits on a per cpu basis to eliminate xcalls. */ struct sf_hment { tte_t hme_tte; /* tte for this hment */ union { struct page *page; /* what page this maps */ struct pa_hment *data; /* pa_hment */ } sf_hment_un; struct sf_hment *hme_next; /* next hment */ struct sf_hment *hme_prev; /* prev hment */ }; struct pa_hment { caddr_t addr; /* va */ uint_t len; /* bytes */ ushort_t flags; /* internal flags */ ushort_t refcnt; /* reference count */ id_t cb_id; /* callback id, table index */ void *pvt; /* handler's private data */ struct sf_hment sfment; /* corresponding dummy sf_hment */ }; #define hme_page sf_hment_un.page #define hme_data sf_hment_un.data #define hme_size(sfhmep) ((int)(TTE_CSZ(&(sfhmep)->hme_tte))) #define PAHME_SZ (sizeof (struct pa_hment)) #define SFHME_SZ (sizeof (struct sf_hment)) #define IS_PAHME(hme) ((hme)->hme_tte.ll == 0) /* * hmeblk_tag structure * structure used to obtain a match on a hme_blk. Currently consists of * the address of the sfmmu struct (or hatid), the base page address of the * hme_blk, and the rehash count. The rehash count is actually only 2 bits * and has the following meaning: * 1 = 8k or 64k hash sequence. * 2 = 512k hash sequence. * 3 = 4M hash sequence. * We require this count because we don't want to get a false hit on a 512K or * 4M rehash with a base address corresponding to a 8k or 64k hmeblk. * Note: The ordering and size of the hmeblk_tag members are implictly known * by the tsb miss handlers written in assembly. Do not change this structure * without checking those routines. See HTAG_SFMMUPSZ define. */ typedef union { struct { uint64_t hblk_basepg: 51, /* hme_blk base pg # */ hblk_rehash: 13; /* rehash number */ sfmmu_t *sfmmup; } hblk_tag_un; uint64_t htag_tag[2]; } hmeblk_tag; #define htag_id hblk_tag_un.sfmmup #define htag_bspage hblk_tag_un.hblk_basepg #define htag_rehash hblk_tag_un.hblk_rehash #define HTAGS_EQ(tag1, tag2) (((tag1.htag_tag[0] ^ tag2.htag_tag[0]) | \ (tag1.htag_tag[1] ^ tag2.htag_tag[1])) == 0) #define HME_REHASH(sfmmup) \ ((sfmmup)->sfmmu_ttecnt[TTE512K] != 0 || \ (sfmmup)->sfmmu_ttecnt[TTE4M] != 0 || \ (sfmmup)->sfmmu_ttecnt[TTE32M] != 0 || \ (sfmmup)->sfmmu_ttecnt[TTE256M] != 0) #endif /* !_ASM */ #define NHMENTS 8 /* # of hments in an 8k hme_blk */ /* needs to be multiple of 2 */ #ifndef _ASM #ifdef HBLK_TRACE #define HBLK_LOCK 1 #define HBLK_UNLOCK 0 #define HBLK_STACK_DEPTH 6 #define HBLK_AUDIT_CACHE_SIZE 16 #define HBLK_LOCK_PATTERN 0xaaaaaaaa #define HBLK_UNLOCK_PATTERN 0xbbbbbbbb struct hblk_lockcnt_audit { int flag; /* lock or unlock */ kthread_id_t thread; int depth; pc_t stack[HBLK_STACK_DEPTH]; }; #endif /* HBLK_TRACE */ /* * Hment block structure. * The hme_blk is the node data structure which the hash structure * mantains. An hme_blk can have 2 different sizes depending on the * number of hments it implicitly contains. When dealing with 64K, 512K, * or 4M hments there is one hment per hme_blk. When dealing with * 8k hments we allocate an hme_blk plus an additional 7 hments to * give us a total of 8 (NHMENTS) hments that can be referenced through a * hme_blk. * * The hmeblk structure contains 2 tte reference counters used to determine if * it is ok to free up the hmeblk. Both counters have to be zero in order * to be able to free up hmeblk. They are protected by cas. * hblk_hmecnt is the number of hments present on pp mapping lists. * hblk_vcnt reflects number of valid ttes in hmeblk. * * The hmeblk now also has per tte lock cnts. This is required because * the counts can be high and there are not enough bits in the tte. When * physio is fixed to not lock the translations we should be able to move * the lock cnt back to the tte. See bug id 1198554. * * Note that xhat_hme_blk's layout follows this structure: hme_blk_misc * and sf_hment are at the same offsets in both structures. Whenever * hme_blk is changed, xhat_hme_blk may need to be updated as well. */ struct hme_blk_misc { ushort_t locked_cnt; /* HAT_LOAD_LOCK ref cnt */ uint_t notused:10; uint_t xhat_bit:1; /* set for an xhat hme_blk */ uint_t shadow_bit:1; /* set for a shadow hme_blk */ uint_t nucleus_bit:1; /* set for a nucleus hme_blk */ uint_t ttesize:3; /* contains ttesz of hmeblk */ }; struct hme_blk { uint64_t hblk_nextpa; /* physical address for hash list */ hmeblk_tag hblk_tag; /* tag used to obtain an hmeblk match */ struct hme_blk *hblk_next; /* on free list or on hash list */ /* protected by hash lock */ struct hme_blk *hblk_shadow; /* pts to shadow hblk */ /* protected by hash lock */ uint_t hblk_span; /* span of memory hmeblk maps */ struct hme_blk_misc hblk_misc; union { struct { ushort_t hblk_hmecount; /* hment on mlists counter */ ushort_t hblk_validcnt; /* valid tte reference count */ } hblk_counts; uint_t hblk_shadow_mask; } hblk_un; #ifdef HBLK_TRACE kmutex_t hblk_audit_lock; /* lock to protect index */ uint_t hblk_audit_index; /* index into audit_cache */ struct hblk_lockcnt_audit hblk_audit_cache[HBLK_AUDIT_CACHE_SIZE]; #endif /* HBLK_AUDIT */ struct sf_hment hblk_hme[1]; /* hment array */ }; #define hblk_lckcnt hblk_misc.locked_cnt #define hblk_xhat_bit hblk_misc.xhat_bit #define hblk_shw_bit hblk_misc.shadow_bit #define hblk_nuc_bit