static void
free_pagelist(PAGELIST_T *pagelist, int actual)
{
vm_page_t *pages;
unsigned int num_pages, i;
vcos_log_trace("free_pagelist - %x, %d", (unsigned int)pagelist, actual);
num_pages =
(pagelist->length + pagelist->offset + PAGE_SIZE - 1) / PAGE_SIZE;
pages = (vm_page_t *)(pagelist->addrs + num_pages);
/* Deal with any partial cache lines (fragments) */
if (pagelist->type >= PAGELIST_READ_WITH_FRAGMENTS) {
FRAGMENTS_T *fragments =
g_fragments_base + (pagelist->type -
PAGELIST_READ_WITH_FRAGMENTS);
int head_bytes, tail_bytes;
if (actual >= 0)
{
/* XXXBSD: might be inefficient */
void *page_address = pmap_mapdev(VM_PAGE_TO_PHYS(pages[0]), PAGE_SIZE*num_pages);
if ((head_bytes = (CACHE_LINE_SIZE - pagelist->offset) & (CACHE_LINE_SIZE - 1)) != 0) {
if (head_bytes > actual)
head_bytes = actual;
memcpy((char *)page_address +
pagelist->offset, fragments->headbuf,
head_bytes);
}
if ((head_bytes < actual) &&
(tail_bytes =
(pagelist->offset + actual) & (CACHE_LINE_SIZE -
1)) != 0) {
memcpy((char *)page_address + PAGE_SIZE*(num_pages - 1) +
((pagelist->offset + actual) & (PAGE_SIZE -
1) & ~(CACHE_LINE_SIZE - 1)),
fragments->tailbuf, tail_bytes);
}
pmap_qremove((vm_offset_t)page_address, PAGE_SIZE*num_pages);
}
mtx_lock(&g_free_fragments_mutex);
*(FRAGMENTS_T **) fragments = g_free_fragments;
g_free_fragments = fragments;
mtx_unlock(&g_free_fragments_mutex);
sema_post(&g_free_fragments_sema);
}
for (i = 0; i < num_pages; i++) {
if (pagelist->type != PAGELIST_WRITE)
vm_page_dirty(pages[i]);
}
vm_page_unhold_pages(pages, num_pages);
free(pagelist, M_VCPAGELIST);
}
/*
* Copy a binary buffer from kernel space to user space.
*
* Returns 0 on success, EFAULT on failure.
*/
int
copyout(const void *kaddr, void *udaddr, size_t len)
{
struct vmspace *vm = curproc->p_vmspace;
struct lwbuf *lwb;
struct lwbuf lwb_cache;
vm_page_t m;
int error;
size_t n;
error = 0;
while (len) {
m = vm_fault_page(&vm->vm_map, trunc_page((vm_offset_t)udaddr),
VM_PROT_READ|VM_PROT_WRITE,
VM_FAULT_NORMAL, &error);
if (error)
break;
n = PAGE_SIZE - ((vm_offset_t)udaddr & PAGE_MASK);
if (n > len)
n = len;
lwb = lwbuf_alloc(m, &lwb_cache);
bcopy(kaddr, (char *)lwbuf_kva(lwb) +
((vm_offset_t)udaddr & PAGE_MASK), n);
len -= n;
udaddr = (char *)udaddr + n;
kaddr = (const char *)kaddr + n;
vm_page_dirty(m);
lwbuf_free(lwb);
vm_page_unhold(m);
}
return (error);
}
/*
* Identify the physical page mapped at the given kernel virtual
* address. Insert this physical page into the given address space at
* the given virtual address, replacing the physical page, if any,
* that already exists there.
*/
static int
vm_pgmoveco(vm_map_t mapa, vm_offset_t kaddr, vm_offset_t uaddr)
{
vm_map_t map = mapa;
vm_page_t kern_pg, user_pg;
vm_object_t uobject;
vm_map_entry_t entry;
vm_pindex_t upindex;
vm_prot_t prot;
boolean_t wired;
KASSERT((uaddr & PAGE_MASK) == 0,
("vm_pgmoveco: uaddr is not page aligned"));
/*
* Herein the physical page is validated and dirtied. It is
* unwired in sf_buf_mext().
*/
kern_pg = PHYS_TO_VM_PAGE(vtophys(kaddr));
kern_pg->valid = VM_PAGE_BITS_ALL;
KASSERT(kern_pg->queue == PQ_NONE && kern_pg->wire_count == 1,
("vm_pgmoveco: kern_pg is not correctly wired"));
if ((vm_map_lookup(&map, uaddr,
VM_PROT_WRITE, &entry, &uobject,
&upindex, &prot, &wired)) != KERN_SUCCESS) {
return(EFAULT);
}
VM_OBJECT_LOCK(uobject);
retry:
if ((user_pg = vm_page_lookup(uobject, upindex)) != NULL) {
if (vm_page_sleep_if_busy(user_pg, TRUE, "vm_pgmoveco"))
goto retry;
vm_page_lock_queues();
pmap_remove_all(user_pg);
vm_page_free(user_pg);
} else {
/*
* Even if a physical page does not exist in the
* object chain's first object, a physical page from a
* backing object may be mapped read only.
*/
if (uobject->backing_object != NULL)
pmap_remove(map->pmap, uaddr, uaddr + PAGE_SIZE);
vm_page_lock_queues();
}
vm_page_insert(kern_pg, uobject, upindex);
vm_page_dirty(kern_pg);
vm_page_unlock_queues();
VM_OBJECT_UNLOCK(uobject);
vm_map_lookup_done(map, entry);
return(KERN_SUCCESS);
}
/*
* Cleanup an XIO so it can be destroyed. The pages associated with the
* XIO are released.
