/* --------------------------------------------------------------------- */
static int
tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
vm_offset_t offset, size_t tlen, struct uio *uio)
{
vm_page_t m;
int error;
VM_OBJECT_LOCK(tobj);
vm_object_pip_add(tobj, 1);
m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
if (m->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
error = vm_pager_get_pages(tobj, &m, 1, 0);
if (error != 0) {
printf("tmpfs get pages from pager error [read]\n");
goto out;
}
} else
vm_page_zero_invalid(m, TRUE);
}
VM_OBJECT_UNLOCK(tobj);
error = uiomove_fromphys(&m, offset, tlen, uio);
VM_OBJECT_LOCK(tobj);
out:
vm_page_lock(m);
vm_page_unwire(m, TRUE);
vm_page_unlock(m);
vm_page_wakeup(m);
vm_object_pip_subtract(tobj, 1);
VM_OBJECT_UNLOCK(tobj);
return (error);
}
int ttm_tt_swapin(struct ttm_tt *ttm)
{
vm_object_t obj;
vm_page_t from_page, to_page;
int i, ret, rv;
obj = ttm->swap_storage;
VM_OBJECT_LOCK(obj);
vm_object_pip_add(obj, 1);
for (i = 0; i < ttm->num_pages; ++i) {
from_page = vm_page_grab(obj, i, VM_ALLOC_NORMAL |
VM_ALLOC_RETRY);
if (from_page->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(obj, i)) {
rv = vm_pager_get_page(obj, &from_page, 1);
if (rv != VM_PAGER_OK) {
vm_page_free(from_page);
ret = -EIO;
goto err_ret;
}
} else {
vm_page_zero_invalid(from_page, TRUE);
}
}
to_page = ttm->pages[i];
if (unlikely(to_page == NULL)) {
ret = -ENOMEM;
vm_page_wakeup(from_page);
goto err_ret;
}
pmap_copy_page(VM_PAGE_TO_PHYS(from_page),
VM_PAGE_TO_PHYS(to_page));
vm_page_wakeup(from_page);
}
vm_object_pip_wakeup(obj);
VM_OBJECT_UNLOCK(obj);
if (!(ttm->page_flags & TTM_PAGE_FLAG_PERSISTENT_SWAP))
vm_object_deallocate(obj);
ttm->swap_storage = NULL;
ttm->page_flags &= ~TTM_PAGE_FLAG_SWAPPED;
return (0);
err_ret:
vm_object_pip_wakeup(obj);
VM_OBJECT_UNLOCK(obj);
return (ret);
}
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);
}
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);
}
/*
* If blocks are contiguous on disk, use this to provide clustered
* read ahead. We will read as many blocks as possible sequentially
* and then parcel them up into logical blocks in the buffer hash table.
*/
static struct buf *
cluster_rbuild(struct vnode *vp, u_quad_t filesize, daddr_t lbn,
daddr_t blkno, long size, int run, int gbflags, struct buf *fbp)
{
struct buf *bp, *tbp;
daddr_t bn;
off_t off;
long tinc, tsize;
int i, inc, j, k, toff;
KASSERT(size == vp->v_mount->mnt_stat.f_iosize,
("cluster_rbuild: size %ld != f_iosize %jd\n",
size, (intmax_t)vp->v_mount->mnt_stat.f_iosize));
/*
* avoid a division
*/
while ((u_quad_t) size * (lbn + run) > filesize) {
--run;
}
if (fbp) {
tbp = fbp;
tbp->b_iocmd = BIO_READ;
} else {
tbp = getblk(vp, lbn, size, 0, 0, gbflags);
if (tbp->b_flags & B_CACHE)
return tbp;
tbp->b_flags |= B_ASYNC | B_RAM;
tbp->b_iocmd = BIO_READ;
}
tbp->b_blkno = blkno;
if( (tbp->b_flags & B_MALLOC) ||
((tbp->b_flags & B_VMIO) == 0) || (run <= 1) )
return tbp;
bp = trypbuf(&cluster_pbuf_freecnt);
if (bp == NULL)
return tbp;
/*
* We are synthesizing a buffer out of vm_page_t's, but
* if the block size is not page aligned then the starting
* address may not be either. Inherit the b_data offset
* from the original buffer.
