diff options
| -rw-r--r-- | sys/uvm/uvm.h | 17 | ||||
| -rw-r--r-- | sys/uvm/uvm_anon.c | 6 | ||||
| -rw-r--r-- | sys/uvm/uvm_aobj.c | 213 | ||||
| -rw-r--r-- | sys/uvm/uvm_extern.h | 15 | ||||
| -rw-r--r-- | sys/uvm/uvm_fault.c | 10 | ||||
| -rw-r--r-- | sys/uvm/uvm_glue.c | 31 | ||||
| -rw-r--r-- | sys/uvm/uvm_km.c | 242 | ||||
| -rw-r--r-- | sys/uvm/uvm_map.c | 118 | ||||
| -rw-r--r-- | sys/uvm/uvm_meter.c | 4 | ||||
| -rw-r--r-- | sys/uvm/uvm_mmap.c | 28 | ||||
| -rw-r--r-- | sys/uvm/uvm_vnode.c | 4 |
11 files changed, 374 insertions, 314 deletions
diff --git a/sys/uvm/uvm.h b/sys/uvm/uvm.h index 40b9f292229..e5410a9e52c 100644 --- a/sys/uvm/uvm.h +++ b/sys/uvm/uvm.h @@ -1,5 +1,5 @@ -/* $OpenBSD: uvm.h,v 1.9 2001/04/10 06:59:12 niklas Exp $ */ -/* $NetBSD: uvm.h,v 1.16 1999/06/21 17:25:11 thorpej Exp $ */ +/* $OpenBSD: uvm.h,v 1.10 2001/06/23 19:24:33 smart Exp $ */ +/* $NetBSD: uvm.h,v 1.17 1999/07/22 22:58:38 thorpej Exp $ */ /* * @@ -147,16 +147,17 @@ UVMHIST_DECL(pdhist); /* * UVM_UNLOCK_AND_WAIT: atomic unlock+wait... front end for the - * (poorly named) thread_sleep_msg function. + * uvm_sleep() function. */ #if defined(MULTIPROCESSOR) || defined(LOCKDEBUG) -#define UVM_UNLOCK_AND_WAIT(event,lock,intr,msg, timo) \ - thread_sleep_msg(event,lock,intr,msg, timo) +#define UVM_UNLOCK_AND_WAIT(event, lock, intr ,msg, timo) \ + uvm_sleep(event, lock, intr, msg, timo) + #else -#define UVM_UNLOCK_AND_WAIT(event,lock,intr,msg, timo) \ - thread_sleep_msg(event,NULL,intr,msg, timo) -#endif +#define UVM_UNLOCK_AND_WAIT(event, lock, intr, msg, timo) \ + uvm_sleep(event, NULL, intr, msg, timo) +#endif /* MULTIPROCESSOR || LOCKDEBUG */ /* * UVM_PAGE_OWN: track page ownership (only if UVM_PAGE_TRKOWN) diff --git a/sys/uvm/uvm_anon.c b/sys/uvm/uvm_anon.c index da7733d32d1..b393d701ddb 100644 --- a/sys/uvm/uvm_anon.c +++ b/sys/uvm/uvm_anon.c @@ -1,5 +1,5 @@ -/* $OpenBSD: uvm_anon.c,v 1.6 2001/01/29 02:07:42 niklas Exp $ */ -/* $NetBSD: uvm_anon.c,v 1.2 1999/03/26 17:34:15 chs Exp $ */ +/* $OpenBSD: uvm_anon.c,v 1.7 2001/06/23 19:24:33 smart Exp $ */ +/* $NetBSD: uvm_anon.c,v 1.3 1999/08/14 06:25:48 ross Exp $ */ /* * @@ -76,6 +76,7 @@ uvm_anon_init() for (lcv = 0 ; lcv < nanon ; lcv++) { anon[lcv].u.an_nxt = uvm.afree; uvm.afree = &anon[lcv]; + simple_lock_init(&uvm.afree->an_lock); } simple_lock_init(&uvm.afreelock); } @@ -108,6 +109,7 @@ uvm_anon_add(pages) simple_lock_init(&anon->an_lock); anon[lcv].u.an_nxt = uvm.afree; uvm.afree = &anon[lcv]; + simple_lock_init(&uvm.afree->an_lock); } simple_unlock(&uvm.afreelock); } diff --git a/sys/uvm/uvm_aobj.c b/sys/uvm/uvm_aobj.c index 34341c5361b..84e8a3539f5 100644 --- a/sys/uvm/uvm_aobj.c +++ b/sys/uvm/uvm_aobj.c @@ -1,5 +1,5 @@ -/* $OpenBSD: uvm_aobj.c,v 1.9 2001/03/22 03:05:54 smart Exp $ */ -/* $NetBSD: uvm_aobj.c,v 1.21 1999/07/07 05:32:26 thorpej Exp $ */ +/* $OpenBSD: uvm_aobj.c,v 1.10 2001/06/23 19:24:33 smart Exp $ */ +/* $NetBSD: uvm_aobj.c,v 1.25 1999/08/21 02:19:05 thorpej Exp $ */ /* * Copyright (c) 1998 Chuck Silvers, Charles D. Cranor and @@ -705,28 +705,211 @@ uao_detach(uobj) } /* - * uao_flush: uh, yea, sure it's flushed. really! + * uao_flush: "flush" pages out of a uvm object + * + * => object should be locked by caller. we may _unlock_ the object + * if (and only if) we need to clean a page (PGO_CLEANIT). + * XXXJRT Currently, however, we don't. In the case of cleaning + * XXXJRT a page, we simply just deactivate it. Should probably + * XXXJRT handle this better, in the future (although "flushing" + * XXXJRT anonymous memory isn't terribly important). + * => if PGO_CLEANIT is not set, then we will neither unlock the object + * or block. + * => if PGO_ALLPAGE is set, then all pages in the object are valid targets + * for flushing. + * => NOTE: we rely on the fact that the object's memq is a TAILQ and + * that new pages are inserted on the