hblk_misc.nucleus_bit #define hblk_ttesz hblk_misc.ttesize #define hblk_hmecnt hblk_un.hblk_counts.hblk_hmecount #define hblk_vcnt hblk_un.hblk_counts.hblk_validcnt #define hblk_shw_mask hblk_un.hblk_shadow_mask #define MAX_HBLK_LCKCNT 0xFFFF #define HMEBLK_ALIGN 0x8 /* hmeblk has to be double aligned */ #ifdef HBLK_TRACE #define HBLK_STACK_TRACE(hmeblkp, lock) \ { \ int flag = lock; /* to pacify lint */ \ int audit_index; \ \ mutex_enter(&hmeblkp->hblk_audit_lock); \ audit_index = hmeblkp->hblk_audit_index; \ hmeblkp->hblk_audit_index = ((hmeblkp->hblk_audit_index + 1) & \ (HBLK_AUDIT_CACHE_SIZE - 1)); \ mutex_exit(&hmeblkp->hblk_audit_lock); \ \ if (flag) \ hmeblkp->hblk_audit_cache[audit_index].flag = \ HBLK_LOCK_PATTERN; \ else \ hmeblkp->hblk_audit_cache[audit_index].flag = \ HBLK_UNLOCK_PATTERN; \ \ hmeblkp->hblk_audit_cache[audit_index].thread = curthread; \ hmeblkp->hblk_audit_cache[audit_index].depth = \ getpcstack(hmeblkp->hblk_audit_cache[audit_index].stack, \ HBLK_STACK_DEPTH); \ } #else #define HBLK_STACK_TRACE(hmeblkp, lock) #endif /* HBLK_TRACE */ #define HMEHASH_FACTOR 16 /* used to calc # of buckets in hme hash */ /* * A maximum number of user hmeblks is defined in order to place an upper * limit on how much nucleus memory is required and to avoid overflowing the * tsbmiss uhashsz and khashsz data areas. The number below corresponds to * the number of buckets required, for an average hash chain length of 4 on * a 16TB machine. */ #define MAX_UHME_BUCKETS (0x1 << 30) #define MAX_KHME_BUCKETS (0x1 << 30) /* * The minimum number of kernel hash buckets. */ #define MIN_KHME_BUCKETS 0x800 /* * The number of hash buckets must be a power of 2. If the initial calculated * value is less than USER_BUCKETS_THRESHOLD we round up to the next greater * power of 2, otherwise we round down to avoid huge over allocations. */ #define USER_BUCKETS_THRESHOLD (1<<22) #define MAX_NUCUHME_BUCKETS 0x4000 #define MAX_NUCKHME_BUCKETS 0x2000 /* * There are 2 locks in the hmehash bucket. The hmehash_mutex is * a regular mutex used to make sure operations on a hash link are only * done by one thread. Any operation which comes into the hat with * a will grab the hmehash_mutex. Normally one would expect * the tsb miss handlers to grab the hash lock to make sure the hash list * is consistent while we traverse it. Unfortunately this can lead to * deadlocks or recursive mutex enters since it is possible for * someone holding the lock to take a tlb/tsb miss. * To solve this problem we have added the hmehash_listlock. This lock * is only grabbed by the tsb miss handlers, vatopfn, and while * adding/removing a hmeblk from the hash list. The code is written to * guarantee we won't take a tlb miss while holding this lock. */ struct hmehash_bucket { kmutex_t hmehash_mutex; uint64_t hmeh_nextpa; /* physical address for hash list */ struct hme_blk *hmeblkp; uint_t hmeh_listlock; }; #endif /* !_ASM */ /* * The tsb miss handlers written in assembly know that sfmmup * is a 64 bit ptr. * * The bspage and re-hash part is 64 bits, with the sfmmup being another 64 * bits. */ #define HTAG_SFMMUPSZ 0 /* Not really used for LP64 */ #define HTAG_REHASHSZ 13 /* * Assembly routines need to be able to get to ttesz */ #define HBLK_SZMASK 0x7 #ifndef _ASM /* * Returns the number of bytes that an hmeblk spans given its tte size */ #define get_hblk_span(hmeblkp) ((hmeblkp)->hblk_span) #define get_hblk_ttesz(hmeblkp) ((hmeblkp)->hblk_ttesz) #define get_hblk_cache(hmeblkp) (((hmeblkp)->hblk_ttesz == TTE8K) ? \ sfmmu8_cache : sfmmu1_cache) #define HMEBLK_SPAN(ttesz) \ ((ttesz == TTE8K)? (TTEBYTES(ttesz) * NHMENTS) : TTEBYTES(ttesz)) #define set_hblk_sz(hmeblkp, ttesz) \ (hmeblkp)->hblk_ttesz = (ttesz); \ (hmeblkp)->hblk_span = HMEBLK_SPAN(ttesz) #define get_hblk_base(hmeblkp) \ ((uintptr_t)(hmeblkp)->hblk_tag.htag_bspage << MMU_PAGESHIFT) #define get_hblk_endaddr(hmeblkp) \ ((caddr_t)(get_hblk_base(hmeblkp) + get_hblk_span(hmeblkp))) #define in_hblk_range(hmeblkp, vaddr) \ (((uintptr_t)(vaddr) >= get_hblk_base(hmeblkp)) && \ ((uintptr_t)(vaddr) < (get_hblk_base(hmeblkp) + \ get_hblk_span(hmeblkp)))) #define tte_to_vaddr(hmeblkp, tte) ((caddr_t)(get_hblk_base(hmeblkp) \ + (TTEBYTES(TTE_CSZ(&tte)) * (tte).tte_hmenum))) #define vaddr_to_vshift(hblktag, vaddr, shwsz) \ ((((uintptr_t)(vaddr) >> MMU_PAGESHIFT) - (hblktag.htag_bspage)) >>\ TTE_BSZS_SHIFT((shwsz) - 1)) #define HME8BLK_SZ (sizeof (struct hme_blk) + \ (NHMENTS - 1) * sizeof (struct sf_hment)) #define HME1BLK_SZ (sizeof (struct hme_blk)) #define H8TOH1 (MMU_PAGESIZE4M / MMU_PAGESIZE) #define H1MIN (2 + MAX_BIGKTSB_TTES) /* nucleus text+data, ktsb */ /* * Hme_blk hash structure * Active mappings are kept in a hash structure of hme_blks. The hash * function is based on (ctx, vaddr) The size of the hash table size is a * power of 2 such that the average hash chain lenth is HMENT_HASHAVELEN. * The hash actually consists of 2 separate hashes. One hash is for the user * address space and the other hash is for the kernel address space. * The number of buckets are calculated at boot time and stored in the global * variables "uhmehash_num" and "khmehash_num". By making the hash table size * a power of 2 we can use a simply & function to derive an index instead of * a divide. * * HME_HASH_FUNCTION(hatid, vaddr, shift) returns a pointer to a hme_hash * bucket. * An hme hash bucket contains a pointer to an hme_blk and the mutex that * protects the link list. * Spitfire supports 4 page sizes. 