*/
void
xio_release(xio_t xio)
{
int i;
vm_page_t m;
for (i = 0; i < xio->xio_npages; ++i) {
m = xio->xio_pages[i];
if (xio->xio_flags & XIOF_WRITE)
vm_page_dirty(m);
vm_page_unhold(m);
}
xio->xio_offset = 0;
xio->xio_npages = 0;
xio->xio_bytes = 0;
xio->xio_error = ENOBUFS;
}
void
i915_gem_object_do_bit_17_swizzle(struct drm_i915_gem_object *obj)
{
int page_count = obj->base.size >> PAGE_SHIFT;
int i;
if (obj->bit_17 == NULL)
return;
for (i = 0; i < page_count; i++) {
char new_bit_17 = VM_PAGE_TO_PHYS(obj->pages[i]) >> 17;
if ((new_bit_17 & 0x1) !=
(test_bit(i, obj->bit_17) != 0)) {
i915_gem_swizzle_page(obj->pages[i]);
vm_page_dirty(obj->pages[i]);
}
}
}
int ttm_tt_swapout(struct ttm_tt *ttm, vm_object_t persistent_swap_storage)
{
vm_object_t obj;
vm_page_t from_page, to_page;
int i;
BUG_ON(ttm->state != tt_unbound && ttm->state != tt_unpopulated);
BUG_ON(ttm->caching_state != tt_cached);
if (!persistent_swap_storage) {
obj = swap_pager_alloc(NULL,
IDX_TO_OFF(ttm->num_pages), VM_PROT_DEFAULT, 0);
if (obj == NULL) {
pr_err("Failed allocating swap storage\n");
return (-ENOMEM);
}
} else
obj = persistent_swap_storage;
VM_OBJECT_LOCK(obj);
vm_object_pip_add(obj, 1);
for (i = 0; i < ttm->num_pages; ++i) {
from_page = ttm->pages[i];
if (unlikely(from_page == NULL))
continue;
to_page = vm_page_grab(obj, i, VM_ALLOC_NORMAL |
VM_ALLOC_RETRY);
pmap_copy_page(VM_PAGE_TO_PHYS(from_page),
VM_PAGE_TO_PHYS(to_page));
to_page->valid = VM_PAGE_BITS_ALL;
vm_page_dirty(to_page);
vm_page_wakeup(to_page);
}
vm_object_pip_wakeup(obj);
VM_OBJECT_UNLOCK(obj);
ttm->bdev->driver->ttm_tt_unpopulate(ttm);
ttm->swap_storage = obj;
ttm->page_flags |= TTM_PAGE_FLAG_SWAPPED;
if (persistent_swap_storage)
ttm->page_flags |= TTM_PAGE_FLAG_PERSISTENT_SWAP;
return 0;
}
int ttm_tt_swapout(struct ttm_tt *ttm, vm_object_t persistent_swap_storage)
{
vm_object_t obj;
vm_page_t from_page, to_page;
int i;
MPASS(ttm->state == tt_unbound || ttm->state == tt_unpopulated);
MPASS(ttm->caching_state == tt_cached);
if (persistent_swap_storage == NULL) {
obj = vm_pager_allocate(OBJT_SWAP, NULL,
IDX_TO_OFF(ttm->num_pages), VM_PROT_DEFAULT, 0,
curthread->td_ucred);
if (obj == NULL) {
printf("[TTM] Failed allocating swap storage\n");
return (-ENOMEM);
}
} else
obj = persistent_swap_storage;
VM_OBJECT_WLOCK(obj);
vm_object_pip_add(obj, 1);
for (i = 0; i < ttm->num_pages; ++i) {
from_page = ttm->pages[i];
if (unlikely(from_page == NULL))
continue;
to_page = vm_page_grab(obj, i, VM_ALLOC_NORMAL);
pmap_copy_page(from_page, to_page);
to_page->valid = VM_PAGE_BITS_ALL;
vm_page_dirty(to_page);
vm_page_xunbusy(to_page);
}
vm_object_pip_wakeup(obj);
VM_OBJECT_WUNLOCK(obj);
ttm->bdev->driver->ttm_tt_unpopulate(ttm);
ttm->swap_storage = obj;
ttm->page_flags |= TTM_PAGE_FLAG_SWAPPED;
if (persistent_swap_storage != NULL)
ttm->page_flags |= TTM_PAGE_FLAG_PERSISTENT_SWAP;
return (0);
}
/*
* Cleanup an XIO so it can be destroyed. The pages associated with the
* XIO are released.
*/
void
xio_release(xio_t xio)
{
int i;
vm_page_t m;
lwkt_gettoken(&vm_token);
crit_enter();
for (i = 0; i < xio->xio_npages; ++i) {
m = xio->xio_pages[i];
if (xio->xio_flags & XIOF_WRITE)
vm_page_dirty(m);
vm_page_unhold(m);
}
crit_exit();
lwkt_reltoken(&vm_token);
xio->xio_offset = 0;
xio->xio_npages = 0;
xio->xio_bytes = 0;
xio->xio_error = ENOBUFS;
}
int
vm_fault_hold(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type,
int fault_flags, vm_page_t *m_hold)
{
vm_prot_t prot;
long ahead, behind;
int alloc_req, era, faultcount, nera, reqpage, result;
boolean_t growstack, is_first_object_locked, wired;
int map_generation;
vm_object_t next_object;
vm_page_t marray[VM_FAULT_READ_MAX];
int hardfault;
struct faultstate fs;
struct vnode *vp;
int locked, error;
hardfault = 0;
growstack = TRUE;
PCPU_INC(cnt.v_vm_faults);
fs.vp = NULL;
faultcount = reqpage = 0;
RetryFault:;
/*
* Find the backing store object and offset into it to begin the
* search.
*/
fs.map = map;
result = vm_map_lookup(&fs.map, vaddr, fault_type, &fs.entry,
&fs.first_object, &fs.first_pindex, &prot, &wired);
if (result != KERN_SUCCESS) {
if (growstack && result == KERN_INVALID_ADDRESS &&
map != kernel_map) {
result = vm_map_growstack(curproc, vaddr);
if (result != KERN_SUCCESS)
return (KERN_FAILURE);
growstack = FALSE;
goto RetryFault;
}
return (result);
}
map_generation = fs.map->timestamp;
if (fs.entry->eflags & MAP_ENTRY_NOFAULT) {
panic("vm_fault: fault on nofault entry, addr: %lx",
(u_long)vaddr);
}
/*
* Make a reference to this object to prevent its disposal while we
* are messing with it. Once we have the reference, the map is free
* to be diddled. Since objects reference their shadows (and copies),
* they will stay around as well.
