/*
* Same as above, but forces the page to be detached from the object
* and go into free pool.
*/
void
sf_ext_free_nocache(void *arg1, void *arg2)
{
struct sf_buf *sf = arg1;
struct sendfile_sync *sfs = arg2;
vm_page_t pg = sf_buf_page(sf);
sf_buf_free(sf);
vm_page_lock(pg);
if (vm_page_unwire(pg, PQ_NONE)) {
vm_object_t obj;
/* Try to free the page, but only if it is cheap to. */
if ((obj = pg->object) == NULL)
vm_page_free(pg);
else if (!vm_page_xbusied(pg) && VM_OBJECT_TRYWLOCK(obj)) {
vm_page_free(pg);
VM_OBJECT_WUNLOCK(obj);
} else
vm_page_deactivate(pg);
}
vm_page_unlock(pg);
if (sfs != NULL) {
mtx_lock(&sfs->mtx);
KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0"));
if (--sfs->count == 0)
cv_signal(&sfs->cv);
mtx_unlock(&sfs->mtx);
}
}
/*
* Detach mapped page and release resources back to the system. Called
* by mbuf(9) code when last reference to a page is freed.
*/
void
sf_ext_free(void *arg1, void *arg2)
{
struct sf_buf *sf = arg1;
struct sendfile_sync *sfs = arg2;
vm_page_t pg = sf_buf_page(sf);
sf_buf_free(sf);
vm_page_lock(pg);
/*
* Check for the object going away on us. This can
* happen since we don't hold a reference to it.
* If so, we're responsible for freeing the page.
*/
if (vm_page_unwire(pg, PQ_INACTIVE) && pg->object == NULL)
vm_page_free(pg);
vm_page_unlock(pg);
if (sfs != NULL) {
mtx_lock(&sfs->mtx);
KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0"));
if (--sfs->count == 0)
cv_signal(&sfs->cv);
mtx_unlock(&sfs->mtx);
}
}
/* --------------------------------------------------------------------- */
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, rv;
VM_OBJECT_LOCK(tobj);
m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
if (m->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
rv = vm_pager_get_pages(tobj, &m, 1, 0);
if (rv != VM_PAGER_OK) {
vm_page_lock(m);
vm_page_free(m);
vm_page_unlock(m);
VM_OBJECT_UNLOCK(tobj);
return (EIO);
}
} else
vm_page_zero_invalid(m, TRUE);
}
VM_OBJECT_UNLOCK(tobj);
error = uiomove_fromphys(&m, offset, tlen, uio);
VM_OBJECT_LOCK(tobj);
vm_page_lock(m);
vm_page_unwire(m, TRUE);
vm_page_unlock(m);
vm_page_wakeup(m);
VM_OBJECT_UNLOCK(tobj);
return (error);
}
static int rtR0MemObjFreeBSDPhysAllocHelper(vm_object_t pObject, u_long cPages,
vm_paddr_t VmPhysAddrHigh, u_long uAlignment,
bool fContiguous, bool fWire, int rcNoMem)
{
if (fContiguous)
{
if (rtR0MemObjFreeBSDContigPhysAllocHelper(pObject, 0, cPages, VmPhysAddrHigh,
uAlignment, fWire) != NULL)
return VINF_SUCCESS;
return rcNoMem;
}
for (vm_pindex_t iPage = 0; iPage < cPages; iPage++)
{
vm_page_t pPage = rtR0MemObjFreeBSDContigPhysAllocHelper(pObject, iPage, 1, VmPhysAddrHigh,
uAlignment, fWire);
if (!pPage)
{
/* Free all allocated pages */
VM_OBJECT_LOCK(pObject);
while (iPage-- > 0)
{
pPage = vm_page_lookup(pObject, iPage);
vm_page_lock_queues();
if (fWire)
vm_page_unwire(pPage, 0);
vm_page_free(pPage);
vm_page_unlock_queues();
}
VM_OBJECT_UNLOCK(pObject);
return rcNoMem;
}
}
return VINF_SUCCESS;
}
/*
* kmem_unback:
*
* Unmap and free the physical pages underlying the specified virtual
* address range.
*
* A physical page must exist within the specified object at each index
* that is being unmapped.