*/
bp->b_flags = B_ASYNC | B_CLUSTER | B_VMIO;
if ((gbflags & GB_UNMAPPED) != 0) {
bp->b_data = unmapped_buf;
} else {
bp->b_data = (char *)((vm_offset_t)bp->b_data |
((vm_offset_t)tbp->b_data & PAGE_MASK));
}
bp->b_iocmd = BIO_READ;
bp->b_iodone = cluster_callback;
bp->b_blkno = blkno;
bp->b_lblkno = lbn;
bp->b_offset = tbp->b_offset;
KASSERT(bp->b_offset != NOOFFSET, ("cluster_rbuild: no buffer offset"));
pbgetvp(vp, bp);
TAILQ_INIT(&bp->b_cluster.cluster_head);
bp->b_bcount = 0;
bp->b_bufsize = 0;
bp->b_npages = 0;
inc = btodb(size);
for (bn = blkno, i = 0; i < run; ++i, bn += inc) {
if (i == 0) {
VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object);
vfs_drain_busy_pages(tbp);
vm_object_pip_add(tbp->b_bufobj->bo_object,
tbp->b_npages);
for (k = 0; k < tbp->b_npages; k++)
vm_page_sbusy(tbp->b_pages[k]);
VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object);
} else {
if ((bp->b_npages * PAGE_SIZE) +
round_page(size) > vp->v_mount->mnt_iosize_max) {
break;
}
tbp = getblk(vp, lbn + i, size, 0, 0, GB_LOCK_NOWAIT |
(gbflags & GB_UNMAPPED));
/* Don't wait around for locked bufs. */
if (tbp == NULL)
break;
/*
* Stop scanning if the buffer is fully valid
* (marked B_CACHE), or locked (may be doing a
* background write), or if the buffer is not
* VMIO backed. The clustering code can only deal
* with VMIO-backed buffers. The bo lock is not
* required for the BKGRDINPROG check since it
* can not be set without the buf lock.
*/
if ((tbp->b_vflags & BV_BKGRDINPROG) ||
(tbp->b_flags & B_CACHE) ||
(tbp->b_flags & B_VMIO) == 0) {
bqrelse(tbp);
break;
}
/*
* The buffer must be completely invalid in order to
* take part in the cluster. If it is partially valid
* then we stop.
*/
off = tbp->b_offset;
tsize = size;
VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object);
for (j = 0; tsize > 0; j++) {
toff = off & PAGE_MASK;
tinc = tsize;
if (toff + tinc > PAGE_SIZE)
tinc = PAGE_SIZE - toff;
VM_OBJECT_ASSERT_WLOCKED(tbp->b_pages[j]->object);
if ((tbp->b_pages[j]->valid &
vm_page_bits(toff, tinc)) != 0)
break;
if (vm_page_xbusied(tbp->b_pages[j]))
break;
vm_object_pip_add(tbp->b_bufobj->bo_object, 1);
vm_page_sbusy(tbp->b_pages[j]);
off += tinc;
tsize -= tinc;
}
if (tsize > 0) {
clean_sbusy:
vm_object_pip_add(tbp->b_bufobj->bo_object, -j);
for (k = 0; k < j; k++)
vm_page_sunbusy(tbp->b_pages[k]);
VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object);
bqrelse(tbp);
break;
}
VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object);
/*
* Set a read-ahead mark as appropriate
*/
if ((fbp && (i == 1)) || (i == (run - 1)))
tbp->b_flags |= B_RAM;
/*
* Set the buffer up for an async read (XXX should
* we do this only if we do not wind up brelse()ing?).
* Set the block number if it isn't set, otherwise
* if it is make sure it matches the block number we
* expect.