tail end of the list. thus, + * we can make a complete pass through the object in one go by starting + * at the head and working towards the tail (new pages are put in + * front of us). + * => NOTE: we are allowed to lock the page queues, so the caller + * must not be holding the lock on them [e.g. pagedaemon had + * better not call us with the queues locked] + * => we return TRUE unless we encountered some sort of I/O error + * XXXJRT currently never happens, as we never directly initiate + * XXXJRT I/O + * + * comment on "cleaning" object and PG_BUSY pages: + * this routine is holding the lock on the object. the only time + * that is can run into a PG_BUSY page that it does not own is if + * some other process has started I/O on the page (e.g. either + * a pagein or a pageout). if the PG_BUSY page is being paged + * in, then it can not be dirty (!PG_CLEAN) because no one has + * had a change to modify it yet. if the PG_BUSY page is being + * paged out then it means that someone else has already started + * cleaning the page for us (how nice!). in this case, if we + * have syncio specified, then after we make our pass through the + * object we need to wait for the other PG_BUSY pages to clear + * off (i.e. we need to do an iosync). also note that once a + * page is PG_BUSY is must stary in its object until it is un-busyed. + * XXXJRT We never actually do this, as we are "flushing" anonymous + * XXXJRT memory, which doesn't have persistent backing store. + * + * note on page traversal: + * we can traverse the pages in an object either by going down the + * linked list in "uobj->memq", or we can go over the address range + * by page doing hash table lookups for each address. depending + * on how many pages are in the object it may be cheaper to do one + * or the other. we set "by_list" to true if we are using memq. + * if the cost of a hash lookup was equal to the cost of the list + * traversal we could compare the number of pages in the start->stop + * range to the total number of pages in the object. however, it + * seems that a hash table lookup is more expensive than the linked + * list traversal, so we multiply the number of pages in the + * start->stop range by a penalty which we define below. */ + +#define UAO_HASH_PENALTY 4 /* XXX: a guess */ + boolean_t -uao_flush(uobj, start, end, flags) +uao_flush(uobj, start, stop, flags) struct uvm_object *uobj; - vaddr_t start, end; + vaddr_t start, stop; int flags; { + struct uvm_aobj *aobj = (struct uvm_aobj *) uobj; + struct vm_page *pp, *ppnext; + boolean_t retval, by_list; + vaddr_t curoff; + UVMHIST_FUNC("uao_flush"); UVMHIST_CALLED(maphist); + + curoff = 0; /* XXX: shut up gcc */ + + retval = TRUE; /* default to success */ + + if (flags & PGO_ALLPAGES) { + start = 0; + stop = aobj->u_pages << PAGE_SHIFT; + by_list = TRUE; /* always go by the list */ + } else { + start = trunc_page(start); + stop = round_page(stop); + if (stop > (aobj->u_pages << PAGE_SHIFT)) { + printf("uao_flush: strange, got an out of range " + "flush (fixed)\n"); + stop = aobj->u_pages << PAGE_SHIFT; + } + by_list = (uobj->uo_npages <= + ((stop - start) >> PAGE_SHIFT) * UAO_HASH_PENALTY); + } + + UVMHIST_LOG(maphist, + " flush start=0x%lx, stop=0x%x, by_list=%d, flags=0x%x", + start, stop, by_list, flags); /* - * anonymous memory doesn't "flush" - */ + * Don't need to do any work here if we're not freeing + * or deactivating pages. + */ + if ((flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) { + UVMHIST_LOG(maphist, + "<- done (no work to do)",0,0,0,0); + return (retval); + } + /* - * XXX - * Deal with: - * - * PGO_DEACTIVATE for sequential access, via uvm_fault(), and - * for MADV_DONTNEED - * - * PGO_FREE for MADV_FREE and MSINVALIDATE + * now do it. note: we must update ppnext in the body of loop or we + * will get stuck. we need to use ppnext because we may free "pp" + * before doing the next loop. */ - return TRUE; + + if (by_list) { + pp = uobj->memq.tqh_first; + } else { + curoff = start; + pp = uvm_pagelookup(uobj, curoff); + } + + ppnext = NULL; /* XXX: shut up gcc */ + uvm_lock_pageq(); /* page queues locked */ + + /* locked: both page queues and uobj */ + for ( ; (by_list && pp != NULL) || + (!by_list && curoff < stop) ; pp = ppnext) { + if (by_list) { + ppnext = pp->listq.tqe_next; + + /* range check */ + if (pp->offset < start || pp->offset >= stop) + continue; + } else { + curoff += PAGE_SIZE; + if (curoff < stop) + ppnext = uvm_pagelookup(uobj, curoff); + + /* null check */ + if (pp == NULL) + continue; + } + + switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { + /* + * XXX In these first 3 cases, we always just + * XXX deactivate the page. We may want to + * XXX handle the different cases more specifically + * XXX in the future. + */ + case PGO_CLEANIT|PGO_FREE: + case PGO_CLEANIT|PGO_DEACTIVATE: + case PGO_DEACTIVATE: + deactivate_it: + /* skip the page if it's loaned or wired */ + if (pp->loan_count != 0 || + pp->wire_count != 0) + continue; + + /* zap all mappings for the page. */ + pmap_page_protect(PMAP_PGARG(pp), + VM_PROT_NONE); + + /* ...and deactivate the page. */ + uvm_pagedeactivate(pp); + + continue; + + case PGO_FREE: + /* + * If there are multiple references to + * the object, just deactivate the page. + */ + if (uobj->uo_refs > 1) + goto deactivate_it; + + /* XXX skip the page if it's loaned or wired */ + if (pp->loan_count != 0 || + pp->wire_count != 0) + continue; + + /* + * mark the page as released if its busy. + */ + if (pp->flags & PG_BUSY) { + pp->flags |= PG_RELEASED; + continue; + } + + /* zap all mappings for the page. */ + pmap_page_protect(PMAP_PGARG(pp), + VM_PROT_NONE); + + uao_dropswap(uobj, pp->offset >> PAGE_SHIFT); + uvm_pagefree(pp); + + continue; + + default: + panic("uao_flush: weird flags"); + } +#ifdef DIAGNOSTIC + panic("uao_flush: unreachable code"); +#endif + } + + uvm_unlock_pageq(); + + UVMHIST_LOG(maphist, + "<- done, rv=%d",retval,0,0,0); + return (retval); } /* diff --git a/sys/uvm/uvm_extern.h b/sys/uvm/uvm_extern.h index bcfd4060cd3..5491dd744c8 100644 --- a/sys/uvm/uvm_extern.h +++ b/sys/uvm/uvm_extern.h @@ -1,5 +1,5 @@ -/* $OpenBSD: uvm_extern.h,v 1.12 2001/05/10 07:59:06 art Exp $ */ -/* $NetBSD: uvm_extern.h,v 1.32 1999/07/02 23:20:58 thorpej Exp $ */ +/* $OpenBSD: uvm_extern.h,v 1.13 2001/06/23 19:24:33 smart Exp $ */ +/* $NetBSD: uvm_extern.h,v 1.34 1999/07/22 22:58:38 thorpej Exp $ */ /* * @@ -136,6 +136,12 @@ #define UVM_PGA_USERESERVE 0x0001 /* + * lockflags that control the locking behavior of various functions. + */ +#define UVM_LK_ENTER 0x00000001 /* map locked on entry */ +#define UVM_LK_EXIT 0x00000002 /* leave map locked on exit */ + +/* * structures */ @@ -150,6 +156,7 @@ struct vm_anon; struct vmspace; struct pmap; struct vnode; +struct simplelock; /* * uvmexp: global data structures that are exported to parts of the kernel @@ -282,6 +289,8 @@ int uvm_fault __P((vm_map_t, vaddr_t, #if defined(KGDB) void uvm_chgkprot __P((caddr_t, size_t, int)); #endif +void uvm_sleep __P((void *, struct simplelock *, boolean_t, + const char *, int)); void uvm_fork __P((struct proc *, struct proc *, boolean_t, void *, size_t)); void uvm_exit __P((struct proc *)); @@ -360,7 +369,7 @@ struct vm_page *uvm_pagealloc_strat __P((struct uvm_object *, #define uvm_pagealloc(obj, off, anon, flags) \ uvm_pagealloc_strat((obj), (off), (anon), (flags), \ UVM_PGA_STRAT_NORMAL, 0) -vaddr_t uvm_pagealloc_contig __P((vaddr_t, vaddr_t, +vaddr_t uvm_pagealloc_contig __P((vaddr_t, vaddr_t, vaddr_t, vaddr_t)); void uvm_pagerealloc __P((struct vm_page *, struct uvm_object *, vaddr_t)); diff --git a/sys/uvm/uvm_fault.c b/sys/uvm/uvm_fault.c index dfbff5be96a..2db0ebe05e1 100644 --- a/sys/uvm/uvm_fault.c +++ b/sys/uvm/uvm_fault.c @@ -1,5 +1,5 @@ -/* $OpenBSD: uvm_fault.c,v 1.13 2001/06/08 08:09:38 art Exp $ */ -/* $NetBSD: uvm_fault.c,v 1.40 1999/07/08 18:11:03 thorpej Exp $ */ +/* $OpenBSD: uvm_fault.c,v 1.14 2001/06/23 19:24:33 smart Exp $ */ +/* $NetBSD: uvm_fault.c,v 1.44 1999/07/22 22:58:38 thorpej Exp $ */ /* * @@ -724,7 +724,7 @@ ReFault: npages = nback + nforw + 1; centeridx = nback; - narrow = FALSE; /* ensure only once per-fault */ + narrow = TRUE; /* ensure only once per-fault */ } else { @@ -843,8 +843,8 @@ ReFault: uvmexp.fltnamap++; pmap_enter(ufi.orig_map->pmap, currva, VM_PAGE_TO_PHYS(anon->u.an_page), - (anon->an_ref > 1) ? - (enter_prot & ~VM_PROT_WRITE) : enter_prot, + (anon->an_ref > 1) ? (enter_prot & ~VM_PROT_WRITE) : + enter_prot, VM_MAPENT_ISWIRED(ufi.entry), 0); } simple_unlock(&anon->an_lock); diff --git a/sys/uvm/uvm_glue.c b/sys/uvm/uvm_glue.c index 6d0d2409b83..03560f5ea94 100644 --- a/sys/uvm/uvm_glue.c +++ b/sys/uvm/uvm_glue.c @@ -1,5 +1,5 @@ -/* $OpenBSD: uvm_glue.c,v 1.15 2001/06/08 08:09:39 art Exp $ */ -/* $NetBSD: uvm_glue.c,v 1.27 1999/07/08 18:11:03 thorpej Exp $ */ +/* $OpenBSD: uvm_glue.c,v 1.16 2001/06/23 19:24:33 smart Exp $ */ +/* $NetBSD: uvm_glue.c,v 1.29 1999/07/25 06:30:36 thorpej Exp $ */ /* * Copyright (c) 1997 Charles D. Cranor and Washington University. @@ -107,6 +107,31 @@ int readbuffers = 0; /* allow KGDB to read kern buffer pool */ /* + * uvm_sleep: atomic unlock and sleep for UVM_UNLOCK_AND_WAIT(). + */ + +void +uvm_sleep(event, slock, canintr, msg, timo) + void *event; + struct simplelock *slock; + boolean_t canintr; + const char *msg; + int timo; +{ + int s, pri; + + pri = PVM; + if (canintr) + pri |= PCATCH; + + s = splhigh(); + if (slock != NULL) + simple_unlock(slock); + (void) tsleep(event, pri, (char *)msg, timo); + splx(s); +} + +/* * uvm_kernacc: can the kernel access a region of memory * * - called from malloc [DIAGNOSTIC], and /dev/kmem driver (mem.c) @@ -298,7 +323,7 @@ uvm_fork(p1, p2, shared, stack, stacksize) panic("uvm_fork: uvm_fault_wire failed: %d", rv); /* - * p_stats currently point at fields in the user struct. Copy + * p_stats currently points at a field in the user struct. Copy * parts of p_stats, and zero out the rest. */ p2->p_stats = &up->u_stats; diff --git a/sys/uvm/uvm_km.c b/sys/uvm/uvm_km.c index 956c913bed5..8841554be91 100644 --- a/sys/uvm/uvm_km.c +++ b/sys/uvm/uvm_km.c @@ -1,5 +1,5 @@ -/* $OpenBSD: uvm_km.c,v 1.9 2001/03/22 03:05:55 smart Exp $ */ -/* $NetBSD: uvm_km.c,v 1.27 1999/06/04 23:38:41 thorpej Exp $ */ +/* $OpenBSD: uvm_km.c,v 1.10 2001/06/23 19:24:33 smart Exp $ */ +/* $NetBSD: uvm_km.c,v 1.31 1999/07/22 22:58:38 thorpej Exp $ */ /* * Copyright (c) 1997 Charles D. Cranor and Washington University. @@ -161,13 +161,6 @@ struct vmi_list vmi_list; simple_lock_data_t vmi_list_slock; /* - * local functions - */ - -static int uvm_km_get __P((struct uvm_object *, vaddr_t, - vm_page_t *, int *, int, vm_prot_t, int, int)); - -/* * local data structues */ @@ -175,233 +168,12 @@ static struct vm_map kernel_map_store; static struct uvm_object kmem_object_store; static struct uvm_object mb_object_store; -static struct uvm_pagerops km_pager = { - NULL, /* init */ - NULL, /* reference */ - NULL, /* detach */ - NULL, /* fault */ - NULL, /* flush */ - uvm_km_get, /* get */ - /* ... rest are NULL */ -}; - /* - * uvm_km_get: pager get function for kernel objects - * - * => currently we do not support pageout to the swap area, so this - * pager is very simple. eventually we may want an anonymous - * object pager which will do paging. - * => XXXCDC: this pager should be phased out in favor of the aobj pager + * All pager operations here are NULL, but the object must have + * a pager ops vector associated with it; various places assume + * it to be so. */ - - -static int -uvm_km_get(uobj, offset, pps, npagesp, centeridx, access_type, advice, flags) - struct uvm_object *uobj; - vaddr_t offset; - struct vm_page **pps; - int *npagesp; - int centeridx, advice, flags; - vm_prot_t access_type; -{ - vaddr_t current_offset; - vm_page_t ptmp; - int lcv, gotpages, maxpages; - boolean_t done; - UVMHIST_FUNC("uvm_km_get"); UVMHIST_CALLED(maphist); - - UVMHIST_LOG(maphist, "flags=%d", flags,0,0,0); - - /* - * 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. 