X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/0c530ab8987f0ae6a1a3d9284f40182b88852816..13f56ec4e58bf8687e2a68032c093c0213dd519b:/osfmk/vm/memory_object.c diff --git a/osfmk/vm/memory_object.c b/osfmk/vm/memory_object.c index f990e664d..de7baff29 100644 --- a/osfmk/vm/memory_object.c +++ b/osfmk/vm/memory_object.c @@ -1,23 +1,29 @@ /* - * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2008 Apple Inc. All rights reserved. * - * @APPLE_LICENSE_HEADER_START@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * - * The contents of this file constitute Original Code as defined in and - * are subject to the Apple Public Source License Version 1.1 (the - * "License"). You may not use this file except in compliance with the - * License. Please obtain a copy of the License at - * http://www.apple.com/publicsource and read it before using this file. + * This file contains Original Code and/or Modifications of Original Code + * as defined in and that are subject to the Apple Public Source License + * Version 2.0 (the 'License'). You may not use this file except in + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * unlawful or unlicensed copies of an Apple operating system, or to + * circumvent, violate, or enable the circumvention or violation of, any + * terms of an Apple operating system software license agreement. * - * This Original Code and all software distributed under the License are - * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER + * Please obtain a copy of the License at + * http://www.opensource.apple.com/apsl/ and read it before using this file. + * + * The Original Code and all software distributed under the License are + * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the - * License for the specific language governing rights and limitations - * under the License. + * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. + * Please see the License for the specific language governing rights and + * limitations under the License. * - * @APPLE_LICENSE_HEADER_END@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* * @OSF_COPYRIGHT@ @@ -94,6 +100,8 @@ #include /* For pmap_clear_modify */ #include /* For kernel_map, vm_move */ #include /* For vm_map_pageable */ +#include /* Needed by some vm_page.h macros */ +#include #if MACH_PAGEMAP #include @@ -103,8 +111,7 @@ memory_object_default_t memory_manager_default = MEMORY_OBJECT_DEFAULT_NULL; -vm_size_t memory_manager_default_cluster = 0; -decl_mutex_data(, memory_manager_default_lock) +decl_lck_mtx_data(, memory_manager_default_lock) /* @@ -133,10 +140,10 @@ decl_mutex_data(, memory_manager_default_lock) typedef int memory_object_lock_result_t; -#define MEMORY_OBJECT_LOCK_RESULT_DONE 0 -#define MEMORY_OBJECT_LOCK_RESULT_MUST_BLOCK 1 -#define MEMORY_OBJECT_LOCK_RESULT_MUST_CLEAN 2 -#define MEMORY_OBJECT_LOCK_RESULT_MUST_RETURN 3 +#define MEMORY_OBJECT_LOCK_RESULT_DONE 0 +#define MEMORY_OBJECT_LOCK_RESULT_MUST_BLOCK 1 +#define MEMORY_OBJECT_LOCK_RESULT_MUST_RETURN 2 +#define MEMORY_OBJECT_LOCK_RESULT_MUST_FREE 3 memory_object_lock_result_t memory_object_lock_page( vm_page_t m, @@ -166,192 +173,151 @@ memory_object_lock_page( { XPR(XPR_MEMORY_OBJECT, "m_o_lock_page, page 0x%X rtn %d flush %d prot %d\n", - (integer_t)m, should_return, should_flush, prot, 0); + m, should_return, should_flush, prot, 0); - /* - * If we cannot change access to the page, - * either because a mapping is in progress - * (busy page) or because a mapping has been - * wired, then give up. - */ - if (m->busy || m->cleaning) - return(MEMORY_OBJECT_LOCK_RESULT_MUST_BLOCK); + if (m->busy || m->cleaning) { + if (m->list_req_pending && + should_return == MEMORY_OBJECT_RETURN_NONE && + should_flush == TRUE) { + + if (m->absent) { + /* + * this is the list_req_pending | absent | busy case + * which originates from vm_fault_page. + * Combine that with should_flush == TRUE and we + * have a case where we need to toss the page from + * the object. + */ + if (!VM_PAGE_WIRED(m)) { + return (MEMORY_OBJECT_LOCK_RESULT_MUST_FREE); + } else { + return (MEMORY_OBJECT_LOCK_RESULT_DONE); + } + } + if (m->pageout || m->cleaning) { + /* + * if pageout is set, page was earmarked by vm_pageout_scan + * to be cleaned and stolen... if cleaning is set, we're + * pre-cleaning pages for a hibernate... + * in either case, we're going + * to take it back since we are being asked to + * flush the page w/o cleaning it (i.e. we don't + * care that it's dirty, we want it gone from + * the cache) and we don't want to stall + * waiting for it to be cleaned for 2 reasons... + * 1 - no use paging it out since we're probably + * shrinking the file at this point or we no + * longer care about the data in the page + * 2 - if we stall, we may casue a deadlock in + * the FS trying to acquire its locks + * on the VNOP_PAGEOUT path presuming that + * those locks are already held on the truncate + * path before calling through to this function + * + * so undo all of the state that vm_pageout_scan + * hung on this page + */ + vm_pageout_queue_steal(m, FALSE); + PAGE_WAKEUP_DONE(m); + } else { + panic("list_req_pending on page %p without