*/
tbp->b_flags |= B_ASYNC;
tbp->b_iocmd = BIO_READ;
if (tbp->b_blkno == tbp->b_lblkno) {
tbp->b_blkno = bn;
} else if (tbp->b_blkno != bn) {
VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object);
goto clean_sbusy;
}
}
/*
* XXX fbp from caller may not be B_ASYNC, but we are going
* to biodone() it in cluster_callback() anyway
*/
BUF_KERNPROC(tbp);
TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head,
tbp, b_cluster.cluster_entry);
VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object);
for (j = 0; j < tbp->b_npages; j += 1) {
vm_page_t m;
m = tbp->b_pages[j];
if ((bp->b_npages == 0) ||
(bp->b_pages[bp->b_npages-1] != m)) {
bp->b_pages[bp->b_npages] = m;
bp->b_npages++;
}
if (m->valid == VM_PAGE_BITS_ALL)
tbp->b_pages[j] = bogus_page;
}
VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object);
/*
* Don't inherit tbp->b_bufsize as it may be larger due to
* a non-page-aligned size. Instead just aggregate using
* 'size'.
*/
if (tbp->b_bcount != size)
printf("warning: tbp->b_bcount wrong %ld vs %ld\n", tbp->b_bcount, size);
if (tbp->b_bufsize != size)
printf("warning: tbp->b_bufsize wrong %ld vs %ld\n", tbp->b_bufsize, size);
bp->b_bcount += size;
bp->b_bufsize += size;
}
/*
* Fully valid pages in the cluster are already good and do not need
* to be re-read from disk. Replace the page with bogus_page
*/
VM_OBJECT_WLOCK(bp->b_bufobj->bo_object);
for (j = 0; j < bp->b_npages; j++) {
VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[j]->object);
if (bp->b_pages[j]->valid == VM_PAGE_BITS_ALL)
bp->b_pages[j] = bogus_page;
}
VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object);
if (bp->b_bufsize > bp->b_kvasize)
panic("cluster_rbuild: b_bufsize(%ld) > b_kvasize(%d)\n",
bp->b_bufsize, bp->b_kvasize);
if (buf_mapped(bp)) {
pmap_qenter(trunc_page((vm_offset_t) bp->b_data),
(vm_page_t *)bp->b_pages, bp->b_npages);
}
return (bp);
}
static int
ttm_bo_vm_fault(vm_object_t vm_obj, vm_ooffset_t offset,
int prot, vm_page_t *mres)
{
struct ttm_buffer_object *bo = vm_obj->handle;
struct ttm_bo_device *bdev = bo->bdev;
struct ttm_tt *ttm = NULL;
vm_page_t m, m1, oldm;
int ret;
int retval = VM_PAGER_OK;
struct ttm_mem_type_manager *man =
&bdev->man[bo->mem.mem_type];
vm_object_pip_add(vm_obj, 1);
oldm = *mres;
if (oldm != NULL) {
vm_page_lock(oldm);
vm_page_remove(oldm);
vm_page_unlock(oldm);
*mres = NULL;
} else
oldm = NULL;
retry:
VM_OBJECT_WUNLOCK(vm_obj);
m = NULL;
reserve:
ret = ttm_bo_reserve(bo, false, false, false, 0);
if (unlikely(ret != 0)) {
if (ret == -EBUSY) {
kern_yield(0);
goto reserve;
}
}
if (bdev->driver->fault_reserve_notify) {
ret = bdev->driver->fault_reserve_notify(bo);
switch (ret) {
case 0:
break;
case -EBUSY:
case -ERESTART:
case -EINTR:
kern_yield(0);
goto reserve;
default:
retval = VM_PAGER_ERROR;
goto out_unlock;
}
}
/*
* Wait for buffer data in transit, due to a pipelined
* move.
*/
mtx_lock(&bdev->fence_lock);
if (test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)) {
/*
* Here, the behavior differs between Linux and FreeBSD.
*
* On Linux, the wait is interruptible (3rd argument to
* ttm_bo_wait). There must be some mechanism to resume
* page fault handling, once the signal is processed.