*
* Bump the paging-in-progress count to prevent size changes (e.g.
* truncation operations) during I/O. This must be done after
* obtaining the vnode lock in order to avoid possible deadlocks.
*/
VM_OBJECT_WLOCK(fs.first_object);
vm_object_reference_locked(fs.first_object);
vm_object_pip_add(fs.first_object, 1);
fs.lookup_still_valid = TRUE;
if (wired)
fault_type = prot | (fault_type & VM_PROT_COPY);
fs.first_m = NULL;
/*
* Search for the page at object/offset.
*/
fs.object = fs.first_object;
fs.pindex = fs.first_pindex;
while (TRUE) {
/*
* If the object is dead, we stop here
*/
if (fs.object->flags & OBJ_DEAD) {
unlock_and_deallocate(&fs);
return (KERN_PROTECTION_FAILURE);
}
/*
* See if page is resident
*/
fs.m = vm_page_lookup(fs.object, fs.pindex);
if (fs.m != NULL) {
/*
* check for page-based copy on write.
* We check fs.object == fs.first_object so
* as to ensure the legacy COW mechanism is
* used when the page in question is part of
* a shadow object. Otherwise, vm_page_cowfault()
* removes the page from the backing object,
* which is not what we want.
*/
vm_page_lock(fs.m);
if ((fs.m->cow) &&
(fault_type & VM_PROT_WRITE) &&
(fs.object == fs.first_object)) {
vm_page_cowfault(fs.m);
unlock_and_deallocate(&fs);
goto RetryFault;
}
/*
* Wait/Retry if the page is busy. We have to do this
* if the page is busy via either VPO_BUSY or
* vm_page_t->busy because the vm_pager may be using
* vm_page_t->busy for pageouts ( and even pageins if
* it is the vnode pager ), and we could end up trying
* to pagein and pageout the same page simultaneously.
*
* We can theoretically allow the busy case on a read
* fault if the page is marked valid, but since such
* pages are typically already pmap'd, putting that
* special case in might be more effort then it is
* worth. We cannot under any circumstances mess
* around with a vm_page_t->busy page except, perhaps,
* to pmap it.
*/
if ((fs.m->oflags & VPO_BUSY) || fs.m->busy) {
/*
* Reference the page before unlocking and
* sleeping so that the page daemon is less
* likely to reclaim it.
*/
vm_page_aflag_set(fs.m, PGA_REFERENCED);
vm_page_unlock(fs.m);
if (fs.object != fs.first_object) {
if (!VM_OBJECT_TRYWLOCK(
fs.first_object)) {
VM_OBJECT_WUNLOCK(fs.object);
VM_OBJECT_WLOCK(fs.first_object);
VM_OBJECT_WLOCK(fs.object);
}
vm_page_lock(fs.first_m);
vm_page_free(fs.first_m);
vm_page_unlock(fs.first_m);
vm_object_pip_wakeup(fs.first_object);
VM_OBJECT_WUNLOCK(fs.first_object);
fs.first_m = NULL;
}
unlock_map(&fs);
if (fs.m == vm_page_lookup(fs.object,
fs.pindex)) {
vm_page_sleep_if_busy(fs.m, TRUE,
"vmpfw");
}
vm_object_pip_wakeup(fs.object);
VM_OBJECT_WUNLOCK(fs.object);
PCPU_INC(cnt.v_intrans);
vm_object_deallocate(fs.first_object);
goto RetryFault;
}
vm_page_remque(fs.m);
vm_page_unlock(fs.m);
/*
* Mark page busy for other processes, and the
* pagedaemon. If it still isn't completely valid
* (readable), jump to readrest, else break-out ( we
* found the page ).
*/
vm_page_busy(fs.m);
if (fs.m->valid != VM_PAGE_BITS_ALL)
goto readrest;
break;
}
/*
* Page is not resident, If this is the search termination
* or the pager might contain the page, allocate a new page.
*/
if (TRYPAGER || fs.object == fs.first_object) {
if (fs.pindex >= fs.object->size) {
unlock_and_deallocate(&fs);
return (KERN_PROTECTION_FAILURE);
}
/*
* Allocate a new page for this object/offset pair.
*
* Unlocked read of the p_flag is harmless. At
* worst, the P_KILLED might be not observed
* there, and allocation can fail, causing
* restart and new reading of the p_flag.
*/
fs.m = NULL;
if (!vm_page_count_severe() || P_KILLED(curproc)) {
#if VM_NRESERVLEVEL > 0
if ((fs.object->flags & OBJ_COLORED) == 0) {
fs.object->flags |= OBJ_COLORED;
fs.object->pg_color = atop(vaddr) -
fs.pindex;
}
#endif
alloc_req = P_KILLED(curproc) ?
VM_ALLOC_SYSTEM : VM_ALLOC_NORMAL;
if (fs.object->type != OBJT_VNODE &&
fs.object->backing_object == NULL)
alloc_req |= VM_ALLOC_ZERO;
fs.m = vm_page_alloc(fs.object, fs.pindex,
alloc_req);
}
if (fs.m == NULL) {
unlock_and_deallocate(&fs);
VM_WAITPFAULT;
goto RetryFault;
} else if (fs.m->valid == VM_PAGE_BITS_ALL)
break;
}
readrest:
/*
* We have found a valid page or we have allocated a new page.
* The page thus may not be valid or may not be entirely
* valid.
*
* Attempt to fault-in the page if there is a chance that the
* pager has it, and potentially fault in additional pages
* at the same time.
*/
if (TRYPAGER) {
int rv;
u_char behavior = vm_map_entry_behavior(fs.entry);
if (behavior == MAP_ENTRY_BEHAV_RANDOM ||
P_KILLED(curproc)) {
behind = 0;
ahead = 0;
} else if (behavior == MAP_ENTRY_BEHAV_SEQUENTIAL) {
behind = 0;
ahead = atop(fs.entry->end - vaddr) - 1;
if (ahead > VM_FAULT_READ_AHEAD_MAX)
ahead = VM_FAULT_READ_AHEAD_MAX;
if (fs.pindex == fs.entry->next_read)
vm_fault_cache_behind(&fs,
VM_FAULT_READ_MAX);
} else {
/*
* If this is a sequential page fault, then
* arithmetically increase the number of pages
* in the read-ahead window. Otherwise, reset
* the read-ahead window to its smallest size.