*/
static int
_kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
{
vm_page_t m, next;
vm_offset_t end, offset;
int domain;
KASSERT(object == kernel_object,
("kmem_unback: only supports kernel object."));
if (size == 0)
return (0);
pmap_remove(kernel_pmap, addr, addr + size);
offset = addr - VM_MIN_KERNEL_ADDRESS;
end = offset + size;
VM_OBJECT_WLOCK(object);
m = vm_page_lookup(object, atop(offset));
domain = vm_phys_domain(m);
for (; offset < end; offset += PAGE_SIZE, m = next) {
next = vm_page_next(m);
vm_page_unwire(m, PQ_NONE);
vm_page_free(m);
}
VM_OBJECT_WUNLOCK(object);
return (domain);
}
vm_page_t
shmem_read_mapping_page(vm_object_t object, vm_pindex_t pindex)
{
vm_page_t m;
int rv;
VM_OBJECT_LOCK_ASSERT_OWNED(object);
m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
if (m->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(object, pindex)) {
rv = vm_pager_get_page(object, &m, 1);
m = vm_page_lookup(object, pindex);
if (m == NULL)
return ERR_PTR(-ENOMEM);
if (rv != VM_PAGER_OK) {
vm_page_free(m);
return ERR_PTR(-ENOMEM);
}
} else {
pmap_zero_page(VM_PAGE_TO_PHYS(m));
m->valid = VM_PAGE_BITS_ALL;
m->dirty = 0;
}
}
vm_page_wire(m);
vm_page_wakeup(m);
return (m);
}
/*
* Release a page we've previously wired.
*/
static void
zbuf_page_free(vm_page_t pp)
{
vm_page_lock(pp);
vm_page_unwire(pp, 0);
if (pp->wire_count == 0 && pp->object == NULL)
vm_page_free(pp);
vm_page_unlock(pp);
}
void
uma_small_free(void *mem, int size, u_int8_t flags)
{
vm_page_t m;
m = PHYS_TO_VM_PAGE(IA64_RR_MASK((u_int64_t)mem));
m->wire_count--;
vm_page_free(m);
atomic_subtract_int(&cnt.v_wire_count, 1);
}
/*
* Release a page busied for a getpages operation. The page may have become
* wired (typically due to being used by the buffer cache) or otherwise been
* soft-busied and cannot be freed in that case. A held page can still be
* freed.
*/
void
vnode_pager_freepage(vm_page_t m)
{
if (m->busy || m->wire_count || (m->flags & PG_NEED_COMMIT)) {
vm_page_activate(m);
vm_page_wakeup(m);
} else {
vm_page_free(m);
}
}
void
uma_small_free(void *mem, int size, u_int8_t flags)
{
vm_page_t m;
m = PHYS_TO_VM_PAGE(TLB_DIRECT_TO_PHYS((vm_offset_t)mem));
m->wire_count--;
vm_page_free(m);
atomic_subtract_int(&cnt.v_wire_count, 1);
}
/*
* 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);
}
void
uma_small_free(void *mem, vm_size_t size, u_int8_t flags)
{
vm_page_t m;
vm_paddr_t pa;
pa = DMAP_TO_PHYS((vm_offset_t)mem);
m = PHYS_TO_VM_PAGE(pa);
m->wire_count--;
vm_page_free(m);
atomic_subtract_int(&vm_cnt.v_wire_count, 1);
}
static void
os_kmem_free(vm_page_t page) // IN
{
os_state *state = &global_state;
os_pmap *pmap = &state->pmap;
if ( !vm_page_lookup(state->vmobject, page->pindex) ) {
return;
}
os_pmap_putindex(pmap, page->pindex);
vm_page_free(page);
}
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);
}
void
uma_small_free(void *mem, vm_size_t size, u_int8_t flags)
{
vm_page_t m;
if (!hw_direct_map)
pmap_remove(kernel_pmap,(vm_offset_t)mem,
(vm_offset_t)mem + PAGE_SIZE);
m = PHYS_TO_VM_PAGE((vm_offset_t)mem);
m->wire_count--;
vm_page_free(m);
atomic_subtract_int(&vm_cnt.v_wire_count, 1);
atomic_subtract_int(&hw_uma_mdpages, 1);
}
int
readdir(ipc_port_t file, struct fs_directory_entry *de, off_t *offp)
{
struct ipc_request_message req;