*
* On FreeBSD, the wait is uninteruptible. This is not a
* problem as we can't end up with an unkillable process
* here, because the wait will eventually time out.
*
* An example of this situation is the Xorg process
* which uses SIGALRM internally. The signal could
* interrupt the wait, causing the page fault to fail
* and the process to receive SIGSEGV.
*/
ret = ttm_bo_wait(bo, false, false, false);
mtx_unlock(&bdev->fence_lock);
if (unlikely(ret != 0)) {
retval = VM_PAGER_ERROR;
goto out_unlock;
}
} else
mtx_unlock(&bdev->fence_lock);
ret = ttm_mem_io_lock(man, true);
if (unlikely(ret != 0)) {
retval = VM_PAGER_ERROR;
goto out_unlock;
}
ret = ttm_mem_io_reserve_vm(bo);
if (unlikely(ret != 0)) {
retval = VM_PAGER_ERROR;
goto out_io_unlock;
}
/*
* Strictly, we're not allowed to modify vma->vm_page_prot here,
* since the mmap_sem is only held in read mode. However, we
* modify only the caching bits of vma->vm_page_prot and
* consider those bits protected by
* the bo->mutex, as we should be the only writers.
* There shouldn't really be any readers of these bits except
* within vm_insert_mixed()? fork?
*
* TODO: Add a list of vmas to the bo, and change the
* vma->vm_page_prot when the object changes caching policy, with
* the correct locks held.
*/
if (!bo->mem.bus.is_iomem) {
/* Allocate all page at once, most common usage */
ttm = bo->ttm;
if (ttm->bdev->driver->ttm_tt_populate(ttm)) {
retval = VM_PAGER_ERROR;
goto out_io_unlock;
}
}
if (bo->mem.bus.is_iomem) {
m = PHYS_TO_VM_PAGE(bo->mem.bus.base + bo->mem.bus.offset +
offset);
KASSERT((m->flags & PG_FICTITIOUS) != 0,
("physical address %#jx not fictitious",
(uintmax_t)(bo->mem.bus.base + bo->mem.bus.offset
+ offset)));
pmap_page_set_memattr(m, ttm_io_prot(bo->mem.placement));
} else {
ttm = bo->ttm;
m = ttm->pages[OFF_TO_IDX(offset)];
if (unlikely(!m)) {
retval = VM_PAGER_ERROR;
goto out_io_unlock;
}
pmap_page_set_memattr(m,
(bo->mem.placement & TTM_PL_FLAG_CACHED) ?
VM_MEMATTR_WRITE_BACK : ttm_io_prot(bo->mem.placement));
}
VM_OBJECT_WLOCK(vm_obj);
if (vm_page_busied(m)) {
vm_page_lock(m);
VM_OBJECT_WUNLOCK(vm_obj);
vm_page_busy_sleep(m, "ttmpbs");
VM_OBJECT_WLOCK(vm_obj);
ttm_mem_io_unlock(man);
ttm_bo_unreserve(bo);
goto retry;
}
m1 = vm_page_lookup(vm_obj, OFF_TO_IDX(offset));
if (m1 == NULL) {
if (vm_page_insert(m, vm_obj, OFF_TO_IDX(offset))) {
VM_OBJECT_WUNLOCK(vm_obj);
VM_WAIT;
VM_OBJECT_WLOCK(vm_obj);
ttm_mem_io_unlock(man);
ttm_bo_unreserve(bo);
goto retry;
}
} else {
KASSERT(m == m1,
("inconsistent insert bo %p m %p m1 %p offset %jx",
bo, m, m1, (uintmax_t)offset));
}
m->valid = VM_PAGE_BITS_ALL;
*mres = m;
vm_page_xbusy(m);
if (oldm != NULL) {
vm_page_lock(oldm);
vm_page_free(oldm);
vm_page_unlock(oldm);
}
out_io_unlock1:
ttm_mem_io_unlock(man);
out_unlock1:
ttm_bo_unreserve(bo);
vm_object_pip_wakeup(vm_obj);
return (retval);
out_io_unlock:
VM_OBJECT_WLOCK(vm_obj);