8k and 64K pages only need one hash. * 512K pages need 2 hashes and 4M pages need 3 hashes. * The 'shift' parameter controls how many bits the vaddr will be shifted in * the hash function. It is calculated in the HME_HASH_SHIFT(ttesz) function * and it varies depending on the page size as follows: * 8k pages: HBLK_RANGE_SHIFT * 64k pages: MMU_PAGESHIFT64K * 512K pages: MMU_PAGESHIFT512K * 4M pages: MMU_PAGESHIFT4M * An assembly version of the hash function exists in sfmmu_ktsb_miss(). All * changes should be reflected in both versions. This function and the TSB * miss handlers are the only places which know about the two hashes. * * HBLK_RANGE_SHIFT controls range of virtual addresses that will fall * into the same bucket for a particular process. It is currently set to * be equivalent to 64K range or one hme_blk. * * The hme_blks in the hash are protected by a per hash bucket mutex * known as SFMMU_HASH_LOCK. * You need to acquire this lock before traversing the hash bucket link * list, while adding/removing a hme_blk to the list, and while * modifying an hme_blk. A possible optimization is to replace these * mutexes by readers/writer lock but right now it is not clear whether * this is a win or not. * * The HME_HASH_TABLE_SEARCH will search the hash table for the * hme_blk that contains the hment that corresponds to the passed * ctx and vaddr. It assumed the SFMMU_HASH_LOCK is held. */ #endif /* ! _ASM */ #define KHATID ksfmmup #define UHMEHASH_SZ uhmehash_num #define KHMEHASH_SZ khmehash_num #define HMENT_HASHAVELEN 4 #define HBLK_RANGE_SHIFT MMU_PAGESHIFT64K /* shift for HBLK_BS_MASK */ #define MAX_HASHCNT 5 #define DEFAULT_MAX_HASHCNT 3 #ifndef _ASM #define HASHADDR_MASK(hashno) TTE_PAGEMASK(hashno) #define HME_HASH_SHIFT(ttesz) \ ((ttesz == TTE8K)? HBLK_RANGE_SHIFT : TTE_PAGE_SHIFT(ttesz)) \ #define HME_HASH_ADDR(vaddr, hmeshift) \ ((caddr_t)(((uintptr_t)(vaddr) >> (hmeshift)) << (hmeshift))) #define HME_HASH_BSPAGE(vaddr, hmeshift) \ (((uintptr_t)(vaddr) >> (hmeshift)) << ((hmeshift) - MMU_PAGESHIFT)) #define HME_HASH_REHASH(ttesz) \ (((ttesz) < TTE512K)? 1 : (ttesz)) #define HME_HASH_FUNCTION(hatid, vaddr, shift) \ ((hatid != KHATID)? \ (&uhme_hash[ (((uintptr_t)(hatid) ^ ((uintptr_t)vaddr >> (shift))) & \ UHMEHASH_SZ) ]): \ (&khme_hash[ (((uintptr_t)(hatid) ^ ((uintptr_t)vaddr >> (shift))) & \ KHMEHASH_SZ) ])) /* * This macro will traverse a hmeblk hash link list looking for an hme_blk * that owns the specified vaddr and hatid. If if doesn't find one , hmeblkp * will be set to NULL, otherwise it will point to the correct hme_blk. * This macro also cleans empty hblks. */ #define HME_HASH_SEARCH_PREV(hmebp, hblktag, hblkp, hblkpa, \ pr_hblk, prevpa, listp) \ { \ struct hme_blk *nx_hblk; \ uint64_t nx_pa; \ \ ASSERT(SFMMU_HASH_LOCK_ISHELD(hmebp)); \ hblkp = hmebp->hmeblkp; \ hblkpa = hmebp->hmeh_nextpa; \ prevpa = 0; \ pr_hblk = NULL; \ while (hblkp) { \ if (HTAGS_EQ(hblkp->hblk_tag, hblktag)) { \ /* found hme_blk */ \ break; \ } \ nx_hblk = hblkp->hblk_next; \ nx_pa = hblkp->hblk_nextpa; \ if (!hblkp->hblk_vcnt && !hblkp->hblk_hmecnt) { \ sfmmu_hblk_hash_rm(hmebp, hblkp, prevpa, pr_hblk); \ sfmmu_hblk_free(hmebp, hblkp, hblkpa, listp); \ } else { \ pr_hblk = hblkp; \ prevpa = hblkpa; \ } \ hblkp = nx_hblk; \ hblkpa = nx_pa; \ } \ } #define HME_HASH_SEARCH(hmebp, hblktag, hblkp, listp) \ { \ struct hme_blk *pr_hblk; \ uint64_t hblkpa, prevpa; \ \ HME_HASH_SEARCH_PREV(hmebp, hblktag, hblkp, hblkpa, pr_hblk, \ prevpa, listp); \ } /* * This macro will traverse a hmeblk hash link list looking for an hme_blk * that owns the specified vaddr and hatid. If if doesn't find one , hmeblkp * will be set to NULL, otherwise it will point to the correct hme_blk. * It doesn't remove empty hblks. */ #define HME_HASH_FAST_SEARCH(hmebp, hblktag, hblkp) \ ASSERT(SFMMU_HASH_LOCK_ISHELD(hmebp)); \ for (hblkp = hmebp->hmeblkp; hblkp; \ hblkp = hblkp->hblk_next) { \ if (HTAGS_EQ(hblkp->hblk_tag, hblktag)) { \ /* found hme_blk */ \ break; \ } \ } \ #define SFMMU_HASH_LOCK(hmebp) \ (mutex_enter(&hmebp->hmehash_mutex)) #define SFMMU_HASH_UNLOCK(hmebp) \ (mutex_exit(&hmebp->hmehash_mutex)) #define SFMMU_HASH_LOCK_TRYENTER(hmebp) \ (mutex_tryenter(&hmebp->hmehash_mutex)) #define SFMMU_HASH_LOCK_ISHELD(hmebp) \ (mutex_owned(&hmebp->hmehash_mutex)) #define SFMMU_XCALL_STATS(ctxnum) \ { \ if (ctxnum == KCONTEXT) { \ SFMMU_STAT(sf_kernel_xcalls); \ } else { \ SFMMU_STAT(sf_user_xcalls); \ } \ } #define astosfmmu(as) ((as)->a_hat) #define sfmmutoctxnum(sfmmup) ((sfmmup)->sfmmu_cnum) #define sfmmutoctx(sfmmup) (&ctxs[sfmmutoctxnum(sfmmup)]) #define hblktosfmmu(hmeblkp) ((sfmmu_t *)(hmeblkp)->hblk_tag.htag_id) #define sfmmutoas(sfmmup) ((sfmmup)->sfmmu_as) #define ctxnumtoctx(ctxnum) (&ctxs[ctxnum]) /* * We use the sfmmu data structure to keep the per as page coloring info. */ #define as_color_bin(as) (astosfmmu(as)->sfmmu_clrbin) #define