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 */ - - for (lcv = 0, current_offset = offset ; - lcv < maxpages ; lcv++, current_offset += PAGE_SIZE) { - - /* do we care about this page? if not, skip it */ - if (pps[lcv] == PGO_DONTCARE) - continue; - - /* lookup page */ - ptmp = uvm_pagelookup(uobj, current_offset); - - /* null? attempt to allocate the page */ - if (ptmp == NULL) { - ptmp = uvm_pagealloc(uobj, current_offset, - NULL, 0); - if (ptmp) { - /* new page */ - ptmp->flags &= ~(PG_BUSY|PG_FAKE); - UVM_PAGE_OWN(ptmp, NULL); - uvm_pagezero(ptmp); - } - } - - /* - * to be useful must get a non-busy, non-released page - */ - if (ptmp == NULL || - (ptmp->flags & (PG_BUSY|PG_RELEASED)) != 0) { - if (lcv == centeridx || - (flags & PGO_ALLPAGES) != 0) - /* need to do a wait or I/O! */ - done = FALSE; - continue; - } - - /* - * useful page: busy/lock it and plug it in our - * result array - */ - - /* caller must un-busy this page */ - ptmp->flags |= PG_BUSY; - UVM_PAGE_OWN(ptmp, "uvm_km_get1"); - pps[lcv] = ptmp; - gotpages++; - - } /* "for" lcv loop */ - - /* - * step 1b: now we've either done everything needed or we - * to unlock and do some waiting or I/O. - */ - - UVMHIST_LOG(maphist, "<- done (done=%d)", done, 0,0,0); - - *npagesp = gotpages; - if (done) - return(VM_PAGER_OK); /* bingo! */ - else - return(VM_PAGER_UNLOCK); /* EEK! Need to - * unlock and I/O */ - } - - /* - * step 2: get non-resident or busy pages. - * object is locked. data structures are unlocked. - */ - - for (lcv = 0, current_offset = offset ; - lcv < maxpages ; lcv++, current_offset += PAGE_SIZE) { - - /* skip over pages we've already gotten or don't want */ - /* skip over pages we don't _have_ to get */ - if (pps[lcv] != NULL || - (lcv != centeridx && (flags & PGO_ALLPAGES) == 0)) - continue; - - /* - * we have yet to locate the current page (pps[lcv]). we - * first look for a page that is already at the current offset. - * if we find a page, we check to see if it is busy or - * released. if that is the case, then we sleep on the page - * until it is no longer busy or released and repeat the - * lookup. if the page we found is neither busy nor - * released, then we busy it (so we own it) and plug it into - * pps[lcv]. this 'break's the following while loop and - * indicates we are ready to move on to the next page in the - * "lcv" loop above. - * - * if we exit the while loop with pps[lcv] still set to NULL, - * then it means that we allocated a new busy/fake/clean page - * ptmp in the object and we need to do I/O to fill in the - * data. - */ - - while (pps[lcv] == NULL) { /* top of "pps" while loop */ - - /* look for a current page */ - ptmp = uvm_pagelookup(uobj, current_offset); - - /* nope? allocate one now (if we can) */ - if (ptmp == NULL) { - - ptmp = uvm_pagealloc(uobj, current_offset, - NULL, 0); - - /* out of RAM? */ - if (ptmp == NULL) { - simple_unlock(&uobj->vmobjlock); - uvm_wait("kmgetwait1"); - simple_lock(&uobj->vmobjlock); - /* goto top of pps while loop */ - continue; - } - - /* - * got new page ready for I/O. break pps - * while loop. pps[lcv] is still NULL. - */ - break; - } - - /* page is there, see if we need to wait on it */ - if ((ptmp->flags & (PG_BUSY|PG_RELEASED)) != 0) { - ptmp->flags |= PG_WANTED; - UVM_UNLOCK_AND_WAIT(ptmp,&uobj->vmobjlock, - FALSE, "uvn_get",0); - simple_lock(&uobj->vmobjlock); - 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). caller must