*/
behind = atop(vaddr - fs.entry->start);
if (behind > VM_FAULT_READ_BEHIND)
behind = VM_FAULT_READ_BEHIND;
ahead = atop(fs.entry->end - vaddr) - 1;
era = fs.entry->read_ahead;
if (fs.pindex == fs.entry->next_read) {
nera = era + behind;
if (nera > VM_FAULT_READ_AHEAD_MAX)
nera = VM_FAULT_READ_AHEAD_MAX;
behind = 0;
if (ahead > nera)
ahead = nera;
if (era == VM_FAULT_READ_AHEAD_MAX)
vm_fault_cache_behind(&fs,
VM_FAULT_CACHE_BEHIND);
} else if (ahead > VM_FAULT_READ_AHEAD_MIN)
ahead = VM_FAULT_READ_AHEAD_MIN;
if (era != ahead)
fs.entry->read_ahead = ahead;
}
/*
* Call the pager to retrieve the data, if any, after
* releasing the lock on the map. We hold a ref on
* fs.object and the pages are VPO_BUSY'd.
*/
unlock_map(&fs);
if (fs.object->type == OBJT_VNODE) {
vp = fs.object->handle;
if (vp == fs.vp)
goto vnode_locked;
else if (fs.vp != NULL) {
vput(fs.vp);
fs.vp = NULL;
}
locked = VOP_ISLOCKED(vp);
if (locked != LK_EXCLUSIVE)
locked = LK_SHARED;
/* Do not sleep for vnode lock while fs.m is busy */
error = vget(vp, locked | LK_CANRECURSE |
LK_NOWAIT, curthread);
if (error != 0) {
vhold(vp);
release_page(&fs);
unlock_and_deallocate(&fs);
error = vget(vp, locked | LK_RETRY |
LK_CANRECURSE, curthread);
vdrop(vp);
fs.vp = vp;
KASSERT(error == 0,
("vm_fault: vget failed"));
goto RetryFault;
}
fs.vp = vp;
}
vnode_locked:
KASSERT(fs.vp == NULL || !fs.map->system_map,
("vm_fault: vnode-backed object mapped by system map"));
/*
* now we find out if any other pages should be paged
* in at this time this routine checks to see if the
* pages surrounding this fault reside in the same
* object as the page for this fault. If they do,
* then they are faulted in also into the object. The
* array "marray" returned contains an array of
* vm_page_t structs where one of them is the
* vm_page_t passed to the routine. The reqpage
* return value is the index into the marray for the
* vm_page_t passed to the routine.
*
* fs.m plus the additional pages are VPO_BUSY'd.
*/
faultcount = vm_fault_additional_pages(
fs.m, behind, ahead, marray, &reqpage);
rv = faultcount ?
vm_pager_get_pages(fs.object, marray, faultcount,
reqpage) : VM_PAGER_FAIL;
if (rv == VM_PAGER_OK) {
/*
* Found the page. Leave it busy while we play
* with it.
*/
/*
* Relookup in case pager changed page. Pager
* is responsible for disposition of old page
* if moved.
*/
fs.m = vm_page_lookup(fs.object, fs.pindex);
if (!fs.m) {
unlock_and_deallocate(&fs);
goto RetryFault;
}
hardfault++;
break; /* break to PAGE HAS BEEN FOUND */
}
/*
* Remove the bogus page (which does not exist at this
* object/offset); before doing so, we must get back
* our object lock to preserve our invariant.
*
* Also wake up any other process that may want to bring
* in this page.
*
* If this is the top-level object, we must leave the
* busy page to prevent another process from rushing
* past us, and inserting the page in that object at
* the same time that we are.
*/
if (rv == VM_PAGER_ERROR)
printf("vm_fault: pager read error, pid %d (%s)\n",
curproc->p_pid, curproc->p_comm);
/*
* Data outside the range of the pager or an I/O error
*/
/*
* XXX - the check for kernel_map is a kludge to work
* around having the machine panic on a kernel space
* fault w/ I/O error.
*/
if (((fs.map != kernel_map) && (rv == VM_PAGER_ERROR)) ||
(rv == VM_PAGER_BAD)) {
vm_page_lock(fs.m);
vm_page_free(fs.m);
vm_page_unlock(fs.m);
fs.m = NULL;
unlock_and_deallocate(&fs);
return ((rv == VM_PAGER_ERROR) ? KERN_FAILURE : KERN_PROTECTION_FAILURE);
}
if (fs.object != fs.first_object) {
vm_page_lock(fs.m);
vm_page_free(fs.m);
vm_page_unlock(fs.m);
fs.m = NULL;
/*
* XXX - we cannot just fall out at this
* point, m has been freed and is invalid!
*/
}
}
/*
* We get here if the object has default pager (or unwiring)
* or the pager doesn't have the page.
*/
if (fs.object == fs.first_object)
fs.first_m = fs.m;
/*
* Move on to the next object. Lock the next object before
* unlocking the current one.
*/
fs.pindex += OFF_TO_IDX(fs.object->backing_object_offset);
next_object = fs.object->backing_object;
if (next_object == NULL) {
/*
* If there's no object left, fill the page in the top
* object with zeros.
*/
if (fs.object != fs.first_object) {
vm_object_pip_wakeup(fs.object);
VM_OBJECT_WUNLOCK(fs.object);
fs.object = fs.first_object;
fs.pindex = fs.first_pindex;
fs.m = fs.first_m;
VM_OBJECT_WLOCK(fs.object);
}
fs.first_m = NULL;
/*
* Zero the page if necessary and mark it valid.
