diff options
| -rw-r--r-- | sys/uvm/uvm_addr.c | 42 | ||||
| -rw-r--r-- | sys/uvm/uvm_amap.c | 194 | ||||
| -rw-r--r-- | sys/uvm/uvm_anon.c | 49 | ||||
| -rw-r--r-- | sys/uvm/uvm_aobj.c | 254 | ||||
| -rw-r--r-- | sys/uvm/uvm_device.c | 55 | ||||
| -rw-r--r-- | sys/uvm/uvm_fault.c | 84 | ||||
| -rw-r--r-- | sys/uvm/uvm_init.c | 42 | ||||
| -rw-r--r-- | sys/uvm/uvm_io.c | 24 | ||||
| -rw-r--r-- | sys/uvm/uvm_km.c | 11 |
9 files changed, 463 insertions, 292 deletions
diff --git a/sys/uvm/uvm_addr.c b/sys/uvm/uvm_addr.c index ca72453cfbb..317a427da49 100644 --- a/sys/uvm/uvm_addr.c +++ b/sys/uvm/uvm_addr.c @@ -1,4 +1,4 @@ -/* $OpenBSD: uvm_addr.c,v 1.29 2020/09/22 14:31:08 mpi Exp $ */ +/* $OpenBSD: uvm_addr.c,v 1.30 2021/03/20 10:24:21 mpi Exp $ */ /* * Copyright (c) 2011 Ariane van der Steldt <ariane@stack.nl> @@ -65,7 +65,9 @@ struct uaddr_rnd_state { #endif }; -/* Definition of a pivot in pivot selector. */ +/* + * Definition of a pivot in pivot selector. + */ struct uaddr_pivot { vaddr_t addr; /* End of prev. allocation. */ int expire;/* Best before date. */ @@ -87,7 +89,11 @@ struct uaddr_pivot_state { extern const struct uvm_addr_functions uaddr_kernel_functions; struct uvm_addr_state uaddr_kbootstrap; -/* Support functions. */ + +/* + * Support functions. + */ + #ifndef SMALL_KERNEL struct vm_map_entry *uvm_addr_entrybyspace(struct uaddr_free_rbtree*, vsize_t); @@ -236,7 +242,9 @@ uvm_addr_fitspace(vaddr_t *min_result, vaddr_t *max_result, if (fspace - before_gap - after_gap < sz) return ENOMEM; - /* Calculate lowest address. */ + /* + * Calculate lowest address. + */ low_addr += before_gap; low_addr = uvm_addr_align_forward(tmp = low_addr, align, offset); if (low_addr < tmp) /* Overflow during alignment. */ @@ -244,7 +252,9 @@ uvm_addr_fitspace(vaddr_t *min_result, vaddr_t *max_result, if (high_addr - after_gap - sz < low_addr) return ENOMEM; - /* Calculate highest address. */ + /* + * Calculate highest address. + */ high_addr -= after_gap + sz; high_addr = uvm_addr_align_backward(tmp = high_addr, align, offset); if (high_addr > tmp) /* Overflow during alignment. */ @@ -341,7 +351,9 @@ uvm_addr_linsearch(struct vm_map *map, struct uvm_addr_state *uaddr, (before_gap & PAGE_MASK) == 0 && (after_gap & PAGE_MASK) == 0); KASSERT(high + sz > high); /* Check for overflow. */ - /* Hint magic. */ + /* + * Hint magic. + */ if (hint == 0) hint = (direction == 1 ? low : high); else if (hint > high) { @@ -463,6 +475,7 @@ uaddr_destroy(struct uvm_addr_state *uaddr) * If hint is set, search will start at the hint position. * Only searches forward. */ + const struct uvm_addr_functions uaddr_lin_functions = { .uaddr_select = &uaddr_lin_select, .uaddr_destroy = &uaddr_destroy, @@ -489,7 +502,9 @@ uaddr_lin_select(struct vm_map *map, struct uvm_addr_state *uaddr, { vaddr_t guard_sz; - /* Deal with guardpages: search for space with one extra page. */ + /* + * Deal with guardpages: search for space with one extra page. + */ guard_sz = ((map->flags & VM_MAP_GUARDPAGES) == 0 ? 0 : PAGE_SIZE); if (uaddr->uaddr_maxaddr - uaddr->uaddr_minaddr - guard_sz < sz) @@ -716,6 +731,7 @@ uaddr_rnd_print(struct uvm_addr_state *uaddr_p, boolean_t full, /* * Kernel allocation bootstrap logic. */ + const struct uvm_addr_functions uaddr_kernel_functions = { .uaddr_select = &uaddr_kbootstrap_select, .uaddr_destroy = &uaddr_kbootstrap_destroy, @@ -839,7 +855,9 @@ uaddr_bestfit_select(struct vm_map *map, struct uvm_addr_state *uaddr_p, if (entry == NULL) return ENOMEM; - /* Walk the tree until we find an entry that fits. */ + /* + * Walk the tree until we find an entry that fits. + */ while (uvm_addr_fitspace(&min, &max, VMMAP_FREE_START(entry), VMMAP_FREE_END(entry), sz, align, offset, 0, guardsz) != 0) { @@ -848,7 +866,9 @@ uaddr_bestfit_select(struct vm_map *map, struct uvm_addr_state *uaddr_p, return ENOMEM; } - /* Return the address that generates the least fragmentation. */ + /* + * Return the address that generates the least fragmentation. + */ *entry_out = entry; *addr_out = (min - VMMAP_FREE_START(entry) <= VMMAP_FREE_END(entry) - guardsz - sz - max ? @@ -1128,7 +1148,9 @@ uaddr_pivot_select(struct vm_map *map, struct uvm_addr_state *uaddr_p, if (pivot->addr == 0 || pivot->entry == NULL || pivot->expire == 0) goto expired; /* Pivot is invalid (null or expired). */ - /* Attempt to use the pivot to map the entry. */ + /* + * Attempt to use the pivot to map the entry. + */ entry = pivot->entry; if (pivot->dir > 0) { if (uvm_addr_fitspace(&min, &max, diff --git a/sys/uvm/uvm_amap.c b/sys/uvm/uvm_amap.c index 7eb20e6a95d..52b2f3998c8 100644 --- a/sys/uvm/uvm_amap.c +++ b/sys/uvm/uvm_amap.c @@ -1,4 +1,4 @@ -/* $OpenBSD: uvm_amap.c,v 1.87 2021/01/19 13:21:36 mpi Exp $ */ +/* $OpenBSD: uvm_amap.c,v 1.88 2021/03/20 10:24:21 mpi Exp $ */ /* $NetBSD: uvm_amap.c,v 1.27 2000/11/25 06:27:59 chs Exp $ */ /* @@ -188,7 +188,7 @@ amap_chunk_free(struct vm_amap *amap, struct vm_amap_chunk *chunk) * when enabled, an array of ints is allocated for the pprefs. this * array is allocated only when a partial reference is added to the * map (either by unmapping part of the amap, or gaining a reference - * to only a part of an amap). if the malloc of the array fails + * to only a part of an amap). if the allocation of the array fails * (M_NOWAIT), then we set the array pointer to PPREF_NONE to indicate * that we tried to do ppref's but couldn't alloc the array so just * give up (after all, this is an optional feature!). @@ -209,12 +209,14 @@ amap_chunk_free(struct vm_amap *amap, struct vm_amap_chunk *chunk) * chunk. note that the "plus one" part is needed because a reference * count of zero is neither positive or negative (need a way to tell * if we've got one zero or a bunch of them). - * + * * here are some in-line functions to help us. */ /* * pp_getreflen: get the reference and length for a specific offset + * + * => ppref's amap must be locked */ static inline void pp_getreflen(int *ppref, int offset, int *refp, int *lenp) @@ -231,6 +233,8 @@ pp_getreflen(int *ppref, int offset, int *refp, int *lenp) /* * pp_setreflen: set the reference and length for a specific offset + * + * => ppref's amap must be locked */ static inline void pp_setreflen(int *ppref, int offset, int ref, int len) @@ -242,7 +246,7 @@ pp_setreflen(int *ppref, int offset, int ref, int len) ppref[offset+1] = len; } } -#endif +#endif /* UVM_AMAP_PPREF */ /* * amap_init: called at boot time to init global amap data structures @@ -276,8 +280,9 @@ amap_init(void) } /* - * amap_alloc1: internal function that allocates an amap, but does not - * init the overlay. + * amap_alloc1: allocate an amap, but do not initialise the overlay. + * + * => Note: lock is not set. */ static inline struct vm_amap * amap_alloc1(int slots, int waitf, int lazyalloc) @@ -408,6 +413,7 @@ amap_lock_alloc(struct vm_amap *amap) * * => caller should ensure sz is a multiple of PAGE_SIZE * => reference count to new amap is set to one + * => new amap is returned unlocked */ struct vm_amap * amap_alloc(vaddr_t sz, int waitf, int lazyalloc) @@ -432,6 +438,7 @@ amap_alloc(vaddr_t sz, int waitf, int lazyalloc) /* * amap_free: free an amap * + * => the amap must be unlocked * => the amap should have a zero reference count and be empty */ void @@ -466,11 +473,9 @@ amap_free(struct vm_amap *amap) /* * amap_wipeout: wipeout all anon's in an amap; then free the amap! * - * => called from amap_unref when the final reference to an amap is - * discarded (i.e. when reference count == 1) + * => Called from amap_unref(), when reference count drops to zero. * => amap must be locked. */ - void amap_wipeout(struct vm_amap *amap) { @@ -483,7 +488,9 @@ amap_wipeout(struct vm_amap *amap) KASSERT(amap->am_ref == 0); if (__predict_false((amap->am_flags & AMAP_SWAPOFF) != 0)) { - /* amap_swap_off will call us again. */ + /* + * Note: amap_swap_off() will call us again. + */ amap_unlock(amap); return; } @@ -503,12 +510,11 @@ amap_wipeout(struct vm_amap *amap) panic("amap_wipeout: corrupt amap"); KASSERT(anon->an_lock == amap->am_lock); + /* + * Drop the reference. + */ refs = --anon->an_ref; if (refs == 0) { - /* - * we had the last reference to a vm_anon. - * free it. - */ uvm_anfree_list(anon, &pgl); } } @@ -516,7 +522,9 @@ amap_wipeout(struct vm_amap *amap) /* free the pages */ uvm_pglistfree(&pgl); - /* now we free the map */ + /* + * Finally, destroy the amap. + */ amap->am_ref = 0; /* ... was one */ amap->am_nused = 0; amap_unlock(amap); @@ -526,10 +534,10 @@ amap_wipeout(struct vm_amap *amap) /* * amap_copy: ensure that a map entry's "needs_copy" flag is false * by copying the amap if necessary. - * + * * => an entry with a null amap pointer will get a new (blank) one. - * => the map that the map entry blocks to must be locked by caller. - * => the amap (if any) currently attached to the entry must be unlocked. + * => the map that the map entry belongs to must be locked by caller. + * => the amap currently attached to "entry" (if any) must be unlocked. * => if canchunk is true, then we may clip the entry into a chunk * => "startva" and "endva" are used only if canchunk is true. they are * used to limit chunking (e.g. if you have a large space that you @@ -550,14 +558,16 @@ amap_copy(struct vm_map *map, struct vm_map_entry *entry, int waitf, KASSERT(map != kernel_map); /* we use sleeping locks */ - /* is there a map to copy? if not, create one from scratch. */ + /* + * Is there an amap to copy? If not, create one. + */ if (entry->aref.ar_amap == NULL) { /* - * check to see if we have a large amap that we can - * chunk. we align startva/endva to chunk-sized + * Check to see if we have a large amap that we can + * chunk. We align startva/endva to chunk-sized * boundaries and then clip to them. * - * if we cannot chunk the amap, allocate it in a way + * If we cannot chunk the amap, allocate it in a way * that makes it grow or shrink dynamically with * the number of slots. */ @@ -584,17 +594,21 @@ amap_copy(struct vm_map *map, struct vm_map_entry *entry, int waitf, } /* - * first check and see if we are the only map entry - * referencing the amap we currently have. if so, then we can - * just take it over rather than copying it. the value can only - * be one if we have the only reference to the amap + * First check and see if we are the only map entry referencing + * he amap we currently have. If so, then just take it over instead + * of copying it. Note that we are reading am_ref without lock held + * as the value value can only be one if we have the only reference + * to the amap (via our locked map). If the value is greater than + * one, then allocate amap and re-check the value. */ if (entry->aref.ar_amap->am_ref == 1) { entry->etype &= ~UVM_ET_NEEDSCOPY; return; } - /* looks like we need to copy the map. */ + /* + * Allocate a new amap (note: not initialised, etc). + */ AMAP_B2SLOT(slots, entry->end - entry->start); if (!UVM_AMAP_SMALL(entry->aref.ar_amap) && entry->aref.ar_amap->am_hashshift != 0) @@ -607,20 +621,22 @@ amap_copy(struct vm_map *map, struct vm_map_entry *entry, int waitf, amap_lock(srcamap); /* - * need to double check reference count now. the reference count - * could have changed while we were in malloc. if the reference count - * dropped down to one we take over the old map rather than - * copying the amap. + * Re-check the reference count with the lock held. If it has + * dropped to one - we can take over the existing map. */ - if (srcamap->am_ref == 1) { /* take it over? */ + if (srcamap->am_ref == 1) { + /* Just take over the existing amap. */ entry->etype &= ~UVM_ET_NEEDSCOPY; amap_unlock(srcamap); - amap->am_ref--; /* drop final reference to map */ - amap_free(amap); /* dispose of new (unused) amap */ + /* Destroy the new (unused) amap. */ + amap->am_ref--; + amap_free(amap); return; } - /* we must copy it now. */ + /* + * Copy the slots. + */ for (lcv = 0; lcv < slots; lcv += n) { srcslot = entry->aref.ar_pageoff + lcv; i = UVM_AMAP_SLOTIDX(lcv); @@ -659,10 +675,9 @@ amap_copy(struct vm_map *map, struct vm_map_entry *entry, int waitf, } /* - * drop our reference to the old amap (srcamap). - * we know that the reference count on srcamap is greater than - * one (we checked above), so there is no way we could drop - * the count to zero. [and no need to worry about freeing it] + * Drop our reference to the old amap (srcamap) and unlock. + * Since the reference count on srcamap is greater than one, + * (we checked above), it cannot drop to zero while it is locked. */ srcamap->am_ref--; KASSERT(srcamap->am_ref > 0); @@ -690,7 +705,9 @@ amap_copy(struct vm_map *map, struct vm_map_entry *entry, int waitf, if (amap->am_lock == NULL) amap_lock_alloc(amap); - /* install new amap. */ + /* + * Install new amap. + */ entry->aref.ar_pageoff = 0; entry->aref.ar_amap = amap; entry->etype &= ~UVM_ET_NEEDSCOPY; @@ -723,9 +740,9 @@ amap_cow_now(struct vm_map *map, struct vm_map_entry *entry) struct vm_amap_chunk *chunk; /* - * note that if we wait, we must ReStart the "lcv" for loop because - * some other process could reorder the anon's in the - * am_anon[] array on us. + * note that if we unlock the amap then we must ReStart the "lcv" for + * loop because some other process could reorder the anon's in the + * am_anon[] array on us while the lock is dropped. */ ReStart: amap_lock(amap); @@ -739,7 +756,10 @@ ReStart: pg = anon->an_page; KASSERT(anon->an_lock == amap->am_lock); - /* page must be resident since parent is wired */ + /* + * The old page must be resident since the parent is + * wired. + */ KASSERT(pg != NULL); /* @@ -750,7 +770,7 @@ ReStart: continue; /* - * if the page is busy then we have to wait for + * If the page is busy, then we have to unlock, wait for * it and then restart. */ if (pg->pg_flags & PG_BUSY) { @@ -760,7 +780,10 @@ ReStart: goto ReStart; } - /* ok, time to do a copy-on-write to a new anon */ + /* + * Perform a copy-on-write. + * First - get a new anon and a page. + */ nanon = uvm_analloc(); if (nanon != NULL) { /* the new anon will share the amap's lock */ @@ -783,18 +806,18 @@ ReStart: } /* - * got it... now we can copy the data and replace anon - * with our new