goto out_io_unlock1;
out_unlock:
VM_OBJECT_WLOCK(vm_obj);
goto out_unlock1;
}
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;
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);
vm_object_pip_add(tobj, 1);
tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
if (tpg->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
error = vm_pager_get_pages(tobj, &tpg, 1, 0);
if (error != 0) {
printf("tmpfs get pages from pager error [write]\n");
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);
out:
if (vobj != NULL)
VM_OBJECT_LOCK(vobj);
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);
if (vpg != NULL)
vm_page_wakeup(vpg);
if (vobj != NULL)
VM_OBJECT_UNLOCK(vobj);
vm_object_pip_subtract(tobj, 1);
VM_OBJECT_UNLOCK(tobj);
return (error);
}
static int
pscnv_gem_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
vm_page_t *mres)
{
struct drm_gem_object *gem_obj = vm_obj->handle;
struct pscnv_bo *bo = gem_obj->driver_private;
struct drm_device *dev = gem_obj->dev;
struct drm_nouveau_private *dev_priv = dev->dev_private;
vm_page_t m = NULL;
vm_page_t oldm;
vm_memattr_t mattr;
vm_paddr_t paddr;
const char *what;
if (bo->chan) {
paddr = dev_priv->fb_phys + offset +
nvc0_fifo_ctrl_offs(dev, bo->chan->cid);
mattr = VM_MEMATTR_UNCACHEABLE;
what = "fifo";
} else switch (bo->flags & PSCNV_GEM_MEMTYPE_MASK) {
case PSCNV_GEM_VRAM_SMALL:
case PSCNV_GEM_VRAM_LARGE:
paddr = dev_priv->fb_phys + bo->map1->start + offset;
mattr = VM_MEMATTR_WRITE_COMBINING;
what = "vram";
break;
case PSCNV_GEM_SYSRAM_SNOOP:
case PSCNV_GEM_SYSRAM_NOSNOOP:
paddr = bo->dmapages[OFF_TO_IDX(offset)];
mattr = VM_MEMATTR_WRITE_BACK;
what = "sysram";
break;
default: return (EINVAL);
}
if (offset >= bo->size) {
if (pscnv_mem_debug > 0)
NV_WARN(dev, "Reading %p + %08llx (%s) is past max size %08llx\n",
bo, offset, what, bo->size);
return (VM_PAGER_ERROR);
}
DRM_LOCK(dev);
if (pscnv_mem_debug > 0)
NV_WARN(dev, "Connecting %p+%08llx (%s) at phys %010llx\n",
bo, offset, what, paddr);
vm_object_pip_add(vm_obj, 1);
if (*mres != NULL) {
oldm = *mres;
vm_page_lock(oldm);
vm_page_remove(oldm);
vm_page_unlock(oldm);
*mres = NULL;
} else
oldm = NULL;
//VM_OBJECT_LOCK(vm_obj);
m = vm_phys_fictitious_to_vm_page(paddr);
if (m == NULL) {
DRM_UNLOCK(dev);
return -EFAULT;
}
KASSERT((m->flags & PG_FICTITIOUS) != 0,
("not fictitious %p", m));
KASSERT(m->wire_count == 1, ("wire_count not 1 %p", m));
if ((m->flags & VPO_BUSY) != 0) {
DRM_UNLOCK(dev);
return -EFAULT;
}
pmap_page_set_memattr(m, mattr);
m->valid = VM_PAGE_BITS_ALL;
*mres = m;
vm_page_lock(m);
vm_page_insert(m, vm_obj, OFF_TO_IDX(offset));
vm_page_unlock(m);
vm_page_busy(m);
printf("fault %p %jx %x phys %x", gem_obj, offset, prot,
m->phys_addr);
DRM_UNLOCK(dev);
if (oldm != NULL) {
vm_page_lock(oldm);
vm_page_free(oldm);
vm_page_unlock(oldm);
}
vm_object_pip_wakeup(vm_obj);
return (VM_PAGER_OK);
}