*/
if ((fs.m->flags & PG_ZERO) == 0) {
pmap_zero_page(fs.m);
} else {
PCPU_INC(cnt.v_ozfod);
}
PCPU_INC(cnt.v_zfod);
fs.m->valid = VM_PAGE_BITS_ALL;
break; /* break to PAGE HAS BEEN FOUND */
} else {
KASSERT(fs.object != next_object,
("object loop %p", next_object));
VM_OBJECT_WLOCK(next_object);
vm_object_pip_add(next_object, 1);
if (fs.object != fs.first_object)
vm_object_pip_wakeup(fs.object);
VM_OBJECT_WUNLOCK(fs.object);
fs.object = next_object;
}
}
KASSERT((fs.m->oflags & VPO_BUSY) != 0,
("vm_fault: not busy after main loop"));
/*
* PAGE HAS BEEN FOUND. [Loop invariant still holds -- the object lock
* is held.]
*/
/*
* If the page is being written, but isn't already owned by the
* top-level object, we have to copy it into a new page owned by the
* top-level object.
*/
if (fs.object != fs.first_object) {
/*
* We only really need to copy if we want to write it.
*/
if ((fault_type & (VM_PROT_COPY | VM_PROT_WRITE)) != 0) {
/*
* This allows pages to be virtually copied from a
* backing_object into the first_object, where the
* backing object has no other refs to it, and cannot
* gain any more refs. Instead of a bcopy, we just
* move the page from the backing object to the
* first object. Note that we must mark the page
* dirty in the first object so that it will go out
* to swap when needed.
*/
is_first_object_locked = FALSE;
if (
/*
* Only one shadow object
*/
(fs.object->shadow_count == 1) &&
/*
* No COW refs, except us
*/
(fs.object->ref_count == 1) &&
/*
* No one else can look this object up
*/
(fs.object->handle == NULL) &&
/*
* No other ways to look the object up
*/
((fs.object->type == OBJT_DEFAULT) ||
(fs.object->type == OBJT_SWAP)) &&
(is_first_object_locked = VM_OBJECT_TRYWLOCK(fs.first_object)) &&
/*
* We don't chase down the shadow chain
*/
fs.object == fs.first_object->backing_object) {
/*
* get rid of the unnecessary page
*/
vm_page_lock(fs.first_m);
vm_page_free(fs.first_m);
vm_page_unlock(fs.first_m);
/*
* grab the page and put it into the
* process'es object. The page is
* automatically made dirty.
*/
vm_page_lock(fs.m);
vm_page_rename(fs.m, fs.first_object, fs.first_pindex);
vm_page_unlock(fs.m);
vm_page_busy(fs.m);
fs.first_m = fs.m;
fs.m = NULL;
PCPU_INC(cnt.v_cow_optim);
} else {
/*
* Oh, well, lets copy it.
*/
pmap_copy_page(fs.m, fs.first_m);
fs.first_m->valid = VM_PAGE_BITS_ALL;
if (wired && (fault_flags &
VM_FAULT_CHANGE_WIRING) == 0) {
vm_page_lock(fs.first_m);
vm_page_wire(fs.first_m);
vm_page_unlock(fs.first_m);
vm_page_lock(fs.m);
vm_page_unwire(fs.m, FALSE);
vm_page_unlock(fs.m);
}
/*
* We no longer need the old page or object.
*/
release_page(&fs);
}
/*
* fs.object != fs.first_object due to above
* conditional
*/
vm_object_pip_wakeup(fs.object);
VM_OBJECT_WUNLOCK(fs.object);
/*
* Only use the new page below...
*/
fs.object = fs.first_object;
fs.pindex = fs.first_pindex;
fs.m = fs.first_m;
if (!is_first_object_locked)
VM_OBJECT_WLOCK(fs.object);
PCPU_INC(cnt.v_cow_faults);
curthread->td_cow++;
} else {
prot &= ~VM_PROT_WRITE;
}
}
/*
* We must verify that the maps have not changed since our last
* lookup.
*/
if (!fs.lookup_still_valid) {
vm_object_t retry_object;
vm_pindex_t retry_pindex;
vm_prot_t retry_prot;
if (!vm_map_trylock_read(fs.map)) {
release_page(&fs);
unlock_and_deallocate(&fs);
goto RetryFault;
}
fs.lookup_still_valid = TRUE;
if (fs.map->timestamp != map_generation) {
result = vm_map_lookup_locked(&fs.map, vaddr, fault_type,
&fs.entry, &retry_object, &retry_pindex, &retry_prot, &wired);
/*
* If we don't need the page any longer, put it on the inactive
* list (the easiest thing to do here). If no one needs it,
* pageout will grab it eventually.
*/
if (result != KERN_SUCCESS) {
release_page(&fs);
unlock_and_deallocate(&fs);
/*
* If retry of map lookup would have blocked then
* retry fault from start.
*/
if (result == KERN_FAILURE)
goto RetryFault;
return (result);
}
if ((retry_object != fs.first_object) ||
(retry_pindex != fs.first_pindex)) {
release_page(&fs);
unlock_and_deallocate(&fs);
goto RetryFault;
}
/*
* Check whether the protection has changed or the object has
* been copied while we left the map unlocked. Changing from
* read to write permission is OK - we leave the page
* write-protected, and catch the write fault. Changing from
* write to read permission means that we can't mark the page
* write-enabled after all.
*/
prot &= retry_prot;
}
}
/*
* If the page was filled by a pager, update the map entry's
* last read offset. Since the pager does not return the
* actual set of pages that it read, this update is based on
* the requested set. Typically, the requested and actual
* sets are the same.
*
* XXX The following assignment modifies the map
* without holding a write lock on it.
*/
if (hardfault)
fs.entry->next_read = fs.pindex + faultcount - reqpage;
if ((prot & VM_PROT_WRITE) != 0 ||
(fault_flags & VM_FAULT_DIRTY) != 0) {
vm_object_set_writeable_dirty(fs.object);
/*
* If this is a NOSYNC mmap we do not want to set VPO_NOSYNC
* if the page is already dirty to prevent data written with
* the expectation of being synced from not being synced.
* Likewise if this entry does not request NOSYNC then make
* sure the page isn't marked NOSYNC. Applications sharing
* data should use the same flags to avoid ping ponging.