as_color_start(as) (astosfmmu(as)->sfmmu_clrstart) typedef struct { char h8[HME8BLK_SZ]; } hblk8_t; typedef struct { char h1[HME1BLK_SZ]; } hblk1_t; typedef struct { ulong_t index; ulong_t len; hblk8_t *list; } nucleus_hblk8_info_t; typedef struct { ulong_t index; ulong_t len; hblk1_t *list; } nucleus_hblk1_info_t; /* * This struct is used for accumlating information about a range * of pages that are unloading so that a single xcall can flush * the entire range from remote tlbs. A function that must demap * a range of virtual addresses declares one of these structures * and initializes using DEMP_RANGE_INIT(). It then passes a pointer to this * struct to the appropriate sfmmu_hblk_* level function which does * all the bookkeeping using the other macros. When the function has * finished the virtual address range, it needs to call DEMAP_RANGE_FLUSH() * macro to take care of any remaining unflushed mappings. * * The maximum range this struct can represent is the number of bits * in the dmr_bitvec field times the pagesize in dmr_pgsz. Currently, only * MMU_PAGESIZE pages are supported. * * Since there are now cases where it's no longer necessary to do * flushes (e.g. when the process isn't runnable because it's swapping * out or exiting) we allow these macros to take a NULL dmr input and do * nothing in that case. */ typedef struct { sfmmu_t *dmr_sfmmup; /* relevent hat */ caddr_t dmr_addr; /* beginning address */ caddr_t dmr_endaddr; /* ending address */ ulong_t dmr_bitvec; /* valid pages found */ ulong_t dmr_bit; /* next page to examine */ ulong_t dmr_maxbit; /* highest page in range */ ulong_t dmr_pgsz; /* page size in range */ } demap_range_t; #define DMR_MAXBIT ((ulong_t)1<<63) /* dmr_bit high bit */ #define DEMAP_RANGE_INIT(sfmmup, dmrp) \ if ((dmrp) != NULL) { \ (dmrp)->dmr_sfmmup = (sfmmup); \ (dmrp)->dmr_bitvec = 0; \ (dmrp)->dmr_maxbit = sfmmu_dmr_maxbit; \ (dmrp)->dmr_pgsz = MMU_PAGESIZE; \ } #define DEMAP_RANGE_PGSZ(dmrp) ((dmrp)? (dmrp)->dmr_pgsz : MMU_PAGESIZE) #define DEMAP_RANGE_CONTINUE(dmrp, addr, endaddr) \ if ((dmrp) != NULL) { \ if ((dmrp)->dmr_bitvec != 0 && (dmrp)->dmr_endaddr != (addr)) \ sfmmu_tlb_range_demap(dmrp); \ (dmrp)->dmr_endaddr = (endaddr); \ } #define DEMAP_RANGE_FLUSH(dmrp) \ if ((dmrp) != NULL) { \ if ((dmrp)->dmr_bitvec != 0) \ sfmmu_tlb_range_demap(dmrp); \ } #define DEMAP_RANGE_MARKPG(dmrp, addr) \ if ((dmrp) != NULL) { \ if ((dmrp)->dmr_bitvec == 0) { \ (dmrp)->dmr_addr = (addr); \ (dmrp)->dmr_bit = 1; \ } \ (dmrp)->dmr_bitvec |= (dmrp)->dmr_bit; \ } #define DEMAP_RANGE_NEXTPG(dmrp) \ if ((dmrp) != NULL && (dmrp)->dmr_bitvec != 0) { \ if ((dmrp)->dmr_bit & (dmrp)->dmr_maxbit) { \ sfmmu_tlb_range_demap(dmrp); \ } else { \ (dmrp)->dmr_bit <<= 1; \ } \ } /* * TSB related structures * * The TSB is made up of tte entries. Both the tag and data are present * in the TSB. The TSB locking is managed as follows: * A software bit in the tsb tag is used to indicate that entry is locked. * If a cpu servicing a tsb miss reads a locked entry the tag compare will * fail forcing the cpu to go to the hat hash for the translation. * The cpu who holds the lock can then modify the data side, and the tag side. * The last write should be to the word containing the lock bit which will * clear the lock and allow the tsb entry to be read. It is assumed that all * cpus reading the tsb will do so with atomic 128-bit loads. An atomic 128 * bit load is required to prevent the following from happening: * * cpu 0 cpu 1 comments * * ldx tag tag unlocked * ldstub lock set lock * stx data * stx tag unlock * ldx tag incorrect tte!!! * * The software also maintains a bit in the tag to indicate an invalid * tsb entry. The purpose of this bit is to allow the tsb invalidate code * to invalidate a tsb entry with a single cas. See code for details. */ union tsb_tag { struct { uint32_t tag_res0:16; /* reserved - context area */ uint32_t tag_inv:1; /* sw - invalid tsb entry */ uint32_t tag_lock:1; /* sw - locked tsb entry */ uint32_t tag_res1:4; /* reserved */ uint32_t tag_va_hi:10; /* va[63:54] */ uint32_t tag_va_lo; /* va[53:22] */ } tagbits; struct tsb_tagints { uint32_t inthi; uint32_t intlo; } tagints; }; #define tag_invalid tagbits.tag_inv #define tag_locked tagbits.tag_lock #define tag_vahi tagbits.tag_va_hi #define tag_valo tagbits.tag_va_lo #define tag_inthi tagints.inthi #define tag_intlo tagints.intlo struct tsbe { union tsb_tag tte_tag; tte_t tte_data; }; /* * A per cpu struct is kept that duplicates some info * used by the tl>0 tsb miss handlers plus it provides * a scratch area. Its purpose is to minimize cache misses * in the tsb miss handler and is 128 bytes (2 e$ lines). * * There should be one allocated per cpu in nucleus memory * and should be aligned on an ecache line boundary. */ struct tsbmiss { sfmmu_t *ksfmmup; /* kernel hat id */ sfmmu_t *usfmmup; /* user hat id */ struct tsbe *tsbptr; /* hardware computed ptr */ struct tsbe *tsbptr4m; /* hardware computed ptr */ uint64_t ismblkpa; struct hmehash_bucket *khashstart; struct hmehash_bucket *uhashstart; uint_t khashsz; uint_t uhashsz; uint16_t dcache_line_mask; /* used to flush dcache */ uint16_t