struct ipc_response_message resp;
struct fs_directory_read_request fsreq;
struct fs_directory_read_response *fsresp;
int error;
memset(&req, 0, sizeof req);
memset(&resp, 0, sizeof resp);
memset(&fsreq, 0, sizeof fsreq);
fsreq.offset = *offp;
req.src = IPC_PORT_UNKNOWN;
req.dst = file;
req.msg = FS_DIRECTORY_MSG_READ;
req.param = 0;
req.data = &fsreq;
req.datalen = sizeof fsreq;
resp.data = true;
error = ipc_request(&req, &resp);
if (error != 0)
return (error);
if (resp.error != 0)
return (resp.error);
/* End of directory. */
if (resp.param == 0) {
de->name[0] = '\0';
*offp = 0;
return (0);
}
/* XXX assert cnt == 1? */
fsresp = resp.page;
strlcpy(de->name, fsresp->entry.name, sizeof de->name);
*offp = fsresp->next;
vm_page_free(resp.page);
return (0);
}
static int
netmap_dev_pager_fault(vm_object_t object, vm_ooffset_t offset,
int prot, vm_page_t *mres)
{
struct netmap_vm_handle_t *vmh = object->handle;
struct netmap_priv_d *priv = vmh->priv;
vm_paddr_t paddr;
vm_page_t page;
vm_memattr_t memattr;
vm_pindex_t pidx;
ND("object %p offset %jd prot %d mres %p",
object, (intmax_t)offset, prot, mres);
memattr = object->memattr;
pidx = OFF_TO_IDX(offset);
paddr = netmap_mem_ofstophys(priv->np_mref, offset);
if (paddr == 0)
return VM_PAGER_FAIL;
if (((*mres)->flags & PG_FICTITIOUS) != 0) {
/*
* If the passed in result page is a fake page, update it with
* the new physical address.
*/
page = *mres;
vm_page_updatefake(page, paddr, memattr);
} else {
/*
* Replace the passed in reqpage page with our own fake page and
* free up the all of the original pages.
*/
#ifndef VM_OBJECT_WUNLOCK /* FreeBSD < 10.x */
#define VM_OBJECT_WUNLOCK VM_OBJECT_UNLOCK
#define VM_OBJECT_WLOCK VM_OBJECT_LOCK
#endif /* VM_OBJECT_WUNLOCK */
VM_OBJECT_WUNLOCK(object);
page = vm_page_getfake(paddr, memattr);
VM_OBJECT_WLOCK(object);
vm_page_free(*mres);
*mres = page;
vm_page_insert(page, object, pidx);
}
page->valid = VM_PAGE_BITS_ALL;
return (VM_PAGER_OK);
}
/*
* vm_contig_pg_free:
*
* Remove pages previously allocated by vm_contig_pg_alloc, and
* assume all references to the pages have been removed, and that
* it is OK to add them back to the free list.
*
* Caller must ensure no races on the page range in question.
* No other requirements.
*/
void
vm_contig_pg_free(int start, u_long size)
{
vm_page_t pga = vm_page_array;
vm_page_t m;
int i;
size = round_page(size);
if (size == 0)
panic("vm_contig_pg_free: size must not be 0");
lwkt_gettoken(&vm_token);
for (i = start; i < (start + size / PAGE_SIZE); i++) {
m = &pga[i];
vm_page_busy(m);
vm_page_free(m);
}
lwkt_reltoken(&vm_token);
}
static int old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset,
int prot, vm_page_t *mres)
{
vm_paddr_t paddr;
vm_page_t page;
vm_offset_t pidx = OFF_TO_IDX(offset);
cdev_t dev;
page = *mres;
dev = object->handle;
paddr = pmap_phys_address(
dev_dmmap(dev, offset, prot, NULL));
KASSERT(paddr != -1,("dev_pager_getpage: map function returns error"));
KKASSERT(object->type == OBJT_DEVICE);
if (page->flags & PG_FICTITIOUS) {
/*
* If the passed in reqpage page is already a fake page,
* update it with the new physical address.
*/
page->phys_addr = paddr;
page->valid = VM_PAGE_BITS_ALL;
} else {
/*
* Replace the passed in reqpage page with our own fake page
* and free up all the original pages.