absent/pageout/cleaning set\n", m); + } + } else + return (MEMORY_OBJECT_LOCK_RESULT_MUST_BLOCK); + } /* * Don't worry about pages for which the kernel * does not have any data. */ - if (m->absent || m->error || m->restart) { - if(m->error && should_flush) { - /* dump the page, pager wants us to */ - /* clean it up and there is no */ - /* relevant data to return */ - if(m->wire_count == 0) { - VM_PAGE_FREE(m); - return(MEMORY_OBJECT_LOCK_RESULT_DONE); - } - } else { - return(MEMORY_OBJECT_LOCK_RESULT_DONE); + if (m->error && should_flush && !VM_PAGE_WIRED(m)) { + /* + * dump the page, pager wants us to + * clean it up and there is no + * relevant data to return + */ + return (MEMORY_OBJECT_LOCK_RESULT_MUST_FREE); } + return (MEMORY_OBJECT_LOCK_RESULT_DONE); } - assert(!m->fictitious); - if (m->wire_count != 0) { + if (VM_PAGE_WIRED(m)) { /* - * If no change would take place - * anyway, return successfully. - * - * No change means: - * Not flushing AND - * No change to page lock [2 checks] AND - * Should not return page - * - * XXX This doesn't handle sending a copy of a wired - * XXX page to the pager, but that will require some - * XXX significant surgery. + * The page is wired... just clean or return the page if needed. + * Wired pages don't get flushed or disconnected from the pmap. */ - if (!should_flush && - (m->page_lock == prot || prot == VM_PROT_NO_CHANGE) && - ! memory_object_should_return_page(m, should_return)) { - - /* - * Restart page unlock requests, - * even though no change took place. - * [Memory managers may be expecting - * to see new requests.] - */ - m->unlock_request = VM_PROT_NONE; - PAGE_WAKEUP(m); - - return(MEMORY_OBJECT_LOCK_RESULT_DONE); - } - - return(MEMORY_OBJECT_LOCK_RESULT_MUST_BLOCK); - } + if (memory_object_should_return_page(m, should_return)) + return (MEMORY_OBJECT_LOCK_RESULT_MUST_RETURN); - /* - * If the page is to be flushed, allow - * that to be done as part of the protection. - */ - - if (should_flush) - prot = VM_PROT_ALL; - - /* - * Set the page lock. - * - * If we are decreasing permission, do it now; - * let the fault handler take care of increases - * (pmap_page_protect may not increase protection). - */ + return (MEMORY_OBJECT_LOCK_RESULT_DONE); + } - if (prot != VM_PROT_NO_CHANGE) { - if ((m->page_lock ^ prot) & prot) { - pmap_page_protect(m->phys_page, VM_PROT_ALL & ~prot); - } -#if 0 - /* code associated with the vestigial - * memory_object_data_unlock + if (should_flush) { + /* + * must do the pmap_disconnect before determining the + * need to return the page... otherwise it's possible + * for the page to go from the clean to the dirty state + * after we've made our decision */ - m->page_lock = prot; - m->lock_supplied = TRUE; - if (prot != VM_PROT_NONE) - m->unusual = TRUE; - else - m->unusual = FALSE; - + if (pmap_disconnect(m->phys_page) & VM_MEM_MODIFIED) + m->dirty = TRUE; + } else { /* - * Restart any past unlock requests, even if no - * change resulted. If the manager explicitly - * requested no protection change, then it is assumed - * to be remembering past requests. + * If we are decreasing permission, do it now; + * let the fault handler take care of increases + * (pmap_page_protect may not increase protection). */ - - m->unlock_request = VM_PROT_NONE; -#endif /* 0 */ - PAGE_WAKEUP(m); + if (prot != VM_PROT_NO_CHANGE) + pmap_page_protect(m->phys_page, VM_PROT_ALL & ~prot); } - /* - * Handle page returning. + * Handle returning dirty or precious pages */ - if (memory_object_should_return_page(m, should_return)) { - /* - * If we weren't planning - * to flush the page anyway, - * we may need to remove the - * page from the pageout - * system and from physical - * maps now. + * we use to do a pmap_disconnect here in support + * of memory_object_lock_request, but that routine + * no longer requires this... in any event, in + * our world, it would turn into a big noop since + * we don't lock the page in any way and as soon + * as we drop the object lock, the page can be + * faulted back into an address space + * + * if (!should_flush) + * pmap_disconnect(m->phys_page); */ - - vm_page_lock_queues(); - VM_PAGE_QUEUES_REMOVE(m); - vm_page_unlock_queues(); - - if (!should_flush) - pmap_disconnect(m->phys_page); - - if (m->dirty) - return(MEMORY_OBJECT_LOCK_RESULT_MUST_CLEAN); - else - return(MEMORY_OBJECT_LOCK_RESULT_MUST_RETURN); + return (MEMORY_OBJECT_LOCK_RESULT_MUST_RETURN); } /* - * Handle flushing + * Handle flushing clean pages */ + if (should_flush) + return (MEMORY_OBJECT_LOCK_RESULT_MUST_FREE); - if (should_flush) { - VM_PAGE_FREE(m); - } else { - /* - * XXX Make clean but not flush a paging hint, - * and deactivate the pages. This is a hack - * because it overloads flush/clean with - * implementation-dependent meaning. This only - * happens to pages that are already clean. - */ - - if (vm_page_deactivate_hint && - (should_return != MEMORY_OBJECT_RETURN_NONE)) { - vm_page_lock_queues(); - vm_page_deactivate(m); - vm_page_unlock_queues(); - } - } + /* + * we use to deactivate clean pages at this point, + * but we do not believe that an msync should change + * the 'age' of a page