*/
if (fs.entry->eflags & MAP_ENTRY_NOSYNC) {
if (fs.m->dirty == 0)
fs.m->oflags |= VPO_NOSYNC;
} else {
fs.m->oflags &= ~VPO_NOSYNC;
}
/*
* If the fault is a write, we know that this page is being
* written NOW so dirty it explicitly to save on
* pmap_is_modified() calls later.
*
* Also tell the backing pager, if any, that it should remove
* any swap backing since the page is now dirty.
*/
if (((fault_type & VM_PROT_WRITE) != 0 &&
(fault_flags & VM_FAULT_CHANGE_WIRING) == 0) ||
(fault_flags & VM_FAULT_DIRTY) != 0) {
vm_page_dirty(fs.m);
vm_pager_page_unswapped(fs.m);
}
}
/*
* Page had better still be busy
*/
KASSERT(fs.m->oflags & VPO_BUSY,
("vm_fault: page %p not busy!", fs.m));
/*
* Page must be completely valid or it is not fit to
* map into user space. vm_pager_get_pages() ensures this.
*/
KASSERT(fs.m->valid == VM_PAGE_BITS_ALL,
("vm_fault: page %p partially invalid", fs.m));
VM_OBJECT_WUNLOCK(fs.object);
/*
* Put this page into the physical map. We had to do the unlock above
* because pmap_enter() may sleep. We don't put the page
* back on the active queue until later so that the pageout daemon
* won't find it (yet).
*/
pmap_enter(fs.map->pmap, vaddr, fault_type, fs.m, prot, wired);
if ((fault_flags & VM_FAULT_CHANGE_WIRING) == 0 && wired == 0)
vm_fault_prefault(fs.map->pmap, vaddr, fs.entry);
VM_OBJECT_WLOCK(fs.object);
vm_page_lock(fs.m);
/*
* If the page is not wired down, then put it where the pageout daemon
* can find it.
*/
if (fault_flags & VM_FAULT_CHANGE_WIRING) {
if (wired)
vm_page_wire(fs.m);
else
vm_page_unwire(fs.m, 1);
} else
vm_page_activate(fs.m);
if (m_hold != NULL) {
*m_hold = fs.m;
vm_page_hold(fs.m);
}
vm_page_unlock(fs.m);
vm_page_wakeup(fs.m);
/*
* Unlock everything, and return
*/
unlock_and_deallocate(&fs);
if (hardfault) {
PCPU_INC(cnt.v_io_faults);
curthread->td_ru.ru_majflt++;
} else
curthread->td_ru.ru_minflt++;
return (KERN_SUCCESS);
}
/*
* Hold each of the physical pages that are mapped by the specified range of
* virtual addresses, ["addr", "addr" + "len"), if those mappings are valid
* and allow the specified types of access, "prot". If all of the implied
* pages are successfully held, then the number of held pages is returned
* together with pointers to those pages in the array "ma". However, if any
* of the pages cannot be held, -1 is returned.
*/
int
vm_fault_quick_hold_pages(vm_map_t map, vm_offset_t addr, vm_size_t len,
vm_prot_t prot, vm_page_t *ma, int max_count)
{
vm_offset_t end, va;
vm_page_t *mp;
int count;
boolean_t pmap_failed;
if (len == 0)
return (0);
end = round_page(addr + len);
addr = trunc_page(addr);
/*
* Check for illegal addresses.
*/
if (addr < vm_map_min(map) || addr > end || end > vm_map_max(map))
return (-1);
count = howmany(end - addr, PAGE_SIZE);
if (count > max_count)
panic("vm_fault_quick_hold_pages: count > max_count");
/*
* Most likely, the physical pages are resident in the pmap, so it is
* faster to try pmap_extract_and_hold() first.
*/
pmap_failed = FALSE;
for (mp = ma, va = addr; va < end; mp++, va += PAGE_SIZE) {
*mp = pmap_extract_and_hold(map->pmap, va, prot);
if (*mp == NULL)
pmap_failed = TRUE;
else if ((prot & VM_PROT_WRITE) != 0 &&
(*mp)->dirty != VM_PAGE_BITS_ALL) {
/*
* Explicitly dirty the physical page. Otherwise, the
* caller's changes may go unnoticed because they are
* performed through an unmanaged mapping or by a DMA
* operation.
*
* The object lock is not held here.
* See vm_page_clear_dirty_mask().
*/
vm_page_dirty(*mp);
}
}
if (pmap_failed) {
/*
* One or more pages could not be held by the pmap. Either no
* page was mapped at the specified virtual address or that
* mapping had insufficient permissions. Attempt to fault in
* and hold these pages.