*/
page = dev_pager_getfake(paddr, object->memattr);
TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist,
page, pageq);
vm_object_hold(object);
vm_page_free(*mres);
if (vm_page_insert(page, object, pidx) == FALSE) {
panic("dev_pager_getpage: page (%p,%016jx) exists",
object, (uintmax_t)pidx);
}
vm_object_drop(object);
}
return (VM_PAGER_OK);
}
static void
unlock_and_deallocate(struct faultstate *fs)
{
vm_object_pip_wakeup(fs->object);
VM_OBJECT_WUNLOCK(fs->object);
if (fs->object != fs->first_object) {
VM_OBJECT_WLOCK(fs->first_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;
}
vm_object_deallocate(fs->first_object);
unlock_map(fs);
if (fs->vp != NULL) {
vput(fs->vp);
fs->vp = NULL;
}
}
static int
privcmd_pg_fault(vm_object_t object, vm_ooffset_t offset,
int prot, vm_page_t *mres)
{
struct privcmd_map *map = object->handle;
vm_pindex_t pidx;
vm_page_t page, oldm;
if (map->mapped != true)
return (VM_PAGER_FAIL);
pidx = OFF_TO_IDX(offset);
if (pidx >= map->size || BIT_ISSET(map->size, pidx, map->err))
return (VM_PAGER_FAIL);
page = PHYS_TO_VM_PAGE(map->phys_base_addr + offset);
if (page == NULL)
return (VM_PAGER_FAIL);
KASSERT((page->flags & PG_FICTITIOUS) != 0,
("not fictitious %p", page));
KASSERT(page->wire_count == 1, ("wire_count not 1 %p", page));
KASSERT(vm_page_busied(page) == 0, ("page %p is busy", page));
if (*mres != NULL) {
oldm = *mres;
vm_page_lock(oldm);
vm_page_free(oldm);
vm_page_unlock(oldm);
*mres = NULL;
}
vm_page_insert(page, object, pidx);
page->valid = VM_PAGE_BITS_ALL;
vm_page_xbusy(page);
*mres = page;
return (VM_PAGER_OK);
}
static void
socow_iodone(void *addr, void *args)
{
struct sf_buf *sf;
vm_page_t pp;
sf = args;
pp = sf_buf_page(sf);
sf_buf_free(sf);
/* remove COW mapping */
vm_page_lock(pp);
vm_page_cowclear(pp);
vm_page_unwire(pp, 0);
/*
* Check for the object going away on us. This can
* happen since we don't hold a reference to it.
* If so, we're responsible for freeing the page.
*/
if (pp->wire_count == 0 && pp->object == NULL)
vm_page_free(pp);
vm_page_unlock(pp);
socow_stats.iodone++;
}
void
uma_small_free(void *mem, int size, u_int8_t flags)
{
pd_entry_t *pd;
pt_entry_t *pt;
if (flags & UMA_SLAB_KMEM)
kmem_free(kmem_map, (vm_offset_t)mem, size);
else {
struct arm_small_page *sp;
if ((vm_offset_t)mem >= KERNBASE) {
mtx_lock(&smallalloc_mtx);
sp = TAILQ_FIRST(&free_pgdesc);
KASSERT(sp != NULL, ("No more free page descriptor ?"));
TAILQ_REMOVE(&free_pgdesc, sp, pg_list);
sp->addr = mem;
pmap_get_pde_pte(kernel_pmap, (vm_offset_t)mem, &pd,
&pt);
if ((*pd & pte_l1_s_cache_mask) ==
pte_l1_s_cache_mode_pt &&
pte_l1_s_cache_mode_pt != pte_l1_s_cache_mode)
TAILQ_INSERT_HEAD(&pages_wt, sp, pg_list);
else
TAILQ_INSERT_HEAD(&pages_normal, sp, pg_list);
mtx_unlock(&smallalloc_mtx);
} else {
vm_page_t m;
vm_paddr_t pa = vtophys((vm_offset_t)mem);
m = PHYS_TO_VM_PAGE(pa);
m->wire_count--;
vm_page_free(m);
atomic_subtract_int(&cnt.v_wire_count, 1);
}
}
}
static int
sg_pager_getpages(vm_object_t object, vm_page_t *m, int count, int reqpage)
{
struct sglist *sg;
vm_page_t m_paddr, page;
vm_pindex_t offset;
vm_paddr_t paddr;
vm_memattr_t memattr;
size_t space;
int i;
VM_OBJECT_ASSERT_WLOCKED(object);
sg = object->handle;
memattr = object->memattr;
VM_OBJECT_WUNLOCK(object);
offset = m[reqpage]->pindex;
/*
* Lookup the physical address of the requested page. An initial
* value of '1' instead of '0' is used so we can assert that the
* page is found since '0' can be a valid page-aligned physical
* address.