one... + * Copy the data and replace anon with the new one. + * Also, setup its lock (share the with amap's lock). */ - uvm_pagecopy(pg, npg); /* old -> new */ - anon->an_ref--; /* can't drop to zero */ + uvm_pagecopy(pg, npg); + anon->an_ref--; KASSERT(anon->an_ref > 0); - chunk->ac_anon[slot] = nanon; /* replace */ + chunk->ac_anon[slot] = nanon; /* - * drop PG_BUSY on new page ... since we have had its - * owner locked the whole time it can't be - * PG_RELEASED | PG_WANTED. + * Drop PG_BUSY on new page. Since its owner was write + * locked all this time - it cannot be PG_RELEASED or + * PG_WANTED. */ atomic_clearbits_int(&npg->pg_flags, PG_BUSY|PG_FAKE); UVM_PAGE_OWN(npg, NULL); @@ -810,6 +833,8 @@ ReStart: * amap_splitref: split a single reference into two separate references * * => called from uvm_map's clip routines + * => origref's map should be locked + * => origref->ar_amap should be unlocked (we will lock) */ void amap_splitref(struct vm_aref *origref, struct vm_aref *splitref, vaddr_t offset) @@ -824,12 +849,11 @@ amap_splitref(struct vm_aref *origref, struct vm_aref *splitref, vaddr_t offset) amap_lock(amap); - /* now: we have a valid am_mapped array. */ if (amap->am_nslot - origref->ar_pageoff - leftslots <= 0) panic("amap_splitref: map size check failed"); #ifdef UVM_AMAP_PPREF - /* Establish ppref before we add a duplicate reference to the amap. */ + /* Establish ppref before we add a duplicate reference to the amap. */ if (amap->am_ppref == NULL) amap_pp_establish(amap); #endif @@ -844,7 +868,9 @@ amap_splitref(struct vm_aref *origref, struct vm_aref *splitref, vaddr_t offset) #ifdef UVM_AMAP_PPREF /* - * amap_pp_establish: add a ppref array to an amap, if possible + * amap_pp_establish: add a ppref array to an amap, if possible. + * + * => amap should be locked by caller* => amap should be locked by caller */ void amap_pp_establish(struct vm_amap *amap) @@ -854,13 +880,12 @@ amap_pp_establish(struct vm_amap *amap) amap->am_ppref = mallocarray(amap->am_nslot, sizeof(int), M_UVMAMAP, M_NOWAIT|M_ZERO); - /* if we fail then we just won't use ppref for this amap */ if (amap->am_ppref == NULL) { - amap->am_ppref = PPREF_NONE; /* not using it */ + /* Failure - just do not use ppref. */ + amap->am_ppref = PPREF_NONE; return; } - /* init ppref */ pp_setreflen(amap->am_ppref, 0, amap->am_ref, amap->am_nslot); } @@ -868,7 +893,8 @@ amap_pp_establish(struct vm_amap *amap) * amap_pp_adjref: adjust reference count to a part of an amap using the * per-page reference count array. * - * => caller must check that ppref != PPREF_NONE before calling + * => caller must check that ppref != PPREF_NONE before calling. + * => map and amap must be locked. */ void amap_pp_adjref(struct vm_amap *amap, int curslot, vsize_t slotlen, int adjval) @@ -883,8 +909,7 @@ amap_pp_adjref(struct vm_amap *amap, int curslot, vsize_t slotlen, int adjval) prevlcv = 0; /* - * first advance to the correct place in the ppref array, - * fragment if needed. + * Advance to the correct place in the array, fragment if needed. */ for (lcv = 0 ; lcv < curslot ; lcv += len) { pp_getreflen(ppref, lcv, &ref, &len); @@ -898,17 +923,17 @@ amap_pp_adjref(struct vm_amap *amap, int curslot, vsize_t slotlen, int adjval) if (lcv != 0) pp_getreflen(ppref, prevlcv, &prevref, &prevlen); else { - /* Ensure that the "prevref == ref" test below always - * fails, since we're starting from the beginning of - * the ppref array; that is, there is no previous - * chunk. + /* + * Ensure that the "prevref == ref" test below always + * fails, since we are starting from the beginning of + * the ppref array; that is, there is no previous chunk. */ prevref = -1; prevlen = 0; } /* - * now adjust reference counts in range. merge the first + * Now adjust reference counts in range. Merge the first * changed entry with the last unchanged entry if possible. */ if (lcv != curslot) @@ -972,12 +997,19 @@ amap_wiperange_chunk(struct vm_amap *amap, struct vm_amap_chunk *chunk, if (refs == 0) { uvm_anfree(anon); } - } + + /* + * done with this anon, next ...! + */ + + } /* end of 'for' loop */ } /* - * amap_wiperange: wipe out a range of an amap - * [different from amap_wipeout because the amap is kept intact] + * amap_wiperange: wipe out a range of an amap. + * Note: different from amap_wipeout because the amap is kept intact. + * + * => Both map and amap must be locked by caller. */ void amap_wiperange(struct vm_amap *amap, int slotoff, int slots) @@ -991,8 +1023,8 @@ amap_wiperange(struct vm_amap *amap, int slotoff, int slots) endbucket = UVM_AMAP_BUCKET(amap, slotoff + slots - 1); /* - * we can either traverse the amap by am_chunks or by am_buckets - * depending on which is cheaper. decide now. + * We can either traverse the amap by am_chunks or by am_buckets. + * Determine which way is less expensive. */ if (UVM_AMAP_SMALL(amap)) amap_wiperange_chunk(amap, &amap->am_small, slotoff, slots); @@ -1110,7 +1142,9 @@ nextamap: } /* - * amap_lookup: look up a page in an amap + * amap_lookup: look up a page in an amap. + * + * => amap should be locked by caller. */ struct vm_anon * amap_lookup(struct vm_aref *aref, vaddr_t offset) @@ -1131,8 +1165,9 @@ amap_lookup(struct vm_aref *aref, vaddr_t offset) } /* - * amap_lookups: look up a range of pages in an amap + * amap_lookups: look up a range of pages in an amap. * + * => amap should be locked by caller. * => XXXCDC: this interface is biased toward array-based amaps. fix. */ void @@ -1184,9 +1219,10 @@ amap_populate(struct vm_aref *aref, vaddr_t offset) } /* - * amap_add: add (or replace) a page to an amap + * amap_add: add (or replace) a page to an amap. * - * => returns 0 if adding the page was successful or 1 when not. + * => amap should be locked by caller. + * => anon must have the lock associated with this amap. */ int amap_add(struct vm_aref *aref, vaddr_t offset, struct vm_anon *anon, @@ -1228,7 +1264,9 @@ amap_add(struct vm_aref *aref, vaddr_t offset, struct vm_anon *anon, } /* - * amap_unadd: remove a page from an amap + * amap_unadd: remove a page from an amap. + * + * => amap should be locked by caller. */ void amap_unadd(struct vm_aref *aref, vaddr_t offset) diff --git a/sys/uvm/uvm_anon.c b/sys/uvm/uvm_anon.c index 2ae8dacc482..e07838ea57d 100644 --- a/sys/uvm/uvm_anon.c +++ b/sys/uvm/uvm_anon.c @@ -1,4 +1,4 @@ -/* $OpenBSD: uvm_anon.c,v 1.52 2021/03/04 09:00:03 mpi Exp $ */ +/* $OpenBSD: uvm_anon.c,v 1.53 2021/03/20 10:24:21 mpi Exp $ */ /* $NetBSD: uvm_anon.c,v 1.10 2000/11/25 06:27:59 chs Exp $ */ /* @@ -42,9 +42,6 @@ struct pool uvm_anon_pool; -/* - * allocate anons - */ void uvm_anon_init(void) { @@ -54,7 +51,9 @@ uvm_anon_init(void) } /* - * allocate an anon + * uvm_analloc: allocate a new anon. + * + * => anon will have no lock associated. */ struct vm_anon * uvm_analloc(void) @@ -93,12 +92,10 @@ uvm_anfree_list(struct vm_anon *anon, struct pglist *pgl) KASSERT(anon->an_lock != NULL); /* - * if page is busy then we just mark it as released (who ever - * has it busy must check for this when they wake up). if the - * page is not busy then we can