in the cache... here is the + * original comment and code concerning this... + * + * XXX Make clean but not flush a paging hint, + * and deactivate the pages. This is a hack + * because it overloads flush/clean with + * implementation-dependent meaning. This only + * happens to pages that are already clean. + * + * if (vm_page_deactivate_hint && (should_return != MEMORY_OBJECT_RETURN_NONE)) + * return (MEMORY_OBJECT_LOCK_RESULT_MUST_DEACTIVATE); + */ - return(MEMORY_OBJECT_LOCK_RESULT_DONE); + return (MEMORY_OBJECT_LOCK_RESULT_DONE); } -#define LIST_REQ_PAGEOUT_PAGES(object, data_cnt, action, po, ro, ioerr, iosync) \ -MACRO_BEGIN \ - \ - register int upl_flags; \ - \ - vm_object_unlock(object); \ - \ - if (iosync) \ - upl_flags = UPL_MSYNC | UPL_IOSYNC; \ - else \ - upl_flags = UPL_MSYNC; \ - \ - (void) memory_object_data_return(object->pager, \ - po, \ - data_cnt, \ - ro, \ - ioerr, \ - (action == MEMORY_OBJECT_LOCK_RESULT_MUST_CLEAN), \ - !should_flush, \ - upl_flags); \ - \ - vm_object_lock(object); \ -MACRO_END + /* * Routine: memory_object_lock_request [user interface] @@ -390,16 +356,8 @@ memory_object_lock_request( vm_prot_t prot) { vm_object_t object; - __unused boolean_t should_flush; - - should_flush = flags & MEMORY_OBJECT_DATA_FLUSH; - - XPR(XPR_MEMORY_OBJECT, - "m_o_lock_request, control 0x%X off 0x%X size 0x%X flags %X prot %X\n", - (integer_t)control, offset, size, - (((should_return&1)<<1)|should_flush), prot); - /* + /* * Check for bogus arguments. */ object = memory_object_control_to_vm_object(control); @@ -417,10 +375,20 @@ memory_object_lock_request( */ vm_object_lock(object); vm_object_paging_begin(object); + + if (flags & MEMORY_OBJECT_DATA_FLUSH_ALL) { + if ((should_return != MEMORY_OBJECT_RETURN_NONE) || offset || object->copy) { + flags &= ~MEMORY_OBJECT_DATA_FLUSH_ALL; + flags |= MEMORY_OBJECT_DATA_FLUSH; + } + } offset -= object->paging_offset; - (void)vm_object_update(object, - offset, size, resid_offset, io_errno, should_return, flags, prot); + if (flags & MEMORY_OBJECT_DATA_FLUSH_ALL) + vm_object_reap_pages(object, REAP_DATA_FLUSH); + else + (void)vm_object_update(object, offset, size, resid_offset, + io_errno, should_return, flags, prot); vm_object_paging_end(object); vm_object_unlock(object); @@ -520,7 +488,7 @@ vm_object_sync( XPR(XPR_VM_OBJECT, "vm_o_sync, object 0x%X, offset 0x%X size 0x%x flush %d rtn %d\n", - (integer_t)object, offset, size, should_flush, should_return); + object, offset, size, should_flush, should_return); /* * Lock the object, and acquire a paging reference to @@ -553,6 +521,40 @@ vm_object_sync( +#define LIST_REQ_PAGEOUT_PAGES(object, data_cnt, po, ro, ioerr, iosync) \ +MACRO_BEGIN \ + \ + int upl_flags; \ + memory_object_t pager; \ + \ + if (object == slide_info.slide_object) { \ + panic("Objects with slid pages not allowed\n"); \ + } \ + \ + if ((pager = (object)->pager) != MEMORY_OBJECT_NULL) { \ + vm_object_paging_begin(object); \ + vm_object_unlock(object); \ + \ + if (iosync) \ + upl_flags = UPL_MSYNC | UPL_IOSYNC; \ + else \ + upl_flags = UPL_MSYNC; \ + \ + (void) memory_object_data_return(pager, \ + po, \ + (memory_object_cluster_size_t)data_cnt, \ + ro, \ + ioerr, \ + FALSE, \ + FALSE, \ + upl_flags); \ + \ + vm_object_lock(object); \ + vm_object_paging_end(object); \ + } \ +MACRO_END + + static int vm_object_update_extent( @@ -568,116 +570,128 @@ vm_object_update_extent( { vm_page_t m; int retval = 0; - vm_size_t data_cnt = 0; vm_object_offset_t paging_offset = 0; - vm_object_offset_t last_offset = offset; + vm_object_offset_t next_offset = offset; memory_object_lock_result_t page_lock_result; - memory_object_lock_result_t pageout_action; - - pageout_action = MEMORY_OBJECT_LOCK_RESULT_DONE; + memory_object_cluster_size_t data_cnt = 0; + struct vm_page_delayed_work dw_array[DEFAULT_DELAYED_WORK_LIMIT]; + struct vm_page_delayed_work *dwp; + int dw_count; + int dw_limit; + + dwp = &dw_array[0]; + dw_count = 0; + dw_limit = DELAYED_WORK_LIMIT(DEFAULT_DELAYED_WORK_LIMIT); for (; offset < offset_end && object->resident_page_count; offset += PAGE_SIZE_64) { /* - * Limit the number of pages to be cleaned at once. + * Limit the number of pages to be cleaned at once to a contiguous + * run, or at most MAX_UPL_TRANSFER size */ - if (data_cnt >= PAGE_SIZE * MAX_UPL_TRANSFER) { - LIST_REQ_PAGEOUT_PAGES(object, data_cnt, - pageout_action, paging_offset, offset_resid, io_errno, should_iosync); - data_cnt = 0; - } + if (data_cnt) { + if ((data_cnt >= PAGE_SIZE * MAX_UPL_TRANSFER) || (next_offset != offset)) { + if (dw_count) { + vm_page_do_delayed_work(object, &dw_array[0], dw_count); + dwp = &dw_array[0]; + dw_count = 0; + } + LIST_REQ_PAGEOUT_PAGES(object, data_cnt, + paging_offset, offset_resid, io_errno, should_iosync); + data_cnt = 0; + } + } while ((m = vm_page_lookup(object, offset)) != VM_PAGE_NULL) { - page_lock_result = memory_object_lock_page(m, should_return, should_flush, prot); - - XPR(XPR_MEMORY_OBJECT, - "m_o_update: lock_page, obj 0x%X offset 0x%X result %d\n", - (integer_t)object, offset, page_lock_result, 0, 0); - - switch (page_lock_result) - { - case MEMORY_OBJECT_LOCK_RESULT_DONE: - /* - * End of a cluster of dirty pages. - */ - if (data_cnt) { - LIST_REQ_PAGEOUT_PAGES(object, - data_cnt, pageout_action, - paging_offset, offset_resid, io_errno, should_iosync); - data_cnt = 0; - continue; - } - break; - - case MEMORY_OBJECT_LOCK_RESULT_MUST_BLOCK: - /* - * Since it is necessary to block, - * clean any dirty pages now. - */ - if (data_cnt) { - LIST_REQ_PAGEOUT_PAGES(object, - data_cnt, pageout_action, - paging_offset, offset_resid, io_errno, should_iosync); - data_cnt = 0; - continue; - } - PAGE_SLEEP(object, m, THREAD_UNINT); - continue; - - case MEMORY_OBJECT_LOCK_RESULT_MUST_CLEAN: - case MEMORY_OBJECT_LOCK_RESULT_MUST_RETURN: - /* - * The clean and return cases are similar. - * - * if this would form a discontiguous block, - * clean the old pages and start anew. - * - * Mark the page busy since we will unlock the - * object if we issue the LIST_REQ_PAGEOUT - */ - m->busy = TRUE; - if (data_cnt && - ((last_offset != offset) || (pageout_action != page_lock_result))) { - LIST_REQ_PAGEOUT_PAGES(object, - data_cnt, pageout_action, - paging_offset, offset_resid, io_errno, should_iosync); - data_cnt = 0; - } - m->busy = FALSE; - - if (m->cleaning) { - PAGE_SLEEP(object, m, THREAD_UNINT); - continue; - } - if (data_cnt == 0) { - pageout_action = page_lock_result; - paging_offset = offset; - } - data_cnt += PAGE_SIZE; - last_offset = offset + PAGE_SIZE_64; - - vm_page_lock_queues(); - /* - * Clean - */ - m->list_req_pending = TRUE; - m->cleaning = TRUE; - - if (should_flush) { - /* - * and add additional state - * for the flush - */ - m->busy = TRUE; - m->pageout = TRUE; - vm_page_wire(m); - } - vm_page_unlock_queues(); - - retval = 1; - break; + + dwp->dw_mask = 0; + + page_lock_result = memory_object_lock_page(m, should_return, should_flush, prot); + + if (data_cnt && page_lock_result != MEMORY_OBJECT_LOCK_RESULT_MUST_RETURN) { + /* + * End of a run of dirty/precious pages. + */ + if (dw_count) { + vm_page_do_delayed_work(object, &dw_array[0], dw_count); + dwp = &dw_array[0]; + dw_count = 0; + } + LIST_REQ_PAGEOUT_PAGES(object, data_cnt, + paging_offset, offset_resid, io_errno, should_iosync); + /* + * LIST_REQ_PAGEOUT_PAGES will drop the object lock which will + * allow the state of page 'm' to change... we need to re-lookup + * the current offset + */ + data_cnt = 0; + continue; + } + + switch (page_lock_result) { + + case MEMORY_OBJECT_LOCK_RESULT_DONE: + break; + + case MEMORY_OBJECT_LOCK_RESULT_MUST_FREE: + dwp->dw_mask |= DW_vm_page_free; + break; + + case MEMORY_OBJECT_LOCK_RESULT_MUST_BLOCK: + PAGE_SLEEP(object, m, THREAD_UNINT); + continue; + + case MEMORY_OBJECT_LOCK_RESULT_MUST_RETURN: + if (data_cnt == 0) + paging_offset = offset; + + data_cnt += PAGE_SIZE; + next_offset = offset + PAGE_SIZE_64; + + /* + * Clean + */ + m->list_req_pending = TRUE; + m->cleaning = TRUE; + + /* + * wired pages shouldn't be flushed and + * since they aren't on any queue, + * no need to remove them + */ + if (!VM_PAGE_WIRED(m)) { + + if (should_flush) { + /* + * add additional state for the flush + */ + m->busy = TRUE; + m->pageout = TRUE; + + dwp->dw_mask |= DW_vm_page_wire; + } + /* + * we use to remove the page from the queues at this + * point, but we do not believe that an msync + * should cause the 'age' of a page to be changed + * + * else + * dwp->dw_mask |= DW_VM_PAGE_QUEUES_REMOVE; + */ + } + retval = 1; + break; + } + if (dwp->dw_mask) { + VM_PAGE_ADD_DELAYED_WORK(dwp, m, dw_count); + + if (dw_count >= dw_limit) { + vm_page_do_delayed_work(object, &dw_array[0], dw_count); + dwp = &dw_array[0]; + dw_count = 0; + } } break; } @@ -686,9 +700,12 @@ vm_object_update_extent( * We have completed the scan for applicable pages. * Clean any pages that have been saved. */ + if (dw_count) + vm_page_do_delayed_work(object, &dw_array[0], dw_count); + if (data_cnt) { - LIST_REQ_PAGEOUT_PAGES(object, - data_cnt, pageout_action, paging_offset, offset_resid, io_errno, should_iosync); + LIST_REQ_PAGEOUT_PAGES(object, data_cnt, + paging_offset, offset_resid, io_errno, should_iosync); } return (retval); } @@ -704,20 +721,21 @@ vm_object_update_extent( */ kern_return_t vm_object_update( - register vm_object_t object, - register vm_object_offset_t offset, - register vm_object_size_t size, - register vm_object_offset_t *resid_offset, - int *io_errno, - memory_object_return_t should_return, - int flags, - vm_prot_t protection) + vm_object_t object, + vm_object_offset_t offset, + vm_object_size_t size, + vm_object_offset_t *resid_offset, + int *io_errno, + memory_object_return_t should_return, + int flags, + vm_prot_t protection) { - vm_object_t copy_object; + vm_object_t copy_object = VM_OBJECT_NULL; boolean_t data_returned = FALSE; boolean_t update_cow; boolean_t should_flush = (flags & MEMORY_OBJECT_DATA_FLUSH) ? TRUE : FALSE; boolean_t should_iosync = (flags & MEMORY_OBJECT_IO_SYNC) ? TRUE : FALSE; + vm_fault_return_t result; int num_of_extents; int n; #define MAX_EXTENTS 8 @@ -751,9 +769,39 @@ vm_object_update( !