*/
space = 0;
paddr = 1;
for (i = 0; i < sg->sg_nseg; i++) {
if (space + sg->sg_segs[i].ss_len <= (offset * PAGE_SIZE)) {
space += sg->sg_segs[i].ss_len;
continue;
}
paddr = sg->sg_segs[i].ss_paddr + offset * PAGE_SIZE - space;
break;
}
KASSERT(paddr != 1, ("invalid SG page index"));
/* If "paddr" is a real page, perform a sanity check on "memattr". */
if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
pmap_page_get_memattr(m_paddr) != memattr) {
memattr = pmap_page_get_memattr(m_paddr);
printf(
"WARNING: A device driver has set \"memattr\" inconsistently.\n");
}
/* Return a fake page for the requested page. */
KASSERT(!(m[reqpage]->flags & PG_FICTITIOUS),
("backing page for SG is fake"));
/* Construct a new fake page. */
page = vm_page_getfake(paddr, memattr);
VM_OBJECT_WLOCK(object);
TAILQ_INSERT_TAIL(&object->un_pager.sgp.sgp_pglist, page, plinks.q);
/* Free the original pages and insert this fake page into the object. */
for (i = 0; i < count; i++) {
if (i == reqpage &&
vm_page_replace(page, object, offset) != m[i])
panic("sg_pager_getpages: invalid place replacement");
vm_page_lock(m[i]);
vm_page_free(m[i]);
vm_page_unlock(m[i]);
}
m[reqpage] = page;
page->valid = VM_PAGE_BITS_ALL;
return (VM_PAGER_OK);
}
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);
}
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;
}
int
exec_map_first_page(struct image_params *imgp)
{
int rv, i, after, initial_pagein;
vm_page_t ma[VM_INITIAL_PAGEIN];
vm_object_t object;
if (imgp->firstpage != NULL)
exec_unmap_first_page(imgp);
object = imgp->vp->v_object;
if (object == NULL)
return (EACCES);
VM_OBJECT_WLOCK(object);
#if VM_NRESERVLEVEL > 0
vm_object_color(object, 0);
#endif
ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
if (ma[0]->valid != VM_PAGE_BITS_ALL) {
vm_page_xbusy(ma[0]);
if (!vm_pager_has_page(object, 0, NULL, &after)) {
vm_page_lock(ma[0]);
vm_page_free(ma[0]);
vm_page_unlock(ma[0]);
VM_OBJECT_WUNLOCK(object);
return (EIO);
}
initial_pagein = min(after, VM_INITIAL_PAGEIN);
KASSERT(initial_pagein <= object->size,
("%s: initial_pagein %d object->size %ju",
__func__, initial_pagein, (uintmax_t )object->size));
for (i = 1; i < initial_pagein; i++) {
if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
if (ma[i]->valid)
break;
if (!vm_page_tryxbusy(ma[i]))
break;
} else {
ma[i] = vm_page_alloc(object, i,
VM_ALLOC_NORMAL);
if (ma[i] == NULL)
break;
}
}
initial_pagein = i;
rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
if (rv != VM_PAGER_OK) {
for (i = 0; i < initial_pagein; i++) {
vm_page_lock(ma[i]);
vm_page_free(ma[i]);
vm_page_unlock(ma[i]);
}
VM_OBJECT_WUNLOCK(object);
return (EIO);
}
vm_page_xunbusy(ma[0]);
for (i = 1; i < initial_pagein; i++)
vm_page_readahead_finish(ma[i]);
}
vm_page_lock(ma[0]);
vm_page_hold(ma[0]);
vm_page_activate(ma[0]);
vm_page_unlock(ma[0]);
VM_OBJECT_WUNLOCK(object);
imgp->firstpage = sf_buf_alloc(ma[0], 0);
imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
return (0);
}
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);
}
/* See: old_dev_pager_fault() in device_pager.c as an example. */
static int
cheri_compositor_cfb_pg_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
vm_page_t *mres)
{
vm_pindex_t pidx;
vm_paddr_t paddr;
vm_page_t page;
struct cfb_vm_object *cfb_vm_obj;
struct cdev *dev;
struct cheri_compositor_softc *sc;
struct cdevsw *csw;
vm_memattr_t memattr;
int ref;
int retval;
pidx = OFF_TO_IDX(offset);
VM_OBJECT_WUNLOCK(vm_obj);
cfb_vm_obj = vm_obj->handle;
dev = cfb_vm_obj->dev;
sc = dev->si_drv1;
retval = VM_PAGER_OK;
CHERI_COMPOSITOR_DEBUG(sc, "vm_obj: %p, offset: %lu, prot: %i", vm_obj,
offset, prot);
csw = dev_refthread(dev, &ref);
if (csw == NULL) {
retval = VM_PAGER_FAIL;
goto done_unlocked;
}
/* Traditional d_mmap() call. */
CHERI_COMPOSITOR_DEBUG(sc, "offset: %lu, nprot: %i", offset, prot);
if (validate_prot_and_offset(sc, cfb_vm_obj->pool->mapped_fd,
prot, offset) != 0) {
retval = VM_PAGER_FAIL;
goto done_unlocked;
}
paddr = calculate_physical_address(sc, cfb_vm_obj->pool, offset);
memattr = VM_MEMATTR_UNCACHEABLE;
CHERI_COMPOSITOR_DEBUG(sc, "paddr: %p, memattr: %i",
(void *) paddr, memattr);
dev_relthread(dev, ref);
/* Sanity checks. */
KASSERT((((*mres)->flags & PG_FICTITIOUS) == 0),
("Expected non-fictitious page."));
/*
* Replace the passed in reqpage page with our own fake page and
* free up the all of the original pages.