un-busy. - */ - ptmp->flags |= PG_BUSY; - UVM_PAGE_OWN(ptmp, "uvm_km_get2"); - pps[lcv] = ptmp; - } - - /* - * if we own the a valid page at the correct offset, pps[lcv] - * will point to it. nothing more to do except go to the - * next page. - */ - - if (pps[lcv]) - continue; /* next lcv */ - - /* - * we have a "fake/busy/clean" page that we just allocated. - * do the needed "i/o" (in this case that means zero it). - */ - - uvm_pagezero(ptmp); - ptmp->flags &= ~(PG_FAKE); - pps[lcv] = ptmp; - - } /* lcv loop */ - - /* - * finally, unlock object and return. - */ - - simple_unlock(&uobj->vmobjlock); - UVMHIST_LOG(maphist, "<- done (OK)",0,0,0,0); - return(VM_PAGER_OK); -} +static struct uvm_pagerops km_pager; /* * uvm_km_init: init kernel maps and objects to reflect reality (i.e. @@ -1106,7 +878,7 @@ uvm_km_alloc_poolpage1(map, obj, waitok) vaddr_t va; again: - pg = uvm_pagealloc(NULL, 0, NULL, 0); + pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE); if (pg == NULL) { if (waitok) { uvm_wait("plpg"); diff --git a/sys/uvm/uvm_map.c b/sys/uvm/uvm_map.c index 14e18ff3f60..bb318eb7371 100644 --- a/sys/uvm/uvm_map.c +++ b/sys/uvm/uvm_map.c @@ -1,5 +1,5 @@ -/* $OpenBSD: uvm_map.c,v 1.15 2001/05/10 14:51:21 art Exp $ */ -/* $NetBSD: uvm_map.c,v 1.63 1999/07/07 21:51:35 thorpej Exp $ */ +/* $OpenBSD: uvm_map.c,v 1.16 2001/06/23 19:24:33 smart Exp $ */ +/* $NetBSD: uvm_map.c,v 1.68 1999/08/21 02:19:05 thorpej Exp $ */ /* * Copyright (c) 1997 Charles D. Cranor and Washington University. @@ -1717,6 +1717,7 @@ uvm_map_protect(map, start, end, new_prot, set_max) boolean_t set_max; { vm_map_entry_t current, entry; + int rv = KERN_SUCCESS; UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist); UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)", map, start, end, new_prot); @@ -1737,11 +1738,13 @@ uvm_map_protect(map, start, end, new_prot, set_max) current = entry; while ((current != &map->header) && (current->start < end)) { - if (UVM_ET_ISSUBMAP(current)) - return (KERN_INVALID_ARGUMENT); + if (UVM_ET_ISSUBMAP(current)) { + rv = KERN_INVALID_ARGUMENT; + goto out; + } if ((new_prot & current->max_protection) != new_prot) { - vm_map_unlock(map); - return (KERN_PROTECTION_FAILURE); + rv = KERN_PROTECTION_FAILURE; + goto out; } current = current->next; } @@ -1773,12 +1776,43 @@ uvm_map_protect(map, start, end, new_prot, set_max) current->protection & MASK(entry)); } + /* + * If the map is configured to lock any future mappings, + * wire this entry now if the old protection was VM_PROT_NONE + * and the new protection is not VM_PROT_NONE. + */ + + if ((map->flags & VM_MAP_WIREFUTURE) != 0 && + VM_MAPENT_ISWIRED(entry) == 0 && + old_prot == VM_PROT_NONE && + new_prot != VM_PROT_NONE) { + if (uvm_map_pageable(map, entry->start, + entry->end, FALSE, + UVM_LK_ENTER|UVM_LK_EXIT) != KERN_SUCCESS) { + /* + * If locking the entry fails, remember the + * error if it's the first one. Note we + * still continue setting the protection in + * the map, but will return the resource + * shortage condition regardless. + * + * XXX Ignore what the actual error is, + * XXX just call it a resource shortage + * XXX so that it doesn't get confused + * XXX what uvm_map_protect() itself would + * XXX normally return. + */ + rv = KERN_RESOURCE_SHORTAGE; + } + } + current = current->next; } + out: vm_map_unlock(map); - UVMHIST_LOG(maphist, "<- done",0,0,0,0); - return(KERN_SUCCESS); + UVMHIST_LOG(maphist, "<- done, rv=%d",rv,0,0,0); + return (rv); } #undef max @@ -1825,10 +1859,6 @@ uvm_map_inherit(map, start, end, new_inheritance) entry = temp_entry->next; } - /* - * XXXJRT: disallow holes? - */ - while ((entry != &map->header) && (entry->start < end)) { UVM_MAP_CLIP_END(map, entry, end); @@ -1871,6 +1901,10 @@ uvm_map_advice(map, start, end, new_advice) entry = temp_entry->next; } + /* + * XXXJRT: disallow holes? + */ + while ((entry != &map->header) && (entry->start < end)) { UVM_MAP_CLIP_END(map, entry, end); @@ -1913,10 +1947,11 @@ uvm_map_advice(map, start, end, new_advice) */ int -uvm_map_pageable(map, start, end, new_pageable, islocked) +uvm_map_pageable(map, start, end, new_pageable, lockflags) vm_map_t map; vaddr_t start, end; - boolean_t new_pageable, islocked; + boolean_t new_pageable; + int lockflags; { vm_map_entry_t entry, start_entry, failed_entry; int rv; @@ -1932,8 +1967,9 @@ uvm_map_pageable(map, start, end, new_pageable, islocked) panic("uvm_map_pageable: map %p not pageable", map); #endif - if (islocked == FALSE) + if ((lockflags & UVM_LK_ENTER) == 0) vm_map_lock(map); + VM_MAP_RANGE_CHECK(map, start, end); /* @@ -1945,7 +1981,8 @@ uvm_map_pageable(map, start, end, new_pageable, islocked) */ if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) { - vm_map_unlock(map); + if ((lockflags & UVM_LK_EXIT) == 0) + vm_map_unlock(map); UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); return (KERN_INVALID_ADDRESS); @@ -1967,7 +2004,8 @@ uvm_map_pageable(map, start, end, new_pageable, islocked) (entry->end < end && (entry->next == &map->header || entry->next->start > entry->end))) { - vm_map_unlock(map); + if ((lockflags & UVM_LK_EXIT) == 0) + vm_map_unlock(map); UVMHIST_LOG(maphist, "<- done (INVALID UNWIRE ARG)",0,0,0,0); return (KERN_INVALID_ARGUMENT); @@ -1987,7 +2025,8 @@ uvm_map_pageable(map, start, end, new_pageable, islocked) uvm_map_entry_unwire(map, entry); entry = entry->next; } - vm_map_unlock(map); + if ((lockflags & UVM_LK_EXIT) == 0) + vm_map_unlock(map); UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); return(KERN_SUCCESS); @@ -2055,7 +2094,8 @@ uvm_map_pageable(map, start, end, new_pageable, islocked) entry->wired_count--; entry = entry->prev; } - vm_map_unlock(map); + if ((lockflags & UVM_LK_EXIT) == 0) + vm_map_unlock(map); UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0); return (KERN_INVALID_ARGUMENT); } @@ -2123,15 +2163,24 @@ uvm_map_pageable(map, start, end, new_pageable, islocked) uvm_map_entry_unwire(map, entry); entry = entry->next; } - vm_map_unlock(map); + if ((lockflags & UVM_LK_EXIT) == 0) + vm_map_unlock(map); UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0); return(rv); } /* We are holding a read lock here. */ - vm_map_unbusy(map); - vm_map_unlock_read(map); - + if ((lockflags & UVM_LK_EXIT) == 0) { + vm_map_unbusy(map); + vm_map_unlock_read(map); + } else { + /* + * Get back to an exclusive (write) lock. + */ + vm_map_upgrade(map); + vm_map_unbusy(map); + } + UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); return(KERN_SUCCESS); } @@ -2323,20 +2372,29 @@ uvm_map_pageable_all(map, flags, limit) /* * first drop the wiring count on all the entries * which haven't actually been wired yet. + * + * Skip VM_PROT_NONE entries like we did above. */ failed_entry = entry; for (/* nothing */; entry != &map->header; - entry = entry->next) + entry = entry->next) { + if (entry->protection == VM_PROT_NONE) + continue; entry->wired_count--; + } /* * now, unwire all the entries that were successfully * wired above. + * + * Skip VM_PROT_NONE entries like we did above. */ for (entry = map->header.next; entry != failed_entry; entry = entry->next) { + if (entry->protection == VM_PROT_NONE) + continue; entry->wired_count--; - if (VM_MAPENT_ISWIRED(entry) == 0) + if (VM_MAPENT_ISWIRED(entry)) uvm_map_entry_unwire(map, entry); } vm_map_unlock(map); @@ -2471,6 +2529,7 @@ uvm_map_clean(map, start, end, flags) case PGO_CLEANIT|PGO_FREE: case PGO_CLEANIT|PGO_DEACTIVATE: case PGO_DEACTIVATE: + deactivate_it: /* skip the page if it's loaned or wired */ if (pg->loan_count != 0 || pg->wire_count != 0) { @@ -2515,12 +2574,19 @@ uvm_map_clean(map, start, end, flags) continue; case PGO_FREE: + /* + * If there are multiple references to + * the amap, just deactivate the