hat_flags; uint32_t itlb_misses; uint32_t dtlb_misses; uint32_t utsb_misses; uint32_t ktsb_misses; uint16_t uprot_traps; uint16_t kprot_traps; /* * scratch[0] -> TSB_TAGACC * scratch[1] -> TSBMISS_HMEBP * scratch[2] -> TSBMISS_HATID */ uintptr_t scratch[3]; uint8_t pad[0x10]; }; /* * A per cpu struct is kept for the use within the tl>0 kpm tsb * miss handler. Some members are duplicates of common data or * the physical addresses of common data. A few members are also * written by the tl>0 kpm tsb miss handler. Its purpose is to * minimize cache misses in the kpm tsb miss handler and occupies * one ecache line. There should be one allocated per cpu in * nucleus memory and it should be aligned on an ecache line * boundary. It is not merged w/ struct tsbmiss since there is * not much to share and the tsbmiss pathes are different, so * a kpm tlbmiss/tsbmiss only touches one cacheline, except for * (DEBUG || SFMMU_STAT_GATHER) where the dtlb_misses counter * of struct tsbmiss is used on every dtlb miss. */ struct kpmtsbm { caddr_t vbase; /* start of address kpm range */ caddr_t vend; /* end of address kpm range */ uchar_t flags; /* flags needed in TL tsbmiss handler */ uchar_t sz_shift; /* for single kpm window */ uchar_t kpmp_shift; /* hash lock shift */ uchar_t kpmp2pshft; /* kpm page to page shift */ uint_t kpmp_table_sz; /* size of kpmp_table or kpmp_stable */ uint64_t kpmp_tablepa; /* paddr of kpmp_table or kpmp_stable */ uint64_t msegphashpa; /* paddr of memseg_phash */ struct tsbe *tsbptr; /* saved ktsb pointer */ uint_t kpm_dtlb_misses; /* kpm tlbmiss counter */ uint_t kpm_tsb_misses; /* kpm tsbmiss counter */ uintptr_t pad[1]; }; extern uint_t tsb_slab_size; extern uint_t tsb_slab_shift; extern uint_t tsb_slab_ttesz; extern uint_t tsb_slab_pamask; #endif /* !_ASM */ /* * Flags for TL kpm tsbmiss handler */ #define KPMTSBM_ENABLE_FLAG 0x01 /* bit copy of kpm_enable */ #define KPMTSBM_TLTSBM_FLAG 0x02 /* use TL tsbmiss handler */ #define KPMTSBM_TSBPHYS_FLAG 0x04 /* use ASI_MEM for TSB update */ /* * The TSB * All TSB sizes supported by the hardware are now supported (8K - 1M). * For kernel TSBs we may go beyond the hardware supported sizes and support * larger TSBs via software. * All TTE sizes are supported in the TSB; the manner in which this is * done is cpu dependent. */ #define TSB_MIN_SZCODE TSB_8K_SZCODE /* min. supported TSB size */ #define TSB_MIN_OFFSET_MASK (TSB_OFFSET_MASK(TSB_MIN_SZCODE)) #define UTSB_MAX_SZCODE TSB_1M_SZCODE /* max. supported TSB size */ #define UTSB_MAX_OFFSET_MASK (TSB_OFFSET_MASK(UTSB_MAX_SZCODE)) #define TSB_FREEMEM_MIN 0x1000 /* 32 mb */ #define TSB_FREEMEM_LARGE 0x10000 /* 512 mb */ #define TSB_8K_SZCODE 0 /* 512 entries */ #define TSB_16K_SZCODE 1 /* 1k entries */ #define TSB_32K_SZCODE 2 /* 2k entries */ #define TSB_64K_SZCODE 3 /* 4k entries */ #define TSB_128K_SZCODE 4 /* 8k entries */ #define TSB_256K_SZCODE 5 /* 16k entries */ #define TSB_512K_SZCODE 6 /* 32k entries */ #define TSB_1M_SZCODE 7 /* 64k entries */ #define TSB_2M_SZCODE 8 /* 128k entries */ #define TSB_4M_SZCODE 9 /* 256k entries */ #define TSB_ENTRY_SHIFT 4 /* each entry = 128 bits = 16 bytes */ #define TSB_ENTRY_SIZE (1 << 4) #define TSB_START_SIZE 9 #define TSB_ENTRIES(tsbsz) (1 << (TSB_START_SIZE + tsbsz)) #define TSB_BYTES(tsbsz) (TSB_ENTRIES(tsbsz) << TSB_ENTRY_SHIFT) #define TSB_OFFSET_MASK(tsbsz) (TSB_ENTRIES(tsbsz) - 1) #define TSB_BASEADDR_MASK ((1 << 12) - 1) /* * sun4u platforms * --------------- * We now support two user TSBs with one TSB base register. * Hence the TSB base register is split up as follows: * * When only one TSB present: * [63 62..42 41..13 12..4 3..0] * ^ ^ ^ ^ ^ * | | | | | * | | | | |_ TSB size code * | | | | * | | | |_ Reserved 0 * | | | * | | |_ TSB VA[41..13] * | | * | |_ VA hole (Spitfire), zeros (Cheetah and beyond) * | * |_ 0 * * When second TSB present: * [63 62..42 41..33 32..29 28..22 21..13 12..4 3..0] * ^ ^ ^ ^ ^ ^ ^ ^ * | | | | | | | | * | | | | | | | |_ First TSB size code * | | | | | | | * | | | | | | |_ Reserved 0 * | | | | | | * | | | | | |_ First TSB's VA[21..13] * | | | | | * | | | | |_ Reserved for future use * | | | | * | | | |_ Second TSB's size code * | | | * | | |_ Second TSB's VA[21..13] * | | * | |_ VA hole (Spitfire) / ones (Cheetah and beyond) * | * |_ 1 * * Note that since we store 21..13 of each TSB's VA, TSBs and their slabs * may be up to 4M in size. For now, only hardware supported TSB sizes * are supported, though the slabs are usually 4M in size. * * sun4v platforms * --------------- * On sun4v platforms, we use two dedicated scratchpad registers as pseudo * hardware TSB base registers to hold up to two different user TSBs. * * Each register contains TSB's physical base and size code information * as follows: * * [63..56 55..13 12..4 3..0] * ^ ^ ^ ^ * | | | | * | | | |_ TSB size code * | | | * | | |_ Reserved 0 * | | * | |_ TSB PA[55..13] * | * | * | * |_ 0 for valid TSB * * Absence of a user TSB (primarily the second user TSB) is indicated by * storing a negative value in the TSB base register. This allows us to * check for presence of a user TSB by simply checking bit# 63. */ #define TSBREG_MSB_SHIFT 32 /* set upper bits */ #define TSBREG_MSB_CONST 0xfffff800 /* set bits 63..43 */ #define TSBREG_FIRTSB_SHIFT 42 /* to clear bits 63:22 */ #define TSBREG_SECTSB_MKSHIFT 20 /* 21:13 --> 41:33 */ #define TSBREG_SECTSB_LSHIFT 22 /* to clear bits 63:42 */ #define TSBREG_SECTSB_RSHIFT (TSBREG_SECTSB_MKSHIFT + TSBREG_SECTSB_LSHIFT) /* sectsb va -> bits 21:13 */ /* after clearing upper bits */ #define TSBREG_SECSZ_SHIFT 29 /* to get sectsb szc to 3:0 */ #define TSBREG_VAMASK_SHIFT 13 /* set up VA mask */ #define BIGKTSB_SZ_MASK 0xf #define TSB_SOFTSZ_MASK BIGKTSB_SZ_MASK #define MIN_BIGKTSB_SZCODE 9 /* 256k entries */ #define MAX_BIGKTSB_SZCODE 11 /* 1024k entries */ #define MAX_BIGKTSB_TTES (TSB_BYTES(MAX_BIGKTSB_SZCODE) / MMU_PAGESIZE4M) #define TAG_VALO_SHIFT 22 /* tag's va are bits 63-22 */ /* * sw bits used on tsb_tag - bit masks used only in assembly * use only a sethi for these fields. */ #define TSBTAG_INVALID 0x00008000 /* tsb_tag.tag_invalid */ #define TSBTAG_LOCKED 0x00004000 /* tsb_tag.tag_locked */ #ifdef _ASM /* * Marker to indicate that this instruction will be hot patched at runtime * to some other value. * This value must be zero since it fills in the imm bits of the target * instructions to be patched */ #define RUNTIME_PATCH (0) /* * V9 defines nop instruction as the following, which we use * at runtime to nullify some instructions we don't want to * execute in the trap handlers on certain platforms. */ #define MAKE_NOP_INSTR(reg) \ sethi %hi(0x1000000), reg /* * Macro to get this CPU's tsbmiss area. */ #define CPU_TSBMISS_AREA(tsbmiss, tmp1) \ CPU_INDEX(tmp1, tsbmiss); /* tmp1 = cpu idx */ \ sethi %hi(tsbmiss_area), tsbmiss; /* tsbmiss base ptr */ \ sllx tmp1, TSBMISS_SHIFT, tmp1; /* byte offset */ \ or tsbmiss, %lo(tsbmiss_area), tsbmiss; \ add tsbmiss, tmp1, tsbmiss /* tsbmiss area of CPU */ /* * Macro to set kernel context + page size codes in DMMU primary context * register. It is only necessary for sun4u because sun4v does not need * page size codes */ #ifdef sun4v #define SET_KCONTEXTREG(reg0, reg1, reg2, reg3, reg4, label1, label2, label3) #else #define SET_KCONTEXTREG(reg0, reg1, reg2, reg3, reg4, label1, label2, label3) \ sethi %hi(kcontextreg), reg0; \ ldx [reg0 + %lo(kcontextreg)], reg0; \ mov MMU_PCONTEXT, reg1; \ ldxa [reg1]ASI_MMU_CTX, reg2; \ xor reg0, reg2, reg2; \ brz reg2, label3; \ srlx reg2, CTXREG_NEXT_SHIFT, reg2; \ rdpr %pstate, reg3; /* disable interrupts */ \ btst PSTATE_IE, reg3; \ /*CSTYLED*/ \ bnz,a,pt %icc, label1; \ wrpr reg3, PSTATE_IE, %pstate; \ /*CSTYLED*/ \ label1:; \ brz reg2, label2; /* need demap if N_pgsz0/1 change */ \ sethi %hi(FLUSH_ADDR), reg4; \ mov DEMAP_ALL_TYPE, reg2; \ stxa %g0, [reg2]ASI_DTLB_DEMAP; \ stxa %g0, [reg2]ASI_ITLB_DEMAP; \ /*CSTYLED*/ \ label2:; \ stxa reg0, [reg1]ASI_MMU_CTX; \ flush reg4; \ btst PSTATE_IE, reg3; \ /*CSTYLED*/ \ bnz,a,pt %icc, label3; \ wrpr %g0, reg3, %pstate; /* restore interrupt state */ \ label3:; #endif #endif /* _ASM */ #ifndef _ASM /* * Page coloring * The p_vcolor field of the page struct (1 byte) is used to store the * virtual page color. This provides for 255 colors. The value zero is * used to mean the page has no color - never been mapped or somehow * purified. */ #define PP_GET_VCOLOR(pp) (((pp)->p_vcolor) - 1) #define PP_NEWPAGE(pp) (!(pp)->p_vcolor) #define PP_SET_VCOLOR(pp, color) \ ((pp)->p_vcolor = ((color) + 1)) /* * As mentioned p_vcolor == 0 means there is no color for this page. * But PP_SET_VCOLOR(pp, color) expects 'color' to be real color minus * one so we define this constant. */ #define NO_VCOLOR (-1) #define addr_to_vcolor(addr) \ (((uint_t)(uintptr_t)(addr) >> MMU_PAGESHIFT) & vac_colors_mask) /* * The field p_index in the psm page structure is for large pages support. * P_index is a bit-vector of the different mapping sizes that a given page * is part of. An hme structure for a large mapping is only added in the * group leader page (first page). All pages covered by a given large mapping * have the corrosponding mapping bit set in their p_index field. This allows * us to only store an explicit hme structure in the leading page which * simplifies the mapping link list management. Furthermore, it provides us * a fast mechanism for determining the largest mapping a page is part of. For * exmaple, a page with a 64K and a 4M mappings has a p_index value of 0x0A. * * Implementation note: even though the first bit in p_index is reserved * for 8K mappings, it is NOT USED by the code and SHOULD NOT be set. * In addition, the upper four bits of the p_index field are used by the * code as temporaries */ /* * Defines for psm page struct fields and large page support */ #define SFMMU_INDEX_SHIFT 6 #define SFMMU_INDEX_MASK ((1 << SFMMU_INDEX_SHIFT) - 1) /* Return the mapping index */ #define PP_MAPINDEX(pp) ((pp)->p_index & SFMMU_INDEX_MASK) /* * These macros rely on the following property: * All pages constituting a large page are covered by a virtually * contiguous set of page_t's. */ /* Return the leader for this mapping size */ #define PP_GROUPLEADER(pp, sz) \ (&(pp)[-(int)(pp->p_pagenum & (TTEPAGES(sz)-1))]) /* Return the root page for this page based on p_szc */ #define PP_PAGEROOT(pp) ((pp)->p_szc == 0 ? (pp) : \ PP_GROUPLEADER((pp), (pp)->p_szc)) #define PP_PAGENEXT_N(pp, n) ((pp) + (n)) #define PP_PAGENEXT(pp) PP_PAGENEXT_N((pp), 1) #define PP_PAGEPREV_N(pp, n) ((pp) - (n)) #define PP_PAGEPREV(pp) PP_PAGEPREV_N((pp), 1) #define PP_ISMAPPED_LARGE(pp) (PP_MAPINDEX(pp) != 0) /* Need function to test the page mappping which takes p_index into account */ #define PP_ISMAPPED(pp) ((pp)->p_mapping || PP_ISMAPPED_LARGE(pp)) /* * Don't call this macro with sz equal to zero. 8K mappings SHOULD NOT * set p_index field. */ #define PAGESZ_TO_INDEX(sz) (1 << (sz)) /* * prototypes for hat assembly routines. Some of these are * known to machine dependent VM code. */ extern uint64_t sfmmu_make_tsbtag(caddr_t); extern struct tsbe * sfmmu_get_tsbe(uint64_t, caddr_t, int, int); extern void sfmmu_load_tsbe(struct tsbe *, uint64_t, tte_t *, int); extern void sfmmu_unload_tsbe(struct tsbe *, uint64_t, int); extern void sfmmu_load_mmustate(sfmmu_t *); extern void sfmmu_ctx_steal_tl1(uint64_t, uint64_t); extern void sfmmu_raise_tsb_exception(uint64_t, uint64_t); #ifndef sun4v extern void sfmmu_itlb_ld(caddr_t, int, tte_t *); extern void sfmmu_dtlb_ld(caddr_t, int, tte_t *); #endif /* sun4v */ extern void sfmmu_copytte(tte_t *, tte_t *); extern int sfmmu_modifytte(tte_t *, tte_t *, tte_t *); extern int sfmmu_modifytte_try(tte_t *, tte_t *, tte_t *); extern pfn_t sfmmu_ttetopfn(tte_t *, caddr_t); extern void sfmmu_hblk_hash_rm(struct hmehash_bucket *, struct hme_blk *, uint64_t, struct hme_blk *); extern void sfmmu_hblk_hash_add(struct hmehash_bucket *, struct hme_blk *, uint64_t); /* * functions exported to machine dependent VM code */ extern void sfmmu_patch_ktsb(void); #ifndef sun4v extern void sfmmu_patch_utsb(void); #endif /* sun4v */ extern pfn_t sfmmu_vatopfn(caddr_t, sfmmu_t *, tte_t *); extern void sfmmu_vatopfn_suspended(caddr_t, sfmmu_t *, tte_t *); #ifdef DEBUG extern void sfmmu_check_kpfn(pfn_t); #else #define sfmmu_check_kpfn(pfn) /* disabled */ #endif /* DEBUG */ extern void sfmmu_memtte(tte_t *, pfn_t, uint_t, int); extern void sfmmu_tteload(struct hat *, tte_t *, caddr_t, page_t *, uint_t); extern void sfmmu_tsbmiss_exception(struct regs *, uintptr_t, uint_t); extern void sfmmu_init_tsbs(void); extern caddr_t sfmmu_ktsb_alloc(caddr_t); extern int sfmmu_getctx_pri(void); extern int sfmmu_getctx_sec(void); extern void sfmmu_setctx_sec(int); extern void sfmmu_inv_tsb(caddr_t, uint_t); extern void sfmmu_init_ktsbinfo(void); extern int sfmmu_setup_4lp(void); extern void sfmmu_patch_mmu_asi(int); extern void sfmmu_init_nucleus_hblks(caddr_t, size_t, int, int); extern void sfmmu_cache_flushall(void); extern pgcnt_t sfmmu_tte_cnt(sfmmu_t *, uint_t); extern void *sfmmu_tsb_segkmem_alloc(vmem_t *, size_t, int); extern void sfmmu_tsb_segkmem_free(vmem_t *, void *, size_t); extern void sfmmu_steal_context(sfmmu_t *, uint8_t *); extern void hat_kern_setup(void); extern int hat_page_relocate(page_t **, page_t **, spgcnt_t *); extern uint_t hat_preferred_pgsz(struct hat *, caddr_t, size_t, int); extern int sfmmu_get_ppvcolor(struct page *); extern int sfmmu_get_addrvcolor(caddr_t); extern int sfmmu_hat_lock_held(sfmmu_t *); /* * Functions exported to xhat_sfmmu.c */ extern kmutex_t *sfmmu_mlist_enter(page_t *); extern void sfmmu_mlist_exit(kmutex_t *); extern int sfmmu_mlist_held(struct page *); extern struct hme_blk *sfmmu_hmetohblk(struct sf_hment *); /* * MMU-specific functions optionally imported from the CPU module */ #pragma weak mmu_large_pages_disabled #pragma weak mmu_set_ctx_page_sizes #pragma weak mmu_preferred_pgsz #pragma weak mmu_check_page_sizes extern int mmu_large_pages_disabled(uint_t); extern void mmu_set_ctx_page_sizes(sfmmu_t *); extern uint_t mmu_preferred_pgsz(sfmmu_t *, caddr_t, size_t); extern void mmu_check_page_sizes(sfmmu_t *, uint64_t *); extern sfmmu_t *ksfmmup; extern struct ctx *ctxs; extern uint_t nctxs; extern caddr_t ktsb_base; extern uint64_t ktsb_pbase; extern int ktsb_sz; extern int ktsb_szcode; extern caddr_t ktsb4m_base; extern uint64_t ktsb4m_pbase; extern int ktsb4m_sz; extern int ktsb4m_szcode; extern uint64_t kpm_tsbbase; extern int kpm_tsbsz; extern int ktsb_phys; extern int enable_bigktsb; #ifndef sun4v extern int utsb_dtlb_ttenum; extern int utsb4m_dtlb_ttenum; #endif /* sun4v */ extern int uhmehash_num; extern int khmehash_num; extern struct hmehash_bucket *uhme_hash; extern struct hmehash_bucket *khme_hash; extern kmutex_t *mml_table; extern uint_t mml_table_sz; extern uint_t mml_shift; extern uint_t hblk_alloc_dynamic; extern struct tsbmiss tsbmiss_area[NCPU]; extern struct kpmtsbm kpmtsbm_area[NCPU]; extern int tsb_max_growsize; #ifndef sun4v extern int dtlb_resv_ttenum; extern caddr_t utsb_vabase; extern caddr_t utsb4m_vabase; #endif /* sun4v */ extern vmem_t *kmem_tsb_default_arena[]; extern int tsb_lgrp_affinity; /* kpm externals */ extern pfn_t sfmmu_kpm_vatopfn(caddr_t); extern void sfmmu_kpm_patch_tlbm(void); extern void sfmmu_kpm_patch_tsbm(void); extern void sfmmu_kpm_load_tsb(caddr_t, tte_t *, int); extern void sfmmu_kpm_unload_tsb(caddr_t, int); extern void sfmmu_kpm_tsbmtl(short *, uint_t *, int); extern int sfmmu_kpm_stsbmtl(char *, uint_t *, int); extern caddr_t kpm_vbase; extern size_t kpm_size; extern struct memseg *memseg_hash[]; extern uint64_t memseg_phash[]; extern kpm_hlk_t *kpmp_table; extern kpm_shlk_t *kpmp_stable; extern uint_t kpmp_table_sz; extern uint_t kpmp_stable_sz; extern uchar_t kpmp_shift; #define PP_ISMAPPED_KPM(pp) ((pp)->p_kpmref > 0) #define IS_KPM_ALIAS_RANGE(vaddr) \ (((vaddr) - kpm_vbase) >> (uintptr_t)kpm_size_shift > 0) #endif /* !