*/
for (mp = ma, va = addr; va < end; mp++, va += PAGE_SIZE)
if (*mp == NULL && vm_fault_hold(map, va, prot,
VM_FAULT_NORMAL, mp) != KERN_SUCCESS)
goto error;
}
return (count);
error:
for (mp = ma; mp < ma + count; mp++)
if (*mp != NULL) {
vm_page_lock(*mp);
vm_page_unhold(*mp);
vm_page_unlock(*mp);
}
return (-1);
}
/*
* This routine takes a user's map, array of pages, number of pages, and flags
* and then does the following:
* - validate that the user has access to those pages (flags indicates read
* or write) - if not fail
* - validate that count is enough to hold range number of pages - if not fail
* - fault in any non-resident pages
* - if the user is doing a read force a write fault for any COWed pages
* - if the user is doing a read mark all pages as dirty
* - hold all pages
*/
int
vm_fault_hold_user_pages(vm_map_t map, vm_offset_t addr, vm_page_t *mp,
int count, vm_prot_t prot)
{
vm_offset_t end, va;
int faults, rv;
pmap_t pmap;
vm_page_t m, *pages;
pmap = vm_map_pmap(map);
pages = mp;
addr &= ~PAGE_MASK;
/*
* Check that virtual address range is legal
* This check is somewhat bogus as on some architectures kernel
* and user do not share VA - however, it appears that all FreeBSD
* architectures define it
*/
end = addr + (count * PAGE_SIZE);
if (end > VM_MAXUSER_ADDRESS) {
log(LOG_WARNING, "bad address passed to vm_fault_hold_user_pages");
return (EFAULT);
}
/*
* First optimistically assume that all pages are resident
* (and R/W if for write) if so just mark pages as held (and
* dirty if for write) and return
*/
vm_page_lock_queues();
for (pages = mp, faults = 0, va = addr; va < end;
va += PAGE_SIZE, pages++) {
/*
* page queue mutex is recursable so this is OK
* it would be really nice if we had an unlocked
* version of this so we were only acquiring the
* pmap lock 1 time as opposed to potentially
* many dozens of times
*/
*pages = m = pmap_extract_and_hold(pmap, va, prot);
if (m == NULL) {
faults++;
continue;
}
/*
* Preemptively mark dirty - the pages
* will never have the modified bit set if
* they are only changed via DMA
*/
if (prot & VM_PROT_WRITE)
vm_page_dirty(m);
}
vm_page_unlock_queues();
if (faults == 0)
return (0);
/*
* Pages either have insufficient permissions or are not present
* trigger a fault where neccessary
*
*/
rv = 0;
for (pages = mp, va = addr; va < end; va += PAGE_SIZE, pages++) {
/*
* Account for a very narrow race where the page may be
* taken away from us before it is held
*/
while (*pages == NULL) {
rv = vm_fault(map, va, prot,
(prot & VM_PROT_WRITE) ? VM_FAULT_DIRTY : VM_FAULT_NORMAL);
if (rv)
goto error;
*pages = pmap_extract_and_hold(pmap, va, prot);
}
}
return (0);
error:
log(LOG_WARNING,
"vm_fault bad return rv=%d va=0x%zx\n", rv, va);
vm_page_lock_queues();
for (pages = mp, va = addr; va < end; va += PAGE_SIZE, pages++)
if (*pages) {
vm_page_unhold(*pages);
*pages = NULL;
}
vm_page_unlock_queues();
return (EFAULT);
}
static int
shm_dotruncate(struct shmfd *shmfd, off_t length)
{
vm_object_t object;
vm_page_t m, ma[1];
vm_pindex_t idx, nobjsize;
vm_ooffset_t delta;
int base, rv;
object = shmfd->shm_object;
VM_OBJECT_LOCK(object);
if (length == shmfd->shm_size) {
VM_OBJECT_UNLOCK(object);
return (0);
}
nobjsize = OFF_TO_IDX(length + PAGE_MASK);
/* Are we shrinking? If so, trim the end. */
if (length < shmfd->shm_size) {
/*
* Disallow any requests to shrink the size if this
* object is mapped into the kernel.
*/
if (shmfd->shm_kmappings > 0) {
VM_OBJECT_UNLOCK(object);
return (EBUSY);
}
/*
* Zero the truncated part of the last page.
*/
base = length & PAGE_MASK;
if (base != 0) {
idx = OFF_TO_IDX(length);
retry:
m = vm_page_lookup(object, idx);
if (m != NULL) {
if ((m->oflags & VPO_BUSY) != 0 ||
m->busy != 0) {
vm_page_sleep(m, "shmtrc");
goto retry;
}
} else if (vm_pager_has_page(object, idx, NULL, NULL)) {
m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
if (m == NULL) {
VM_OBJECT_UNLOCK(object);
VM_WAIT;
VM_OBJECT_LOCK(object);
goto retry;
} else if (m->valid != VM_PAGE_BITS_ALL) {
ma[0] = m;
rv = vm_pager_get_pages(object, ma, 1,
0);
m = vm_page_lookup(object, idx);
} else
/* A cached page was reactivated. */
rv = VM_PAGER_OK;
vm_page_lock(m);
if (rv == VM_PAGER_OK) {
vm_page_deactivate(m);
vm_page_unlock(m);
vm_page_wakeup(m);
} else {
vm_page_free(m);
vm_page_unlock(m);
VM_OBJECT_UNLOCK(object);
return (EIO);
}
}
if (m != NULL) {
pmap_zero_page_area(m, base, PAGE_SIZE - base);
KASSERT(m->valid == VM_PAGE_BITS_ALL,
("shm_dotruncate: page %p is invalid", m));
vm_page_dirty(m);
vm_pager_page_unswapped(m);
}
}
delta = ptoa(object->size - nobjsize);
/* Toss in memory pages. */
if (nobjsize < object->size)
vm_object_page_remove(object, nobjsize, object->size,
0);
/* Toss pages from swap. */
if (object->type == OBJT_SWAP)
swap_pager_freespace(object, nobjsize, delta);
/* Free the swap accounted for shm */
swap_release_by_cred(delta, object->cred);
object->charge -= delta;
} else {
/* Attempt to reserve the swap */
delta = ptoa(nobjsize - object->size);
if (!swap_reserve_by_cred(delta, object->cred)) {
VM_OBJECT_UNLOCK(object);
return (ENOMEM);
}
object->charge += delta;
}
shmfd->shm_size = length;
mtx_lock(&shm_timestamp_lock);
vfs_timestamp(&shmfd->shm_ctime);
shmfd->shm_mtime = shmfd->shm_ctime;
mtx_unlock(&shm_timestamp_lock);
object->size = nobjsize;
VM_OBJECT_UNLOCK(object);
return (0);
}
static int
tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
{
vm_pindex_t idx;
vm_page_t vpg, tpg;
vm_offset_t offset;
off_t addr;
size_t tlen;
int error, rv;
error = 0;
addr = uio->uio_offset;
idx = OFF_TO_IDX(addr);
offset = addr & PAGE_MASK;
tlen = MIN(PAGE_SIZE - offset, len);
if ((vobj == NULL) ||
(vobj->resident_page_count == 0 && vobj->cache == NULL)) {
vpg = NULL;
goto nocache;
}
VM_OBJECT_LOCK(vobj);
lookupvpg:
if (((vpg = vm_page_lookup(vobj, idx)) != NULL) &&
vm_page_is_valid(vpg, offset, tlen)) {
if ((vpg->oflags & VPO_BUSY) != 0) {
/*
* Reference the page before unlocking and sleeping so
* that the page daemon is less likely to reclaim it.