free it now. + * If the page is busy, mark it as PG_RELEASED, so + * that uvm_anon_release(9) would release it later. */ if ((pg->pg_flags & PG_BUSY) != 0) { - /* tell them to dump it when done */ atomic_setbits_int(&pg->pg_flags, PG_RELEASED); rw_obj_hold(anon->an_lock); return; @@ -127,13 +124,11 @@ uvm_anfree_list(struct vm_anon *anon, struct pglist *pgl) } anon->an_lock = NULL; - /* free any swap resources. */ - uvm_anon_dropswap(anon); - /* - * now that we've stripped the data areas from the anon, free the anon - * itself! + * Free any swap resources, leave a page replacement hint. */ + uvm_anon_dropswap(anon); + KASSERT(anon->an_page == NULL); KASSERT(anon->an_swslot == 0); @@ -154,9 +149,10 @@ uvm_anwait(void) } /* - * fetch an anon's page. + * uvm_anon_pagein: fetch an anon's page. * - * => returns TRUE if pagein was aborted due to lack of memory. + * => anon must be locked, and is unlocked upon return. + * => returns true if pagein was aborted due to lack of memory. */ boolean_t @@ -168,20 +164,26 @@ uvm_anon_pagein(struct vm_amap *amap, struct vm_anon *anon) KASSERT(rw_write_held(anon->an_lock)); KASSERT(anon->an_lock == amap->am_lock); + /* + * Get the page of the anon. + */ rv = uvmfault_anonget(NULL, amap, anon); switch (rv) { case VM_PAGER_OK: KASSERT(rw_write_held(anon->an_lock)); break; + case VM_PAGER_ERROR: case VM_PAGER_REFAULT: + /* - * nothing more to do on errors. - * VM_PAGER_REFAULT can only mean that the anon was freed, - * so again there's nothing to do. + * Nothing more to do on errors. + * VM_PAGER_REFAULT means that the anon was freed. */ + return FALSE; + default: #ifdef DIAGNOSTIC panic("anon_pagein: uvmfault_anonget -> %d", rv); @@ -191,8 +193,7 @@ uvm_anon_pagein(struct vm_amap *amap, struct vm_anon *anon) } /* - * ok, we've got the page now. - * mark it as dirty, clear its swslot and un-busy it. + * Mark the page as dirty and clear its swslot. */ pg = anon->an_page; if (anon->an_swslot > 0) { @@ -201,7 +202,9 @@ uvm_anon_pagein(struct vm_amap *amap, struct vm_anon *anon) anon->an_swslot = 0; atomic_clearbits_int(&pg->pg_flags, PG_CLEAN); - /* deactivate the page (to put it on a page queue) */ + /* + * Deactivate the page (to put it on a page queue). + */ pmap_clear_reference(pg); pmap_page_protect(pg, PROT_NONE); uvm_lock_pageq(); @@ -213,7 +216,7 @@ uvm_anon_pagein(struct vm_amap *amap, struct vm_anon *anon) } /* - * uvm_anon_dropswap: release any swap resources from this anon. + * uvm_anon_dropswap: release any swap resources from this anon. * * => anon must be locked or have a reference count of 0. */ diff --git a/sys/uvm/uvm_aobj.c b/sys/uvm/uvm_aobj.c index d6f0b8cb597..9c694abc9f2 100644 --- a/sys/uvm/uvm_aobj.c +++ b/sys/uvm/uvm_aobj.c @@ -1,4 +1,4 @@ -/* $OpenBSD: uvm_aobj.c,v 1.91 2021/03/04 09:00:03 mpi Exp $ */ +/* $OpenBSD: uvm_aobj.c,v 1.92 2021/03/20 10:24:21 mpi Exp $ */ /* $NetBSD: uvm_aobj.c,v 1.39 2001/02/18 21:19:08 chs Exp $ */ /* @@ -48,56 +48,45 @@ #include <uvm/uvm.h> /* - * an aobj manages anonymous-memory backed uvm_objects. in addition - * to keeping the list of resident pages, it also keeps a list of - * allocated swap blocks. depending on the size of the aobj this list - * of allocated swap blocks is either stored in an array (small objects) - * or in a hash table (large objects). + * An anonymous UVM object (aobj) manages anonymous-memory. In addition to + * keeping the list of resident pages, it may also keep a list of allocated + * swap blocks. Depending on the size of the object, this list is either + * stored in an array (small objects) or in a hash table (large objects). */ /* - * local structures - */ - -/* - * for hash tables, we break the address space of the aobj into blocks - * of UAO_SWHASH_CLUSTER_SIZE pages. we require the cluster size to - * be a power of two. + * Note: for hash tables, we break the address space of the aobj into blocks + * of UAO_SWHASH_CLUSTER_SIZE pages, which shall be a power of two. */ #define UAO_SWHASH_CLUSTER_SHIFT 4 #define UAO_SWHASH_CLUSTER_SIZE (1 << UAO_SWHASH_CLUSTER_SHIFT) -/* get the "tag" for this page index */ +/* Get the "tag" for this page index. */ #define UAO_SWHASH_ELT_TAG(PAGEIDX) \ ((PAGEIDX) >> UAO_SWHASH_CLUSTER_SHIFT) -/* given an ELT and a page index, find the swap slot */ +/* Given an ELT and a page index, find the swap slot. */ #define UAO_SWHASH_ELT_PAGESLOT_IDX(PAGEIDX) \ ((PAGEIDX) & (UAO_SWHASH_CLUSTER_SIZE - 1)) #define UAO_SWHASH_ELT_PAGESLOT(ELT, PAGEIDX) \ ((ELT)->slots[(PAGEIDX) & (UAO_SWHASH_CLUSTER_SIZE - 1)]) -/* given an ELT, return its pageidx base */ +/* Given an ELT, return its pageidx base. */ #define UAO_SWHASH_ELT_PAGEIDX_BASE(ELT) \ ((ELT)->tag << UAO_SWHASH_CLUSTER_SHIFT) -/* - * the swhash hash function - */ +/* The hash function. */ #define UAO_SWHASH_HASH(AOBJ, PAGEIDX) \ (&(AOBJ)->u_swhash[(((PAGEIDX) >> UAO_SWHASH_CLUSTER_SHIFT) \ & (AOBJ)->u_swhashmask)]) /* - * the swhash threshold determines if we will use an array or a + * The threshold which determines whether we will use an array or a * hash table to store the list of allocated swap blocks. */ - #define UAO_SWHASH_THRESHOLD (UAO_SWHASH_CLUSTER_SIZE * 4) -/* - * the number of buckets in a swhash, with an upper bound - */ +/* The number of buckets in a hash, with an upper bound. */ #define UAO_SWHASH_MAXBUCKETS 256 #define UAO_SWHASH_BUCKETS(pages) \ (min((pages) >> UAO_SWHASH_CLUSTER_SHIFT, UAO_SWHASH_MAXBUCKETS)) @@ -149,14 +138,8 @@ struct uvm_aobj { LIST_ENTRY(uvm_aobj) u_list; /* global list of aobjs */ }; -/* - * uvm_aobj_pool: pool of uvm_aobj structures - */ struct pool uvm_aobj_pool; -/* - * local functions - */ static struct uao_swhash_elt *uao_find_swhash_elt(struct uvm_aobj *, int, boolean_t); static int uao_find_swslot(struct uvm_aobj *, int); @@ -223,17 +206,20 @@ uao_find_swhash_elt(struct uvm_aobj *aobj, int pageidx, boolean_t create) swhash = UAO_SWHASH_HASH(aobj, pageidx); /* first hash to get bucket */ page_tag = UAO_SWHASH_ELT_TAG(pageidx); /* tag to search for */ - /* now search the bucket for the requested tag */ + /* + * now search the bucket for the requested tag + */ LIST_FOREACH(elt, swhash, list) { if (elt->tag == page_tag) return(elt); } - /* fail now if we are not allowed to create a new entry in the bucket */ if (!create) return NULL; - /* allocate a new entry for the bucket and init/insert it in */ + /* + * allocate a new entry for the bucket and init/insert it in + */ elt = pool_get(&uao_swhash_elt_pool, PR_NOWAIT | PR_ZERO); /* * XXX We cannot sleep here as the hash table might disappear @@ -258,11 +244,15 @@ inline static int uao_find_swslot(struct uvm_aobj *aobj, int pageidx) { - /* if noswap flag is set, then we never return a slot */ + /* + * if noswap flag is set, then we never return a slot + */ if (aobj->u_flags & UAO_FLAG_NOSWAP) return(0); - /* if hashing, look in hash table. */ + /* + * if hashing, look in hash table. + */ if (aobj->u_pages > UAO_SWHASH_THRESHOLD) { struct uao_swhash_elt *elt = uao_find_swhash_elt(aobj, pageidx, FALSE); @@ -273,7 +263,9 @@ uao_find_swslot(struct uvm_aobj *aobj, int pageidx) return(0); } - /* otherwise, look in the array */ + /* + * otherwise, look in the array + */ return(aobj->u_swslots[pageidx]); } @@ -281,6 +273,8 @@ uao_find_swslot(struct uvm_aobj *aobj, int pageidx) * uao_set_swslot: set the swap slot for a page in an aobj. * * => setting a slot to zero frees the slot + * => we return the old slot number, or -1 if we failed to allocate + * memory to record the new slot number */ int uao_set_swslot(struct uvm_object *uobj, int pageidx, int slot) @@ -290,18 +284,21 @@ uao_set_swslot(struct uvm_object *uobj, int pageidx, int slot) KERNEL_ASSERT_LOCKED(); - /* if noswap flag is set, then we can't set a slot */ + /* + * if noswap flag is set, then we can't set a slot + */ if (aobj->u_flags & UAO_FLAG_NOSWAP) { if (slot == 0) return(0); /* a clear is ok */ /* but a set is not */ printf("uao_set_swslot: uobj = %p\n", uobj); - panic("uao_set_swslot: attempt to set a slot" - " on a NOSWAP object"); + panic("uao_set_swslot: attempt to set a slot on a NOSWAP object"); } - /* are we using a hash table? if so, add it in the hash. */ + /* + * are we using a hash table? if so, add it in the hash. + */ if (aobj->u_pages > UAO_SWHASH_THRESHOLD) { /* * Avoid allocating an entry just to free it again if @@ -322,12 +319,11 @@ uao_set_swslot(struct uvm_object *uobj, int pageidx, int slot) * now adjust the elt's reference counter and free it if we've * dropped it to zero. */ - /* an allocation? */ if (slot) { if (oldslot == 0) elt->count++; - } else { /* freeing slot ... */ - if (oldslot) /* to be safe */ + } else { + if (oldslot) elt->count--; if (elt->count == 0) { @@ -335,7 +331,7 @@ uao_set_swslot(struct uvm_object *uobj, int pageidx, int slot) pool_put(&uao_swhash_elt_pool, elt); } } - } else { + } else { /* we are using an array */ oldslot = aobj->u_swslots[pageidx]; aobj->u_swslots[pageidx] = slot; @@ -393,12 +389,15 @@ uao_free(struct uvm_aobj *aobj) } else { int i; - /* free the array */ + /* + * free the array + */ for (i = 0; i < aobj->u_pages; i++) { int slot = aobj->u_swslots[i]; if (slot) { uvm_swap_free(slot, 1); + /* this page is no longer only in swap. */ atomic_dec_int(&uvmexp.swpgonly); } @@ -406,7 +405,9 @@ uao_free(struct uvm_aobj *aobj) free(aobj->u_swslots, M_UVMAOBJ, aobj->u_pages * sizeof(int)); } - /* finally free the aobj itself */ + /* + * finally free the aobj itself + */ pool_put(&uvm_aobj_pool, aobj); } @@ -702,36 +703,39 @@ uao_grow(struct uvm_object *uobj, int pages) struct uvm_object * uao_create(vsize_t size, int flags) { - static struct uvm_aobj kernel_object_store; /* home of kernel_object */ - static int kobj_alloced = 0; /* not allocated yet */ + static struct uvm_aobj kernel_object_store; + static int kobj_alloced = 0; int pages = round_page(size) >> PAGE_SHIFT; int refs = UVM_OBJ_KERN; int mflags; struct uvm_aobj *aobj; - /* malloc a new aobj unless we are asked for the kernel object */ - if (flags & UAO_FLAG_KERNOBJ) { /* want kernel object? */ + /* + * Allocate a new aobj, unless kernel object is requested. + */ + if (flags & UAO_FLAG_KERNOBJ) { if (kobj_alloced) panic("uao_create: kernel object already allocated"); aobj = &kernel_object_store; aobj->u_pages = pages; - aobj->u_flags = UAO_FLAG_NOSWAP; /* no swap to start */ - /* we are special, we never die */ + aobj->u_flags = UAO_FLAG_NOSWAP; kobj_alloced = UAO_FLAG_KERNOBJ; } else if (flags & UAO_FLAG_KERNSWAP) { aobj = &kernel_object_store; if (kobj_alloced != UAO_FLAG_KERNOBJ) panic("uao_create: asked to enable swap on kernel object"); kobj_alloced = UAO_FLAG_KERNSWAP; - } else { /* normal object */ + } else { aobj = pool_get(&uvm_aobj_pool, PR_WAITOK); aobj->u_pages = pages; - aobj->u_flags = 0; /* normal object */ - refs = 1; /* normal object so 1 ref */ + aobj->u_flags = 0; + refs = 1; } - /* allocate hash/array if necessary */ + /* + * allocate hash/array if necessary + */ if (flags == 0 || (flags & (UAO_FLAG_KERNSWAP | UAO_FLAG_CANFAIL))) { if (flags) mflags = M_NOWAIT; @@ -768,9 +772,14 @@ uao_create(vsize_t size, int flags) } } + /* + * Initialise UVM object. + */ uvm_objinit(&aobj->u_obj, &aobj_pager, refs); - /* now that aobj is ready, add it to the global list */ + /* + * now that aobj is ready, add it to the global list + */ mtx_enter(&uao_list_lock); LIST_INSERT_HEAD(&uao_list, aobj, u_list); mtx_leave(&uao_list_lock); @@ -799,7 +808,7 @@ uao_init(void) } /* - * uao_reference: add a ref to an aobj + * uao_reference: hold a reference to an anonymous UVM object. */ void uao_reference(struct uvm_object *uobj) @@ -808,23 +817,20 @@ uao_reference(struct uvm_object *uobj) uao_reference_locked(uobj); } -/* - * uao_reference_locked: add a ref to an aobj - */ void uao_reference_locked(struct uvm_object *uobj) { - /* kernel_object already has plenty of references, leave it alone. */ + /* Kernel object is persistent. */ if (UVM_OBJ_IS_KERN_OBJECT(uobj)) return; - uobj->uo_refs++; /* bump! */ + uobj->uo_refs++; } /* - * uao_detach: drop a reference to an aobj + * uao_detach: drop a reference to an anonymous UVM object. */ void uao_detach(struct uvm_object *uobj) @@ -845,26 +851,34 @@ uao_detach_locked(struct uvm_object *uobj) struct uvm_aobj *aobj = (struct uvm_aobj *)uobj; struct vm_page *pg; - /* detaching from kernel_object is a noop. */ + /* + * Detaching from kernel_object is a NOP. + */ if (UVM_OBJ_IS_KERN_OBJECT(uobj)) { return; } - uobj->uo_refs--; /* drop ref! */ - if (uobj->uo_refs) { /* still more refs? */ + /* + * Drop the reference. If it was the last one, destroy the object. + */ + uobj->uo_refs--; + if (uobj->uo_refs) { return; } - /* remove the aobj from the global list. */ + /* + * Remove the aobj from the global list. + */ mtx_enter(&uao_list_lock); LIST_REMOVE(aobj, u_list); mtx_leave(&uao_list_lock); /* - * Free all pages left in the object. If they're busy, wait - * for them to become available before we kill it. - * Release swap resources then free the page. - */ + * Free all the pages left in the aobj. For each page, when the + * page is no longer busy (and thus after any disk I/O that it is + * involved in is complete), release any swap resources and free + * the page itself. + */ uvm_lock_pageq(); while((pg = RBT_ROOT(uvm_objtree, &uobj->memt)) != NULL) { if (pg->pg_flags & PG_BUSY) { @@ -880,12 +894,14 @@ uao_detach_locked(struct uvm_object *uobj) } uvm_unlock_pageq(); - /* finally, free the rest. */ + /* + * Finally, free the anonymous UVM object itself. + */ uao_free(aobj); } /* - * uao_flush: "flush" pages out of a uvm object + * uao_flush: flush pages out of a uvm object * * => if PGO_CLEANIT is not set, then we will not block. * => if PGO_ALLPAGE is set, then all pages in the object are valid targets @@ -958,15 +974,11 @@ uao_flush(struct uvm_object *uobj, voff_t start, voff_t stop, int flags) /* FALLTHROUGH */ case PGO_DEACTIVATE: deactivate_it: - /* skip the page if it's wired */ if (pp->wire_count != 0) continue; uvm_lock_pageq(); - /* zap all mappings for the page. */ pmap_page_protect(pp, PROT_NONE); - - /* ...and deactivate the page. */ uvm_pagedeactivate(pp); uvm_unlock_pageq(); @@ -983,9 +995,16 @@ uao_flush(struct uvm_object *uobj, voff_t start, voff_t stop, int flags) if (pp->wire_count != 0) continue; - /* zap all mappings for the page. */ + /* + * free the swap slot and the page. + */ pmap_page_protect(pp, PROT_NONE); + /* + * freeing swapslot here is not strictly necessary. + * however, leaving it here doesn't save much + * because we need to update swap accounting anyway. + */ uao_dropswap(uobj, pp->offset >> PAGE_SHIFT); uvm_lock_pageq(); uvm_pagefree(pp); @@ -1029,12 +1048,17 @@ uao_get(struct uvm_object *uobj, voff_t offset, struct vm_page **pps, KERNEL_ASSERT_LOCKED(); - /* get number of pages */ + /* + * get number of pages + */ maxpages = *npagesp; - /* step 1: handled the case where fault data structures are locked. */ if (flags & PGO_LOCKED) { - /* step 1a: get pages that are already resident. */ + /* + * step 1a: get pages that are already resident. only do + * this if the data structures are locked (i.e. the first + * time through). + */ done = TRUE; /* be optimistic */ gotpages = 0; /* # of pages we got so far */ @@ -1065,7 +1089,9 @@ uao_get(struct uvm_object *uobj, voff_t offset, struct vm_page **pps, } } - /* to be useful must get a non-busy page */ + /* + * to be useful must get a non-busy page + */ if (ptmp == NULL || (ptmp->pg_flags & PG_BUSY) != 0) { if (lcv == centeridx || @@ -1076,10 +1102,8 @@ uao_get(struct uvm_object *uobj, voff_t offset, struct vm_page **pps, } /* - * useful page: busy it and plug it in our - * result array + * useful page: plug it in our result array */ - /* caller must un-busy this page */ atomic_setbits_int(&ptmp->pg_flags, PG_BUSY); UVM_PAGE_OWN(ptmp, "uao_get1"); pps[lcv] = ptmp; @@ -1146,8 +1170,7 @@ uao_get(struct uvm_object *uobj, voff_t offset, struct vm_page **pps, /* out of RAM? */ if (ptmp == NULL) { uvm_wait("uao_getpage"); - /* goto top of pps while loop */ - continue; + continue; } /* @@ -1169,12 +1192,10 @@ uao_get(struct uvm_object *uobj, voff_t offset, struct vm_page **pps, tsleep_nsec(ptmp, PVM, "uao_get", INFSLP); continue; /* goto top of pps while loop */ } - - /* - * if we get here then the page has become resident and - * unbusy between steps 1 and 2. we busy it now (so we - * own it) and set pps[lcv] (so that we exit the while - * loop). + + /* + * if we get here then the page is resident and + * unbusy. we busy it now (so we own it). */ /* we own it, caller must un-busy */ atomic_setbits_int(&ptmp->pg_flags, PG_BUSY); @@ -1200,10 +1221,14 @@ uao_get(struct uvm_object *uobj, voff_t offset, struct vm_page **pps, /* page hasn't existed before, just zero it. */ uvm_pagezero(ptmp); } else { - /* page in the swapped-out page. */ + /* + * page in the swapped-out page. + */ rv = uvm_swap_get(ptmp, swslot, PGO_SYNCIO); - /* I/O done. check for errors. */ + /* + * I/O done. check for errors. + */ if (rv != VM_PAGER_OK) { /* * remove the swap slot from the aobj @@ -1228,18 +1253,16 @@ uao_get(struct uvm_object *uobj, voff_t offset, struct vm_page **pps, } } - /* + /* * we got the page! clear the fake flag (indicates valid * data now in page) and plug into our result array. note - * that page is still busy. + * that page is still busy. * * it is the callers job to: * => check if the page is released * => unbusy the page * => activate the page */ - - /* data is valid ... */ atomic_clearbits_int(&ptmp->pg_flags, PG_FAKE); pmap_clear_modify(ptmp); /* ... and clean */ pps[lcv] = ptmp; @@ -1274,7 +1297,9 @@ uao_swap_off(int startslot, int endslot) { struct uvm_aobj *aobj, *nextaobj, *prevaobj = NULL; - /* walk the list of all aobjs. */ + /* + * Walk the list of all anonymous UVM objects. + */ mtx_enter(&uao_list_lock); for (aobj = LIST_FIRST(&uao_list); @@ -1325,7 +1350,9 @@ uao_swap_off(int startslot, int endslot) prevaobj = aobj; } - /* done with traversal, unlock the list */ + /* + * done with traversal, unlock the list + */ mtx_leave(&uao_list_lock); if (prevaobj) { uao_detach_locked(&prevaobj->u_obj); @@ -1357,8 +1384,10 @@ restart: for (i = 0; i < UAO_SWHASH_CLUSTER_SIZE; i++) { int slot = elt->slots[i]; - /* if slot isn't in range, skip it. */ - if (slot < startslot || + /* + * if the slot isn't in range, skip it. + */ + if (slot < startslot || slot >= endslot) { continue; } @@ -1384,12 +1413,16 @@ restart: for (i = 0; i < aobj->u_pages; i++) { int slot = aobj->u_swslots[i]; - /* if the slot isn't in range, skip it */ + /* + * if the slot isn't in range, skip it + */ if (slot < startslot || slot >= endslot) { continue; } - /* process the page. */ + /* + * process the page. + */ rv = uao_pagein_page(aobj, i); if (rv) { return rv; @@ -1401,8 +1434,9 @@ restart: } /* - * page in a page from an aobj. used for swap_off. - * returns TRUE if pagein was aborted due to lack of memory. + * uao_pagein_page: page in a single page from an anonymous UVM object. + * + * => Returns TRUE if pagein was aborted due to lack of memory. */ static boolean_t uao_pagein_page(struct uvm_aobj *aobj, int pageidx) @@ -1438,7 +1472,9 @@ uao_pagein_page(struct uvm_aobj *aobj, int pageidx) atomic_clearbits_int(&pg->pg_flags, PG_BUSY|PG_CLEAN|PG_FAKE); UVM_PAGE_OWN(pg, NULL); - /* deactivate the page (to put it on a page queue). */ + /* + * deactivate the page (to put it on a page queue). + */ pmap_clear_reference(pg); uvm_lock_pageq(); uvm_pagedeactivate(pg); diff --git a/sys/uvm/uvm_device.c b/sys/uvm/uvm_device.c index 33452c43573..81dccae9363 100644 --- a/sys/uvm/uvm_device.c +++ b/sys/uvm/uvm_device.c @@ -1,4 +1,4 @@ -/* $OpenBSD: uvm_device.c,v 1.60 2020/11/06 11:52:39 mpi Exp $ */ +/* $OpenBSD: uvm_device.c,v 1.61 2021/03/20 10:24:21 mpi Exp $ */ /* $NetBSD: uvm_device.c,v 1.30 2000/11/25 06:27:59 chs Exp $ */ /* @@ -79,6 +79,11 @@ const struct uvm_pagerops uvm_deviceops = { }; /* + * the ops! + */ + + +/* * udv_attach * * get a VM object that is associated with a device. allocate a new @@ -97,14 +102,18 @@ udv_attach(dev_t device, vm_prot_t accessprot, voff_t off, vsize_t size) struct uvm_object *obj; #endif - /* before we do anything, ensure this device supports mmap */ + /* + * before we do anything, ensure this device supports mmap + */ mapfn = cdevsw[major(device)].d_mmap; if (mapfn == NULL || mapfn == (paddr_t (*)(dev_t, off_t, int)) enodev || mapfn == (paddr_t (*)(dev_t, off_t, int)) nullop) return(NULL); - /* Negative offsets on the object are not allowed. */ + /* + * Negative offsets on the object are not allowed. + */ if (off < 0) return(NULL); @@ -126,16 +135,22 @@ udv_attach(dev_t device, vm_prot_t accessprot, voff_t off, vsize_t size) off += PAGE_SIZE; size -= PAGE_SIZE; } - /* keep looping until we get it */ + /* + * keep looping until we get it + */ for (;;) { - /* first, attempt to find it on the main list */ + /* + * first, attempt to find it on the main list + */ mtx_enter(&udv_lock); LIST_FOREACH(lcv, &udv_list, u_list) { if (device == lcv->u_device) break; } - /* got it on main list. put a hold on it and unlock udv_lock. */ + /* + * got it on main list. put a hold on it and unlock udv_lock. + */ if (lcv) { /* * if someone else has a hold on