*/
page = vm_page_getfake(paddr, memattr);
VM_OBJECT_WLOCK(vm_obj);
vm_page_lock(*mres);
vm_page_free(*mres);
vm_page_unlock(*mres);
*mres = page;
vm_page_insert(page, vm_obj, pidx);
page->valid = VM_PAGE_BITS_ALL;
/* Success! */
retval = VM_PAGER_OK;
goto done;
done_unlocked:
VM_OBJECT_WLOCK(vm_obj);
done:
CHERI_COMPOSITOR_DEBUG(sc, "Finished with mres: %p (retval: %i)", *mres,
retval);
return (retval);
}
int
socow_setup(struct mbuf *m0, struct uio *uio)
{
struct sf_buf *sf;
vm_page_t pp;
struct iovec *iov;
struct vmspace *vmspace;
struct vm_map *map;
vm_offset_t offset, uva;
socow_stats.attempted++;
vmspace = curproc->p_vmspace;
map = &vmspace->vm_map;
uva = (vm_offset_t) uio->uio_iov->iov_base;
offset = uva & PAGE_MASK;
/*
* Verify that access to the given address is allowed from user-space.
*/
if (vm_fault_quick((caddr_t)uva, VM_PROT_READ) < 0)
return (0);
/*
* verify page is mapped & not already wired for i/o
*/
pp = pmap_extract_and_hold(map->pmap, uva, VM_PROT_READ);
if (pp == NULL) {
socow_stats.fail_not_mapped++;
return(0);
}
/*
* set up COW
*/
vm_page_lock(pp);
if (vm_page_cowsetup(pp) != 0) {
vm_page_unhold(pp);
vm_page_unlock(pp);
return (0);
}
/*
* wire the page for I/O
*/
vm_page_wire(pp);
vm_page_unhold(pp);
vm_page_unlock(pp);
/*
* Allocate an sf buf
*/
sf = sf_buf_alloc(pp, SFB_CATCH);
if (sf == NULL) {
vm_page_lock(pp);
vm_page_cowclear(pp);
vm_page_unwire(pp, 0);
/*
* Check for the object going away on us. This can
* happen since we don't hold a reference to it.
* If so, we're responsible for freeing the page.
*/
if (pp->wire_count == 0 && pp->object == NULL)
vm_page_free(pp);
vm_page_unlock(pp);
socow_stats.fail_sf_buf++;
return(0);
}
/*
* attach to mbuf
*/
MEXTADD(m0, sf_buf_kva(sf), PAGE_SIZE, socow_iodone,
(void*)sf_buf_kva(sf), sf, M_RDONLY, EXT_SFBUF);
m0->m_len = PAGE_SIZE - offset;
m0->m_data = (caddr_t)sf_buf_kva(sf) + offset;
socow_stats.success++;
iov = uio->uio_iov;
iov->iov_base = (char *)iov->iov_base + m0->m_len;
iov->iov_len -= m0->m_len;
uio->uio_resid -= m0->m_len;
uio->uio_offset += m0->m_len;
if (iov->iov_len == 0) {
uio->uio_iov++;
uio->uio_iovcnt--;
}
return(m0->m_len);
}