*/
vm_page_reference(vpg);
vm_page_sleep(vpg, "tmfsmw");
goto lookupvpg;
}
vm_page_busy(vpg);
vm_page_undirty(vpg);
VM_OBJECT_UNLOCK(vobj);
error = uiomove_fromphys(&vpg, offset, tlen, uio);
} else {
if (__predict_false(vobj->cache != NULL))
vm_page_cache_free(vobj, idx, idx + 1);
VM_OBJECT_UNLOCK(vobj);
vpg = NULL;
}
nocache:
VM_OBJECT_LOCK(tobj);
tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
if (tpg->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
rv = vm_pager_get_pages(tobj, &tpg, 1, 0);
if (rv != VM_PAGER_OK) {
vm_page_lock(tpg);
vm_page_free(tpg);
vm_page_unlock(tpg);
error = EIO;
goto out;
}
} else
vm_page_zero_invalid(tpg, TRUE);
}
VM_OBJECT_UNLOCK(tobj);
if (vpg == NULL)
error = uiomove_fromphys(&tpg, offset, tlen, uio);
else {
KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid"));
pmap_copy_page(vpg, tpg);
}
VM_OBJECT_LOCK(tobj);
if (error == 0) {
KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
("parts of tpg invalid"));
vm_page_dirty(tpg);
}
vm_page_lock(tpg);
vm_page_unwire(tpg, TRUE);
vm_page_unlock(tpg);
vm_page_wakeup(tpg);
out:
VM_OBJECT_UNLOCK(tobj);
if (vpg != NULL) {
VM_OBJECT_LOCK(vobj);
vm_page_wakeup(vpg);
VM_OBJECT_UNLOCK(vobj);
}
return (error);
}
static int
uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
{
vm_page_t m;
vm_pindex_t idx;
size_t tlen;
int error, offset, rv;
idx = OFF_TO_IDX(uio->uio_offset);
offset = uio->uio_offset & PAGE_MASK;
tlen = MIN(PAGE_SIZE - offset, len);
VM_OBJECT_WLOCK(obj);
/*
* Parallel reads of the page content from disk are prevented
* by exclusive busy.
*
* Although the tmpfs vnode lock is held here, it is
* nonetheless safe to sleep waiting for a free page. The
* pageout daemon does not need to acquire the tmpfs vnode
* lock to page out tobj's pages because tobj is a OBJT_SWAP
* type object.
*/
m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL);
if (m->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(obj, idx, NULL, NULL)) {
rv = vm_pager_get_pages(obj, &m, 1, 0);
m = vm_page_lookup(obj, idx);
if (m == NULL) {
printf(
"uiomove_object: vm_obj %p idx %jd null lookup rv %d\n",
obj, idx, rv);
VM_OBJECT_WUNLOCK(obj);
return (EIO);
}
if (rv != VM_PAGER_OK) {
printf(
"uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n",
obj, idx, m->valid, rv);
vm_page_lock(m);
vm_page_free(m);
vm_page_unlock(m);
VM_OBJECT_WUNLOCK(obj);
return (EIO);
}
} else
vm_page_zero_invalid(m, TRUE);
}
vm_page_xunbusy(m);
vm_page_lock(m);
vm_page_hold(m);
vm_page_unlock(m);
VM_OBJECT_WUNLOCK(obj);
error = uiomove_fromphys(&m, offset, tlen, uio);
if (uio->uio_rw == UIO_WRITE && error == 0) {
VM_OBJECT_WLOCK(obj);
vm_page_dirty(m);
VM_OBJECT_WUNLOCK(obj);
}
vm_page_lock(m);
vm_page_unhold(m);
if (m->queue == PQ_NONE) {
vm_page_deactivate(m);
} else {
/* Requeue to maintain LRU ordering. */
vm_page_requeue(m);
}
vm_page_unlock(m);
return (error);
}
static void
free_pagelist(BULKINFO_T *bi, int actual)
{
vm_page_t*pages;
unsigned int num_pages, i;
void *page_address;
PAGELIST_T *pagelist;
pagelist = bi->pagelist;
vchiq_log_trace(vchiq_arm_log_level,
"free_pagelist - %x, %d", (unsigned int)pagelist, actual);
num_pages =
(pagelist->length + pagelist->offset + PAGE_SIZE - 1) /
PAGE_SIZE;
pages = (vm_page_t*)(pagelist->addrs + num_pages);
/* Deal with any partial cache lines (fragments) */
if (pagelist->type >= PAGELIST_READ_WITH_FRAGMENTS) {
FRAGMENTS_T *fragments = g_fragments_base +
(pagelist->type - PAGELIST_READ_WITH_FRAGMENTS);
int head_bytes, tail_bytes;
head_bytes = (CACHE_LINE_SIZE - pagelist->offset) &
(CACHE_LINE_SIZE - 1);
tail_bytes = (pagelist->offset + actual) &
(CACHE_LINE_SIZE - 1);
if ((actual >= 0) && (head_bytes != 0)) {
if (head_bytes > actual)
head_bytes = actual;
memcpy((char *)bi->buf,
fragments->headbuf,
head_bytes);
}
if ((actual >= 0) && (head_bytes < actual) &&
(tail_bytes != 0)) {
memcpy((char *)bi->buf + actual - tail_bytes,
fragments->tailbuf, tail_bytes);
}
down(&g_free_fragments_mutex);
*(FRAGMENTS_T **) fragments = g_free_fragments;
g_free_fragments = fragments;
up(&g_free_fragments_mutex);
up(&g_free_fragments_sema);
}
for (i = 0; i < num_pages; i++) {
if (pagelist->type != PAGELIST_WRITE)
vm_page_dirty(pages[i]);
}
vm_page_unhold_pages(pages, num_pages);
bus_dmamap_unload(bi->pagelist_dma_tag, bi->pagelist_dma_map);
bus_dmamem_free(bi->pagelist_dma_tag, bi->pagelist, bi->pagelist_dma_map);
bus_dmamap_destroy(bi->pagelist_dma_tag, bi->pagelist_dma_map);
bus_dma_tag_destroy(bi->pagelist_dma_tag);
free(bi, M_VCPAGELIST);
}