it, sleep and start @@ -153,7 +168,9 @@ udv_attach(dev_t device, vm_prot_t accessprot, voff_t off, vsize_t size) lcv->u_flags |= UVM_DEVICE_HOLD; mtx_leave(&udv_lock); - /* bump reference count, unhold, return. */ + /* + * bump reference count, unhold, return. + */ lcv->u_obj.uo_refs++; mtx_enter(&udv_lock); @@ -164,7 +181,9 @@ udv_attach(dev_t device, vm_prot_t accessprot, voff_t off, vsize_t size) return(&lcv->u_obj); } - /* did not find it on main list. need to malloc a new one. */ + /* + * Did not find it on main list. Need to allocate a new one. + */ mtx_leave(&udv_lock); /* NOTE: we could sleep in the following malloc() */ udv = malloc(sizeof(*udv), M_TEMP, M_WAITOK); @@ -229,7 +248,9 @@ udv_detach(struct uvm_object *uobj) KERNEL_ASSERT_LOCKED(); - /* loop until done */ + /* + * loop until done + */ again: if (uobj->uo_refs > 1) { uobj->uo_refs--; @@ -237,7 +258,9 @@ again: } KASSERT(uobj->uo_npages == 0 && RBT_EMPTY(uvm_objtree, &uobj->memt)); - /* is it being held? if so, wait until others are done. */ + /* + * is it being held? if so, wait until others are done. + */ mtx_enter(&udv_lock); if (udv->u_flags & UVM_DEVICE_HOLD) { udv->u_flags |= UVM_DEVICE_WANTED; @@ -250,7 +273,9 @@ again: goto again; } - /* got it! nuke it now. */ + /* + * got it! nuke it now. + */ LIST_REMOVE(udv, u_list); if (udv->u_flags & UVM_DEVICE_WANTED) wakeup(udv); @@ -310,7 +335,9 @@ udv_fault(struct uvm_faultinfo *ufi, vaddr_t vaddr, vm_page_t *pps, int npages, return(VM_PAGER_ERROR); } - /* get device map function. */ + /* + * get device map function. + */ device = udv->u_device; mapfn = cdevsw[major(device)].d_mmap; @@ -325,7 +352,9 @@ udv_fault(struct uvm_faultinfo *ufi, vaddr_t vaddr, vm_page_t *pps, int npages, /* pmap va = vaddr (virtual address of pps[0]) */ curr_va = vaddr; - /* loop over the page range entering in as needed */ + /* + * loop over the page range entering in as needed + */ retval = VM_PAGER_OK; for (lcv = 0 ; lcv < npages ; lcv++, curr_offset += PAGE_SIZE, curr_va += PAGE_SIZE) { diff --git a/sys/uvm/uvm_fault.c b/sys/uvm/uvm_fault.c index bf1bfcd32a6..a4d0337d469 100644 --- a/sys/uvm/uvm_fault.c +++ b/sys/uvm/uvm_fault.c @@ -1,4 +1,4 @@ -/* $OpenBSD: uvm_fault.c,v 1.118 2021/03/12 14:15:49 jsg Exp $ */ +/* $OpenBSD: uvm_fault.c,v 1.119 2021/03/20 10:24:21 mpi Exp $ */ /* $NetBSD: uvm_fault.c,v 1.51 2000/08/06 00:22:53 thorpej Exp $ */ /* @@ -230,24 +230,32 @@ static void uvmfault_amapcopy(struct uvm_faultinfo *ufi) { for (;;) { - /* no mapping? give up. */ + /* + * no mapping? give up. + */ if (uvmfault_lookup(ufi, TRUE) == FALSE) return; - /* copy if needed. */ + /* + * copy if needed. + */ if (UVM_ET_ISNEEDSCOPY(ufi->entry)) amap_copy(ufi->map, ufi->entry, M_NOWAIT, UVM_ET_ISSTACK(ufi->entry) ? FALSE : TRUE, ufi->orig_rvaddr, ufi->orig_rvaddr + 1); - /* didn't work? must be out of RAM. sleep. */ + /* + * didn't work? must be out of RAM. unlock and sleep. + */ if (UVM_ET_ISNEEDSCOPY(ufi->entry)) { uvmfault_unlockmaps(ufi, TRUE); uvm_wait("fltamapcopy"); continue; } - /* got it! */ + /* + * got it! unlock and return. + */ uvmfault_unlockmaps(ufi, TRUE); return; } @@ -341,11 +349,11 @@ uvmfault_anonget(struct uvm_faultinfo *ufi, struct vm_amap *amap, uvmfault_unlockall(ufi, amap, NULL); /* - * we are passing a PG_BUSY+PG_FAKE+PG_CLEAN - * page into the uvm_swap_get function with - * all data structures unlocked. note that - * it is ok to read an_swslot here because - * we hold PG_BUSY on the page. + * Pass a PG_BUSY+PG_FAKE+PG_CLEAN page into + * the uvm_swap_get() function with all data + * structures unlocked. Note that it is OK + * to read an_swslot here, because we hold + * PG_BUSY on the page. */ counters_inc(uvmexp_counters, pageins); error = uvm_swap_get(pg, anon->an_swslot, @@ -393,6 +401,9 @@ uvmfault_anonget(struct uvm_faultinfo *ufi, struct vm_amap *amap, if (pg->pg_flags & PG_RELEASED) { pmap_page_protect(pg, PROT_NONE); KASSERT(anon->an_ref == 0); + /* + * Released while we had unlocked amap. + */ if (locked) uvmfault_unlockall(ufi, NULL, NULL); uvm_anon_release(anon); /* frees page for us */ @@ -418,7 +429,7 @@ uvmfault_anonget(struct uvm_faultinfo *ufi, struct vm_amap *amap, /* * Note: page was never !PG_BUSY, so it * cannot be mapped and thus no need to - * pmap_page_protect it... + * pmap_page_protect() it. */ uvm_lock_pageq(); uvm_pagefree(pg); @@ -432,8 +443,7 @@ uvmfault_anonget(struct uvm_faultinfo *ufi, struct vm_amap *amap, } /* - * must be OK, clear modify (already PG_CLEAN) - * and activate + * We have successfully read the page, activate it. */ pmap_clear_modify(pg); uvm_lock_pageq(); @@ -776,7 +786,9 @@ uvm_fault_check(struct uvm_faultinfo *ufi, struct uvm_faultctx *flt, if (amap) uvmfault_anonflush(*ranons, nback); - /* flush object? */ + /* + * flush object? + */ if (uobj) { voff_t uoff; @@ -843,7 +855,7 @@ uvm_fault_upper_lookup(struct uvm_faultinfo *ufi, } /* - * unmapped or center page. check if any anon at this level. + * unmapped or center page. check if any anon at this level. */ if (amap == NULL || anons[lcv] == NULL) { pages[lcv] = NULL; @@ -903,6 +915,7 @@ uvm_fault_upper(struct uvm_faultinfo *ufi, struct uvm_faultctx *flt, struct vm_page *pg = NULL; int error, ret; + /* locked: maps(read), amap, anon */ KASSERT(rw_write_held(amap->am_lock)); KASSERT(anon->an_lock == amap->am_lock); @@ -1037,7 +1050,9 @@ uvm_fault_upper(struct uvm_faultinfo *ufi, struct uvm_faultctx *flt, return ERESTART; } - /* ... update the page queues. */ + /* + * ... update the page queues. + */ uvm_lock_pageq(); if (fault_type == VM_FAULT_WIRE) { @@ -1181,6 +1196,7 @@ uvm_fault_lower(struct uvm_faultinfo *ufi, struct uvm_faultctx *flt, * (PGO_LOCKED). */ if (uobj == NULL) { + /* zero fill; don't care neighbor pages */ uobjpage = NULL; } else { uobjpage = uvm_fault_lower_lookup(ufi, flt, pages); @@ -1236,7 +1252,9 @@ uvm_fault_lower(struct uvm_faultinfo *ufi, struct uvm_faultctx *flt, 0, flt->access_type & MASK(ufi->entry), ufi->entry->advice, PGO_SYNCIO); - /* recover from I/O */ + /* + * recover from I/O + */ if (result != VM_PAGER_OK) { KASSERT(result != VM_PAGER_PEND); @@ -1343,7 +1361,9 @@ uvm_fault_lower(struct uvm_faultinfo *ufi, struct uvm_faultctx *flt, * out of memory resources? */ if (anon == NULL || pg == NULL) { - /* arg! must unbusy our page and fail or sleep. */ + /* + * arg! must unbusy our page and fail or sleep. + */ if (uobjpage != PGO_DONTCARE) { uvm_lock_pageq(); uvm_pageactivate(uobjpage); @@ -1378,7 +1398,9 @@ uvm_fault_lower(struct uvm_faultinfo *ufi, struct uvm_faultctx *flt, return ERESTART; } - /* fill in the data */ + /* + * fill in the data + */ if (uobjpage != PGO_DONTCARE) { counters_inc(uvmexp_counters, flt_prcopy); /* copy page [pg now dirty] */ @@ -1559,9 +1581,12 @@ uvm_fault_unwire_locked(vm_map_t map, vaddr_t start, vaddr_t end) * the PAs from the pmap. we also lock out the page daemon so that * we can call uvm_pageunwire. */ + uvm_lock_pageq(); - /* find the beginning map entry for the region. */ + /* + * find the beginning map entry for the region. + */ KASSERT(start >= vm_map_min(map) && end <= vm_map_max(map)); if (uvm_map_lookup_entry(map, start, &entry) == FALSE) panic("uvm_fault_unwire_locked: address not in map"); @@ -1570,7 +1595,9 @@ uvm_fault_unwire_locked(vm_map_t map, vaddr_t start, vaddr_t end) if (pmap_extract(pmap, va, &pa) == FALSE) continue; - /* find the map entry for the current address. */ + /* + * find the map entry for the current address. + */ KASSERT(va >= entry->start); while (va >= entry->end) { next = RBT_NEXT(uvm_map_addr, entry); @@ -1578,7 +1605,9 @@ uvm_fault_unwire_locked(vm_map_t map, vaddr_t start, vaddr_t end) entry = next; } - /* if the entry is no longer wired, tell the pmap. */ + /* + * if the entry is no longer wired, tell the pmap. + */ if (VM_MAPENT_ISWIRED(entry) == 0) pmap_unwire(pmap, va); @@ -1646,7 +1675,9 @@ uvmfault_lookup(struct uvm_faultinfo *ufi, boolean_t write_lock) { vm_map_t tmpmap; - /* init ufi values for lookup. */ + /* + * init ufi values for lookup. + */ ufi->map = ufi->orig_map; ufi->size = ufi->orig_size; @@ -1688,11 +1719,14 @@ uvmfault_lookup(struct uvm_faultinfo *ufi, boolean_t write_lock) continue; } - /* got it! */ + /* + * got it! + */ ufi->mapv = ufi->map->timestamp; return(TRUE); - } + } /* while loop */ + /*NOTREACHED*/ } diff --git a/sys/uvm/uvm_init.c b/sys/uvm/uvm_init.c index 7342321d5a3..b15c4d7f9af 100644 --- a/sys/uvm/uvm_init.c +++ b/sys/uvm/uvm_init.c @@ -1,4 +1,4 @@ -/* $OpenBSD: uvm_init.c,v 1.41 2020/12/28 14:01:23 mpi Exp $ */ +/* $OpenBSD: uvm_init.c,v 1.42 2021/03/20 10:24:21 mpi Exp $ */ /* $NetBSD: uvm_init.c,v 1.14 2000/06/27 17:29:23 mrg Exp $ */ /* @@ -69,41 +69,41 @@ vaddr_t vm_min_kernel_address; /* * uvm_init: init the VM system. called from kern/init_main.c. */ + void uvm_init(void) { vaddr_t kvm_start, kvm_end; - /* step 0: ensure that the hardware set the page size */ + /* + * Ensure that the hardware set the page size. + */ if (uvmexp.pagesize == 0) { panic("uvm_init: page size not set"); } - - /* step 1: set up stats. */ averunnable.fscale = FSCALE; /* - * step 2: init the page sub-system. this includes allocating the - * vm_page structures, and setting up all the page queues (and - * locks). available memory will be put in the "free" queue. - * kvm_start and kvm_end will be set to the area of kernel virtual - * memory which is available for general use. + * Init the page sub-system. This includes allocating the vm_page + * structures, and setting up all the page queues (and locks). + * Available memory will be put in the "free" queue, kvm_start and + * kvm_end will be set to the area of kernel virtual memory which + * is available for general use. */ uvm_page_init(&kvm_start, &kvm_end); /* - * step 3: init the map sub-system. allocates the static pool of - * vm_map_entry structures that are used for "special" kernel maps - * (e.g. kernel_map, kmem_map, etc...). + * Init the map sub-system. + * + * Allocates the static pool of vm_map_entry structures that are + * used for "special" kernel maps (e.g. kernel_map, kmem_map, etc...). */ uvm_map_init(); /* - * step 4: setup the kernel's virtual memory data structures. this - * includes setting up the kernel_map/kernel_object and the kmem_map/ - * kmem_object. + * Setup the kernel's virtual memory data structures. This includes + * setting up the kernel_map/kernel_object. */ - uvm_km_init(vm_min_kernel_address, kvm_start, kvm_end); /* @@ -112,7 +112,7 @@ uvm_init(void) uvmfault_init(); /* - * step 5: init the pmap module. the pmap module is free to allocate + * Init the pmap module. The pmap module is free to allocate * memory for its private use (e.g. pvlists). */ pmap_init(); @@ -123,8 +123,8 @@ uvm_init(void) uvm_km_page_init(); /* - * step 7: init the kernel memory allocator. after this call the - * kernel memory allocator (malloc) can be used. + * Make kernel memory allocators ready for use. + * After this call the malloc memory allocator can be used. */ kmeminit(); @@ -134,7 +134,7 @@ uvm_init(void) dma_alloc_init(); /* - * step 8: init all pagers and the pager_map. + * Init all pagers and the pager_map. */ uvm_pager_init(); @@ -172,7 +172,7 @@ uvm_init(void) panic("uvm_init: cannot reserve dead beef @0x%x", DEADBEEF1); #endif /* - * init anonymous memory systems + * Init anonymous memory systems. */ uvm_anon_init(); diff --git a/sys/uvm/uvm_io.c b/sys/uvm/uvm_io.c index 6ebc1cb9c0b..96a9ba543b1 100644 --- a/sys/uvm/uvm_io.c +++ b/sys/uvm/uvm_io.c @@ -1,4 +1,4 @@ -/* $OpenBSD: uvm_io.c,v 1.26 2016/01/09 11:34:27 kettenis Exp $ */ +/* $OpenBSD: uvm_io.c,v 1.27 2021/03/20 10:24:21 mpi Exp $ */ /* $NetBSD: uvm_io.c,v 1.12 2000/06/27 17:29:23 mrg Exp $ */ /* @@ -87,9 +87,14 @@ uvm_io(vm_map_t map, struct uio *uio, int flags) if (flags & UVM_IO_FIXPROT) extractflags |= UVM_EXTRACT_FIXPROT; - /* step 1: main loop... while we've got data to move */ + /* + * step 1: main loop... while we've got data to move + */ for (/*null*/; togo > 0 ; pageoffset = 0) { - /* step 2: extract mappings from the map into kernel_map */ + + /* + * step 2: extract mappings from the map into kernel_map + */ error = uvm_map_extract(map, baseva, chunksz, &kva, extractflags); if (error) { @@ -105,7 +110,9 @@ uvm_io(vm_map_t map, struct uio *uio, int flags) break; } - /* step 3: move a chunk of data */ + /* + * step 3: move a chunk of data + */ sz = chunksz - pageoffset; if (sz > togo) sz = togo; @@ -113,7 +120,10 @@ uvm_io(vm_map_t map, struct uio *uio, int flags) togo -= sz; baseva += chunksz; - /* step 4: unmap the area of kernel memory */ + + /* + * step 4: unmap the area of kernel memory + */ vm_map_lock(kernel_map); TAILQ_INIT(&dead_entries); uvm_unmap_remove(kernel_map, kva, kva+chunksz, @@ -121,10 +131,6 @@ uvm_io(vm_map_t map, struct uio *uio, int flags) vm_map_unlock(kernel_map); uvm_unmap_detach(&dead_entries, AMAP_REFALL); - /* - * We defer checking the error return from uiomove until - * here so that we won't leak memory. - */ if (error) break; } diff --git a/sys/uvm/uvm_km.c b/sys/uvm/uvm_km.c index aea66f7ec26..af6e17963a5 100644 --- a/sys/uvm/uvm_km.c +++ b/sys/uvm/uvm_km.c @@ -1,4 +1,4 @@ -/* $OpenBSD: uvm_km.c,v 1.141 2021/03/12 14:15:49 jsg Exp $ */ +/* $OpenBSD: uvm_km.c,v 1.142 2021/03/20 10:24:21 mpi Exp $ */ /* $NetBSD: uvm_km.c,v 1.42 2001/01/14 02:10:01 thorpej Exp $ */ /* @@ -223,7 +223,9 @@ uvm_km_suballoc(struct vm_map *map, vaddr_t *min, vaddr_t *max, vsize_t size, uvm_map_setup(submap, vm_map_pmap(map), *min, *max, flags); } - /* now let uvm_map_submap plug in it... */ + /* + * now let uvm_map_submap plug in it... + */ if (uvm_map_submap(map, *min, *max, submap) != 0) panic("uvm_km_suballoc: submap allocation failed"); @@ -541,8 +543,9 @@ uvm_km_valloc_align(struct vm_map *map, vsize_t size, vsize_t align, int flags) size = round_page(size); kva = vm_map_min(map); /* hint */ - /* allocate some virtual space, demand filled by kernel_object. */ - + /* + * allocate some virtual space. will be demand filled by kernel_object. + */ if (__predict_false(uvm_map(map, &kva, size, uvm.kernel_object, UVM_UNKNOWN_OFFSET, align, UVM_MAPFLAG(PROT_READ | PROT_WRITE, PROT_READ | PROT_WRITE, |