(flags & MEMORY_OBJECT_DATA_PURGE))) || (flags & MEMORY_OBJECT_COPY_SYNC); + if (update_cow || (flags & (MEMORY_OBJECT_DATA_PURGE | MEMORY_OBJECT_DATA_SYNC))) { + int collisions = 0; + + while ((copy_object = object->copy) != VM_OBJECT_NULL) { + /* + * need to do a try here since we're swimming upstream + * against the normal lock ordering... however, we need + * to hold the object stable until we gain control of the + * copy object so we have to be careful how we approach this + */ + if (vm_object_lock_try(copy_object)) { + /* + * we 'won' the lock on the copy object... + * no need to hold the object lock any longer... + * take a real reference on the copy object because + * we're going to call vm_fault_page on it which may + * under certain conditions drop the lock and the paging + * reference we're about to take... the reference + * will keep the copy object from going away if that happens + */ + vm_object_unlock(object); + vm_object_reference_locked(copy_object); + break; + } + vm_object_unlock(object); - if((((copy_object = object->copy) != NULL) && update_cow) || - (flags & MEMORY_OBJECT_DATA_SYNC)) { + collisions++; + mutex_pause(collisions); + + vm_object_lock(object); + } + } + if ((copy_object != VM_OBJECT_NULL && update_cow) || (flags & MEMORY_OBJECT_DATA_SYNC)) { vm_map_size_t i; vm_map_size_t copy_size; vm_map_offset_t copy_offset; @@ -761,81 +809,94 @@ vm_object_update( vm_page_t page; vm_page_t top_page; kern_return_t error = 0; + struct vm_object_fault_info fault_info; + + if (copy_object != VM_OBJECT_NULL) { + /* + * translate offset with respect to shadow's offset + */ + copy_offset = (offset >= copy_object->vo_shadow_offset) ? + (vm_map_offset_t)(offset - copy_object->vo_shadow_offset) : + (vm_map_offset_t) 0; + + if (copy_offset > copy_object->vo_size) + copy_offset = copy_object->vo_size; + + /* + * clip size with respect to shadow offset + */ + if (offset >= copy_object->vo_shadow_offset) { + copy_size = size; + } else if (size >= copy_object->vo_shadow_offset - offset) { + copy_size = size - (copy_object->vo_shadow_offset - offset); + } else { + copy_size = 0; + } + + if (copy_offset + copy_size > copy_object->vo_size) { + if (copy_object->vo_size >= copy_offset) { + copy_size = copy_object->vo_size - copy_offset; + } else { + copy_size = 0; + } + } + copy_size+=copy_offset; - if(copy_object != NULL) { - /* translate offset with respect to shadow's offset */ - copy_offset = (offset >= copy_object->shadow_offset)? - (vm_map_offset_t)(offset - copy_object->shadow_offset) : - (vm_map_offset_t) 0; - if(copy_offset > copy_object->size) - copy_offset = copy_object->size; - - /* clip size with respect to shadow offset */ - if (offset >= copy_object->shadow_offset) { - copy_size = size; - } else if (size >= copy_object->shadow_offset - offset) { - copy_size = size - - (copy_object->shadow_offset - offset); - } else { - copy_size = 0; - } - - if (copy_offset + copy_size > copy_object->size) { - if (copy_object->size >= copy_offset) { - copy_size = copy_object->size - copy_offset; - } else { - copy_size = 0; - } - } - - copy_size+=copy_offset; - - vm_object_unlock(object); - vm_object_lock(copy_object); } else { copy_object = object; copy_size = offset + size; copy_offset = offset; } + fault_info.interruptible = THREAD_UNINT; + fault_info.behavior = VM_BEHAVIOR_SEQUENTIAL; + fault_info.user_tag = 0; + fault_info.lo_offset = copy_offset; + fault_info.hi_offset = copy_size; + fault_info.no_cache = FALSE; + fault_info.stealth = TRUE; + fault_info.io_sync = FALSE; + fault_info.cs_bypass = FALSE; + fault_info.mark_zf_absent = FALSE; vm_object_paging_begin(copy_object); - for (i=copy_offset; iobject, top_page); - PAGE_WAKEUP_DONE(page); - vm_page_lock_queues(); - if (!page->active && !page->inactive) - vm_page_activate(page); - vm_page_unlock_queues(); vm_object_lock(copy_object); vm_object_paging_begin(copy_object); - } else { - PAGE_WAKEUP_DONE(page); - vm_page_lock_queues(); - if (!page->active && !page->inactive) - vm_page_activate(page); + } + if (!page->active && + !page->inactive && + !page->throttled) { + vm_page_lockspin_queues(); + if (!page->active && + !page->inactive && + !page->throttled) + vm_page_deactivate(page); vm_page_unlock_queues(); } + PAGE_WAKEUP_DONE(page); break; case VM_FAULT_RETRY: prot = VM_PROT_WRITE|VM_PROT_READ; @@ -853,34 +914,45 @@ vm_object_update( vm_object_lock(copy_object); vm_object_paging_begin(copy_object); goto RETRY_COW_OF_LOCK_REQUEST; - case VM_FAULT_FICTITIOUS_SHORTAGE: - vm_page_more_fictitious(); - prot = VM_PROT_WRITE|VM_PROT_READ; - vm_object_lock(copy_object); - vm_object_paging_begin(copy_object); - goto RETRY_COW_OF_LOCK_REQUEST; + case VM_FAULT_SUCCESS_NO_VM_PAGE: + /* success but no VM page: fail */ + vm_object_paging_end(copy_object); + vm_object_unlock(copy_object); + /*FALLTHROUGH*/ case VM_FAULT_MEMORY_ERROR: + if (object != copy_object) + vm_object_deallocate(copy_object); vm_object_lock(object); goto BYPASS_COW_COPYIN; + default: + panic("vm_object_update: unexpected error 0x%x" + " from vm_fault_page()\n", result); } } vm_object_paging_end(copy_object); - if(copy_object != object) { + } + if ((flags & (MEMORY_OBJECT_DATA_SYNC | MEMORY_OBJECT_COPY_SYNC))) { + if (copy_object != VM_OBJECT_NULL && copy_object != object) { vm_object_unlock(copy_object); + vm_object_deallocate(copy_object); vm_object_lock(object); } + return KERN_SUCCESS; } - if((flags & (MEMORY_OBJECT_DATA_SYNC | MEMORY_OBJECT_COPY_SYNC))) { - return KERN_SUCCESS; - } - if(((copy_object = object->copy) != NULL) && - (flags & MEMORY_OBJECT_DATA_PURGE)) { - copy_object->shadow_severed = TRUE; - copy_object->shadowed = FALSE; - copy_object->shadow = NULL; - /* delete the ref the COW was holding on the target object */ - vm_object_deallocate(object); + if (copy_object != VM_OBJECT_NULL && copy_object != object) { + if ((flags & MEMORY_OBJECT_DATA_PURGE)) { + copy_object->shadow_severed = TRUE; + copy_object->shadowed = FALSE; + copy_object->shadow = NULL; + /* + * delete the ref the COW was holding on the target object + */ + vm_object_deallocate(object); + } + vm_object_unlock(copy_object); + vm_object_deallocate(copy_object); + vm_object_lock(object); } BYPASS_COW_COPYIN: @@ -1026,7 +1098,7 @@ kern_return_t memory_object_synchronize_completed( memory_object_control_t control, memory_object_offset_t offset, - vm_offset_t length) + memory_object_size_t length) { vm_object_t object; msync_req_t msr; @@ -1035,7 +1107,7 @@ memory_object_synchronize_completed( XPR(XPR_MEMORY_OBJECT, "m_o_sync_completed, object 0x%X, offset 0x%X length 0x%X\n", - (integer_t)object, offset, length, 0, 0); + object, offset, length, 0, 0); /* * Look for bogus arguments @@ -1076,7 +1148,6 @@ vm_object_set_attributes_common( boolean_t may_cache, memory_object_copy_strategy_t copy_strategy, boolean_t temporary, - memory_object_cluster_size_t cluster_size, boolean_t silent_overwrite, boolean_t advisory_pageout) { @@ -1084,7 +1155,7 @@ vm_object_set_attributes_common( XPR(XPR_MEMORY_OBJECT, "m_o_set_attr_com, object 0x%X flg %x strat %d\n", - (integer_t)object, (may_cache&1)|((temporary&1)<1), copy_strategy, 0, 0); + object, (may_cache&1)|((temporary&1)<1), copy_strategy, 0, 0); if (object == VM_OBJECT_NULL) return(KERN_INVALID_ARGUMENT); @@ -1110,17 +1181,6 @@ vm_object_set_attributes_common( may_cache = TRUE; if (temporary) temporary = TRUE; - if (cluster_size != 0) { - int pages_per_cluster; - pages_per_cluster = atop_32(cluster_size); - /* - * Cluster size must be integral multiple of page size, - * and be a power of 2 number of pages. - */ - if ((cluster_size & (PAGE_SIZE-1)) || - ((pages_per_cluster-1) & pages_per_cluster)) - return KERN_INVALID_ARGUMENT; - } vm_object_lock(object); @@ -1134,12 +1194,6 @@ vm_object_set_attributes_common( object->temporary = temporary; object->silent_overwrite = silent_overwrite; object->advisory_pageout = advisory_pageout; - if (cluster_size == 0) - cluster_size = PAGE_SIZE; - object->cluster_size = cluster_size; - - assert(cluster_size >= PAGE_SIZE && - cluster_size % PAGE_SIZE == 0); /* * Wake up anyone waiting for the ready attribute @@ -1176,7 +1230,6 @@ memory_object_change_attributes( boolean_t temporary; boolean_t may_cache; boolean_t invalidate; - memory_object_cluster_size_t cluster_size; memory_object_copy_strategy_t copy_strategy; boolean_t silent_overwrite; boolean_t advisory_pageout; @@ -1195,7 +1248,6 @@ memory_object_change_attributes( #if notyet invalidate = object->invalidate; #endif - cluster_size = object->cluster_size; vm_object_unlock(object); switch (flavor) { @@ -1248,7 +1300,6 @@ memory_object_change_attributes( perf = (memory_object_perf_info_t) attributes; may_cache = perf->may_cache; - cluster_size = round_page_32(perf->cluster_size); break; } @@ -1266,7 +1317,6 @@ memory_object_change_attributes( may_cache = attr->may_cache; copy_strategy = attr->copy_strategy; - cluster_size = page_size; break; } @@ -1284,7 +1334,6 @@ memory_object_change_attributes( copy_strategy = attr->copy_strategy; may_cache = attr->may_cache_object; - cluster_size = attr->cluster_size; temporary = attr->temporary; break; @@ -1313,7 +1362,6 @@ memory_object_change_attributes( may_cache, copy_strategy, temporary, - cluster_size, silent_overwrite, advisory_pageout)); } @@ -1390,7 +1438,7 @@ memory_object_get_attributes( } perf = (memory_object_perf_info_t) attributes; - perf->cluster_size = object->cluster_size; + perf->cluster_size = PAGE_SIZE; perf->may_cache = object->can_persist; *count = MEMORY_OBJECT_PERF_INFO_COUNT; @@ -1425,7 +1473,7 @@ memory_object_get_attributes( attr = (memory_object_attr_info_t) attributes; attr->copy_strategy = object->copy_strategy; - attr->cluster_size = object->cluster_size; + attr->cluster_size = PAGE_SIZE; attr->may_cache_object = object->can_persist; attr->temporary = object->temporary; @@ -1476,7 +1524,9 @@ memory_object_iopl_request( if(*upl_size == 0) { if(offset >= named_entry->size) return(KERN_INVALID_RIGHT); - *upl_size = named_entry->size - offset; + *upl_size = (upl_size_t)(named_entry->size - offset); + if (*upl_size != named_entry->size - offset) + return KERN_INVALID_ARGUMENT; } if(caller_flags & UPL_COPYOUT_FROM) { if((named_entry->protection & VM_PROT_READ) @@ -1602,7 +1652,7 @@ memory_object_upl_request( object = memory_object_control_to_vm_object(control); if (object == VM_OBJECT_NULL) - return (KERN_INVALID_ARGUMENT); + return (KERN_TERMINATED); return vm_object_upl_request(object, offset, @@ -1651,6 +1701,27 @@ memory_object_super_upl_request( cntrl_flags); } +kern_return_t +memory_object_cluster_size(memory_object_control_t control, memory_object_offset_t *start, + vm_size_t *length, uint32_t *io_streaming, memory_object_fault_info_t fault_info) +{ + vm_object_t object; + + object = memory_object_control_to_vm_object(control); + + if (object == VM_OBJECT_NULL || object->paging_offset > *start) + return (KERN_INVALID_ARGUMENT); + + *start -= object->paging_offset; + + vm_object_cluster_size(object, (vm_object_offset_t *)start, length, (vm_object_fault_info_t)fault_info, io_streaming); + + *start += object->paging_offset; + + return (KERN_SUCCESS); +} + + int vm_stat_discard_cleared_reply = 0; int vm_stat_discard_cleared_unset = 0; int vm_stat_discard_cleared_too_late = 0; @@ -1669,11 +1740,12 @@ kern_return_t host_default_memory_manager( host_priv_t host_priv, memory_object_default_t *default_manager, - memory_object_cluster_size_t cluster_size) + __unused memory_object_cluster_size_t cluster_size) { memory_object_default_t current_manager; memory_object_default_t new_manager; memory_object_default_t returned_manager; + kern_return_t result = KERN_SUCCESS; if (host_priv == HOST_PRIV_NULL) return(KERN_INVALID_HOST); @@ -1681,16 +1753,28 @@ host_default_memory_manager( assert(host_priv == &realhost); new_manager = *default_manager; - mutex_lock(&memory_manager_default_lock); + lck_mtx_lock(&memory_manager_default_lock); current_manager = memory_manager_default; + returned_manager = MEMORY_OBJECT_DEFAULT_NULL; if (new_manager == MEMORY_OBJECT_DEFAULT_NULL) { /* * Retrieve the current value. */ - memory_object_default_reference(current_manager); returned_manager = current_manager; + memory_object_default_reference(returned_manager); } else { + + /* + * If this is the first non-null manager, start + * up the internal pager support. + */ + if (current_manager == MEMORY_OBJECT_DEFAULT_NULL) { + result = vm_pageout_internal_start(); + if (result != KERN_SUCCESS) + goto out; + } + /* * Retrieve the current value, * and replace it with the supplied value. @@ -1698,33 +1782,34 @@ host_default_memory_manager( * but we have to take a reference on the new * one. */ - returned_manager = current_manager; memory_manager_default = new_manager; memory_object_default_reference(new_manager); - if (cluster_size % PAGE_SIZE != 0) { -#if 0 - mutex_unlock(&memory_manager_default_lock); - return KERN_INVALID_ARGUMENT; -#else - cluster_size = round_page_32(cluster_size); -#endif - } - memory_manager_default_cluster = cluster_size; - /* * In case anyone's been waiting for a memory * manager to be established, wake them up. */ thread_wakeup((event_t) &memory_manager_default); - } - mutex_unlock(&memory_manager_default_lock); +#ifndef CONFIG_FREEZE + /* + * Now that we have a default pager for anonymous memory, + * reactivate all the throttled pages (i.e. dirty pages with + * no pager). + */ + if (current_manager == MEMORY_OBJECT_DEFAULT_NULL) + { + vm_page_reactivate_all_throttled(); + } +#endif + } + out: + lck_mtx_unlock(&memory_manager_default_lock); *default_manager = returned_manager; - return(KERN_SUCCESS); + return(result); } /* @@ -1736,25 +1821,24 @@ host_default_memory_manager( */ __private_extern__ memory_object_default_t -memory_manager_default_reference( - memory_object_cluster_size_t *cluster_size) +memory_manager_default_reference(void) { memory_object_default_t current_manager; - mutex_lock(&memory_manager_default_lock); + lck_mtx_lock(&memory_manager_default_lock); current_manager = memory_manager_default; while (current_manager == MEMORY_OBJECT_DEFAULT_NULL) { wait_result_t res; - res = thread_sleep_mutex((event_t) &memory_manager_default, - &memory_manager_default_lock, - THREAD_UNINT); + res = lck_mtx_sleep(&memory_manager_default_lock, + LCK_SLEEP_DEFAULT, + (event_t) &memory_manager_default, + THREAD_UNINT); assert(res == THREAD_AWAKENED); current_manager = memory_manager_default; } memory_object_default_reference(current_manager); - *cluster_size = memory_manager_default_cluster; - mutex_unlock(&memory_manager_default_lock); + lck_mtx_unlock(&memory_manager_default_lock); return current_manager; } @@ -1776,18 +1860,18 @@ memory_manager_default_check(void) { memory_object_default_t current; - mutex_lock(&memory_manager_default_lock); + lck_mtx_lock(&memory_manager_default_lock); current = memory_manager_default; if (current == MEMORY_OBJECT_DEFAULT_NULL) { static boolean_t logged; /* initialized to 0 */ boolean_t complain = !logged; logged = TRUE; - mutex_unlock(&memory_manager_default_lock); + lck_mtx_unlock(&memory_manager_default_lock); if (complain) printf("Warning: No default memory manager\n"); return(KERN_FAILURE); } else { - mutex_unlock(&memory_manager_default_lock); + lck_mtx_unlock(&memory_manager_default_lock); return(KERN_SUCCESS); } } @@ -1796,7 +1880,7 @@ __private_extern__ void memory_manager_default_init(void) { memory_manager_default = MEMORY_OBJECT_DEFAULT_NULL; - mutex_init(&memory_manager_default_lock, 