_ASM */ /* sfmmu_kpm_tsbmtl flags */ #define KPMTSBM_STOP 0 #define KPMTSBM_START 1 /* kpm_smallpages kp_mapped values */ #define KPM_MAPPEDS -1 /* small mapping valid, no conflict */ #define KPM_MAPPEDSC 1 /* small mapping valid, conflict */ /* Physical memseg address NULL marker */ #define MSEG_NULLPTR_PA -1 /* * Memseg hash defines for kpm trap level tsbmiss handler. * Must be in sync w/ page.h . */ #define SFMMU_MEM_HASH_SHIFT 0x9 #define SFMMU_N_MEM_SLOTS 0x200 #define SFMMU_MEM_HASH_ENTRY_SHIFT 3 #ifndef _ASM #if (SFMMU_MEM_HASH_SHIFT != MEM_HASH_SHIFT) #error SFMMU_MEM_HASH_SHIFT != MEM_HASH_SHIFT #endif #if (SFMMU_N_MEM_SLOTS != N_MEM_SLOTS) #error SFMMU_N_MEM_SLOTS != N_MEM_SLOTS #endif /* Physical memseg address NULL marker */ #define SFMMU_MEMSEG_NULLPTR_PA -1 /* * Check KCONTEXT to be zero, asm parts depend on that assumption. */ #if (KCONTEXT != 0) #error KCONTEXT != 0 #endif #endif /* !_ASM */ #endif /* _KERNEL */ #ifndef _ASM /* * ctx, hmeblk, mlistlock and other stats for sfmmu */ struct sfmmu_global_stat { int sf_tsb_exceptions; /* # of tsb exceptions */ int sf_tsb_raise_exception; /* # tsb exc. w/o TLB flush */ int sf_pagefaults; /* # of pagefaults */ int sf_uhash_searches; /* # of user hash searches */ int sf_uhash_links; /* # of user hash links */ int sf_khash_searches; /* # of kernel hash searches */ int sf_khash_links; /* # of kernel hash links */ int sf_swapout; /* # times hat swapped out */ int sf_ctxfree; /* ctx alloc from free list */ int sf_ctxdirty; /* ctx alloc from dirty list */ int sf_ctxsteal; /* ctx allocated by steal */ int sf_tsb_alloc; /* # TSB allocations */ int sf_tsb_allocfail; /* # times TSB alloc fail */ int sf_tsb_sectsb_create; /* # times second TSB added */ int sf_tteload8k; /* calls to sfmmu_tteload */ int sf_tteload64k; /* calls to sfmmu_tteload */ int sf_tteload512k; /* calls to sfmmu_tteload */ int sf_tteload4m; /* calls to sfmmu_tteload */ int sf_tteload32m; /* calls to sfmmu_tteload */ int sf_tteload256m; /* calls to sfmmu_tteload */ int sf_tsb_load8k; /* # times loaded 8K tsbent */ int sf_tsb_load4m; /* # times loaded 4M tsbent */ int sf_hblk_hit; /* found hblk during tteload */ int sf_hblk8_ncreate; /* static hblk8's created */ int sf_hblk8_nalloc; /* static hblk8's allocated */ int sf_hblk1_ncreate; /* static hblk1's created */ int sf_hblk1_nalloc; /* static hblk1's allocated */ int sf_hblk_slab_cnt; /* sfmmu8_cache slab creates */ int sf_hblk_reserve_cnt; /* hblk_reserve usage */ int sf_hblk_recurse_cnt; /* hblk_reserve owner reqs */ int sf_hblk_reserve_hit; /* hblk_reserve hash hits */ int sf_get_free_success; /* reserve list allocs */ int sf_get_free_throttle; /* fails due to throttling */ int sf_get_free_fail; /* fails due to empty list */ int sf_put_free_success; /* reserve list frees */ int sf_put_free_fail; /* fails due to full list */ int sf_pgcolor_conflict; /* VAC conflict resolution */ int sf_uncache_conflict; /* VAC conflict resolution */ int sf_unload_conflict; /* VAC unload resolution */ int sf_ism_uncache; /* VAC conflict resolution */ int sf_ism_recache; /* VAC conflict resolution */ int sf_recache; /* VAC conflict resolution */ int sf_steal_count; /* # of hblks stolen */ int sf_pagesync; /* # of pagesyncs */ int sf_clrwrt; /* # of clear write perms */ int sf_pagesync_invalid; /* pagesync with inv tte */ int sf_kernel_xcalls; /* # of kernel cross calls */ int sf_user_xcalls; /* # of user cross calls */ int sf_tsb_grow; /* # of user tsb grows */ int sf_tsb_shrink; /* # of user tsb shrinks */ int sf_tsb_resize_failures; /* # of user tsb resize */ int sf_tsb_reloc; /* # of user tsb relocations */ int sf_user_vtop; /* # of user vatopfn calls */ int sf_ctx_swap; /* # times switched MMU ctxs */ int sf_tlbflush_all; /* # times flush all TLBs */ int sf_tlbflush_ctx; /* # times flush TLB ctx */ int sf_tlbflush_deferred; /* # times !flush ctx imm. */ int sf_tlb_reprog_pgsz; /* # times switch TLB pgsz */ }; struct sfmmu_tsbsize_stat { int sf_tsbsz_8k; int sf_tsbsz_16k; int sf_tsbsz_32k; int sf_tsbsz_64k; int sf_tsbsz_128k; int sf_tsbsz_256k; int sf_tsbsz_512k; int sf_tsbsz_1m; int sf_tsbsz_2m; int sf_tsbsz_4m; }; struct sfmmu_percpu_stat { int sf_itlb_misses; /* # of itlb misses */ int sf_dtlb_misses; /* # of dtlb misses */ int sf_utsb_misses; /* # of user tsb misses */ int sf_ktsb_misses; /* # of kernel tsb misses */ int sf_tsb_hits; /* # of tsb hits */ int sf_umod_faults; /* # of mod (prot viol) flts */ int sf_kmod_faults; /* # of mod (prot viol) flts */ }; #define SFMMU_STAT(stat) sfmmu_global_stat.stat++; #define SFMMU_STAT_ADD(stat, amount) sfmmu_global_stat.stat += amount; #define SFMMU_STAT_SET(stat, count) sfmmu_global_stat.stat = count; #endif /* !_ASM */ #ifdef __cplusplus } #endif #endif /* _VM_HAT_SFMMU_H */