page. + */ + if (amap_refs(amap) > 1) + goto deactivate_it; + /* XXX skip the page if it's wired */ if (pg->wire_count != 0) { simple_unlock(&anon->an_lock); continue; } - amap_unadd(&entry->aref, offset); + amap_unadd(¤t->aref, offset); refs = --anon->an_ref; simple_unlock(&anon->an_lock); if (refs == 0) diff --git a/sys/uvm/uvm_meter.c b/sys/uvm/uvm_meter.c index b208c4740ac..c397087d7bb 100644 --- a/sys/uvm/uvm_meter.c +++ b/sys/uvm/uvm_meter.c @@ -1,5 +1,5 @@ -/* $OpenBSD: uvm_meter.c,v 1.9 2001/01/29 02:07:46 niklas Exp $ */ -/* $NetBSD: uvm_meter.c,v 1.8 1999/03/25 18:48:53 mrg Exp $ */ +/* $OpenBSD: uvm_meter.c,v 1.10 2001/06/23 19:24:34 smart Exp $ */ +/* $NetBSD: uvm_meter.c,v 1.10 1999/07/25 06:30:36 thorpej Exp $ */ /* * Copyright (c) 1997 Charles D. Cranor and Washington University. diff --git a/sys/uvm/uvm_mmap.c b/sys/uvm/uvm_mmap.c index 791a056d936..d52dfa4e539 100644 --- a/sys/uvm/uvm_mmap.c +++ b/sys/uvm/uvm_mmap.c @@ -1,5 +1,5 @@ -/* $OpenBSD: uvm_mmap.c,v 1.14 2001/06/08 08:09:39 art Exp $ */ -/* $NetBSD: uvm_mmap.c,v 1.30 1999/07/08 00:52:45 thorpej Exp $ */ +/* $OpenBSD: uvm_mmap.c,v 1.15 2001/06/23 19:24:34 smart Exp $ */ +/* $NetBSD: uvm_mmap.c,v 1.35 1999/07/17 21:35:50 thorpej Exp $ */ /* * Copyright (c) 1997 Charles D. Cranor and Washington University. @@ -93,7 +93,7 @@ sys_sbrk(p, v, retval) { #if 0 struct sys_sbrk_args /* { - syscallarg(int) incr; + syscallarg(intptr_t) incr; } */ *uap = v; #endif @@ -198,16 +198,16 @@ sys_mincore(p, v, retval) lim = end < entry->end ? end : entry->end; /* - * Special case for mapped devices; these are always - * considered resident. + * Special case for objects with no "real" pages. Those + * are always considered resident (mapped devices). */ if (UVM_ET_ISOBJ(entry)) { - extern struct uvm_pagerops uvm_deviceops; /* XXX */ #ifdef DIAGNOSTIC if (UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) panic("mincore: user map has kernel object"); #endif - if (entry->object.uvm_obj->pgops == &uvm_deviceops) { + if (entry->object.uvm_obj->pgops->pgo_releasepg + == NULL) { for (/* nothing */; start < lim; start += PAGE_SIZE, vec++) subyte(vec, 1); @@ -215,8 +215,8 @@ sys_mincore(p, v, retval) } } - uobj = entry->object.uvm_obj; /* top layer */ - amap = entry->aref.ar_amap; /* bottom layer */ + amap = entry->aref.ar_amap; /* top layer */ + uobj = entry->object.uvm_obj; /* bottom layer */ if (amap != NULL) amap_lock(amap); @@ -621,7 +621,9 @@ sys_msync(p, v, retval) /* * translate MS_ flags into PGO_ flags */ - uvmflags = PGO_CLEANIT | (flags & MS_INVALIDATE) ? PGO_FREE : 0; + uvmflags = PGO_CLEANIT; + if (flags & MS_INVALIDATE) + uvmflags |= PGO_FREE; if (flags & MS_SYNC) uvmflags |= PGO_SYNCIO; else @@ -980,7 +982,7 @@ sys_mlock(p, v, retval) #endif error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, FALSE, - FALSE); + 0); return (error == KERN_SUCCESS ? 0 : ENOMEM); } @@ -1027,7 +1029,7 @@ sys_munlock(p, v, retval) #endif error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, TRUE, - FALSE); + 0); return (error == KERN_SUCCESS ? 0 : ENOMEM); } @@ -1258,7 +1260,7 @@ uvm_mmap(map, addr, size, prot, maxprot, flags, handle, foff, locklimit) goto bad; } retval = uvm_map_pageable(map, *addr, *addr + size, - FALSE, TRUE); + FALSE, UVM_LK_ENTER); if (retval != KERN_SUCCESS) { /* unmap the region! */ (void) uvm_unmap(map, *addr, *addr + size); diff --git a/sys/uvm/uvm_vnode.c b/sys/uvm/uvm_vnode.c index 8219e4faf87..12327f97836 100644 --- a/sys/uvm/uvm_vnode.c +++ b/sys/uvm/uvm_vnode.c @@ -1,5 +1,5 @@ -/* $OpenBSD: uvm_vnode.c,v 1.12 2001/03/22 03:05:57 smart Exp $ */ -/* $NetBSD: uvm_vnode.c,v 1.23 1999/04/11 04:04:11 chs Exp $ */ +/* $OpenBSD: uvm_vnode.c,v 1.13 2001/06/23 19:24:34 smart Exp $ */ +/* $NetBSD: uvm_vnode.c,v 1.25 1999/07/22 22:58:39 thorpej Exp $ */ /* * Copyright (c) 1997 Charles D. Cranor and Washington University. |