0); + lck_mtx_init(&memory_manager_default_lock, &vm_object_lck_grp, &vm_object_lck_attr); } @@ -1849,7 +1933,40 @@ memory_object_range_op( offset_beg, offset_end, ops, - range); + (uint32_t *) range); +} + + +void +memory_object_mark_used( + memory_object_control_t control) +{ + vm_object_t object; + + if (control == NULL) + return; + + object = memory_object_control_to_vm_object(control); + + if (object != VM_OBJECT_NULL) + vm_object_cache_remove(object); +} + + +void +memory_object_mark_unused( + memory_object_control_t control, + __unused boolean_t rage) +{ + vm_object_t object; + + if (control == NULL) + return; + + object = memory_object_control_to_vm_object(control); + + if (object != VM_OBJECT_NULL) + vm_object_cache_add(object); } @@ -1872,6 +1989,37 @@ memory_object_pages_resident( return (KERN_SUCCESS); } +kern_return_t +memory_object_signed( + memory_object_control_t control, + boolean_t is_signed) +{ + vm_object_t object; + + object = memory_object_control_to_vm_object(control); + if (object == VM_OBJECT_NULL) + return KERN_INVALID_ARGUMENT; + + vm_object_lock(object); + object->code_signed = is_signed; + vm_object_unlock(object); + + return KERN_SUCCESS; +} + +boolean_t +memory_object_is_slid( + memory_object_control_t control) +{ + vm_object_t object = VM_OBJECT_NULL; + vm_object_t slide_object = slide_info.slide_object; + + object = memory_object_control_to_vm_object(control); + if (object == VM_OBJECT_NULL) + return FALSE; + + return (object == slide_object); +} static zone_t mem_obj_control_zone; @@ -1882,6 +2030,8 @@ memory_object_control_bootstrap(void) i = (vm_size_t) sizeof (struct memory_object_control); mem_obj_control_zone = zinit (i, 8192*i, 4096, "mem_obj_control"); + zone_change(mem_obj_control_zone, Z_CALLERACCT, FALSE); + zone_change(mem_obj_control_zone, Z_NOENCRYPT, TRUE); return; } @@ -2039,14 +2189,16 @@ kern_return_t memory_object_data_request memory_object_t memory_object, memory_object_offset_t offset, memory_object_cluster_size_t length, - vm_prot_t desired_access + vm_prot_t desired_access, + memory_object_fault_info_t fault_info ) { return (memory_object->mo_pager_ops->memory_object_data_request)( memory_object, offset, length, - desired_access); + desired_access, + fault_info); } /* Routine memory_object_data_return */ @@ -2054,7 +2206,7 @@ kern_return_t memory_object_data_return ( memory_object_t memory_object, memory_object_offset_t offset, - vm_size_t size, + memory_object_cluster_size_t size, memory_object_offset_t *resid_offset, int *io_error, boolean_t dirty, @@ -2078,7 +2230,7 @@ kern_return_t memory_object_data_initialize ( memory_object_t memory_object, memory_object_offset_t offset, - vm_size_t size + memory_object_cluster_size_t size ) { return (memory_object->mo_pager_ops->memory_object_data_initialize)( @@ -2092,7 +2244,7 @@ kern_return_t memory_object_data_unlock ( memory_object_t memory_object, memory_object_offset_t offset, - vm_size_t size, + memory_object_size_t size, vm_prot_t desired_access ) { @@ -2108,7 +2260,7 @@ kern_return_t memory_object_synchronize ( memory_object_t memory_object, memory_object_offset_t offset, - vm_size_t size, + memory_object_size_t size, vm_sync_t sync_flags ) { @@ -2119,16 +2271,61 @@ kern_return_t memory_object_synchronize sync_flags); } -/* Routine memory_object_unmap */ -kern_return_t memory_object_unmap + +/* + * memory_object_map() is called by VM (in vm_map_enter() and its variants) + * each time a "named" VM object gets mapped directly or indirectly + * (copy-on-write mapping). A "named" VM object has an extra reference held + * by the pager to keep it alive until the pager decides that the + * memory object (and its VM object) can be reclaimed. + * VM calls memory_object_last_unmap() (in vm_object_deallocate()) when all + * the mappings of that memory object have been removed. + * + * For a given VM object, calls to memory_object_map() and memory_object_unmap() + * are serialized (through object->mapping_in_progress), to ensure that the + * pager gets a consistent view of the mapping status of the memory object. + * + * This allows the pager to keep track of how many times a memory object + * has been mapped and with which protections, to decide when it can be + * reclaimed. + */ + +/* Routine memory_object_map */ +kern_return_t memory_object_map +( + memory_object_t memory_object, + vm_prot_t prot +) +{ + return (memory_object->mo_pager_ops->memory_object_map)( + memory_object, + prot); +} + +/* Routine memory_object_last_unmap */ +kern_return_t memory_object_last_unmap ( memory_object_t memory_object ) { - return (memory_object->mo_pager_ops->memory_object_unmap)( + return (memory_object->mo_pager_ops->memory_object_last_unmap)( memory_object); } +/* Routine memory_object_data_reclaim */ +kern_return_t memory_object_data_reclaim +( + memory_object_t memory_object, + boolean_t reclaim_backing_store +) +{ + if (memory_object->mo_pager_ops->memory_object_data_reclaim == NULL) + return KERN_NOT_SUPPORTED; + return (memory_object->mo_pager_ops->memory_object_data_reclaim)( + memory_object, + reclaim_backing_store); +} + /* Routine memory_object_create */ kern_return_t memory_object_create (