Redo the page table page allocation on MIPS, as suggested by

alc@.

The UMA zone based allocation is replaced by a scheme that creates
a new free page list for the KSEG0 region, and a new function
in sys/vm that allocates pages from a specific free page list.

This also fixes a race condition introduced by the UMA based page table
page allocation code. Dropping the page queue and pmap locks before
the call to uma_zfree, and re-acquiring them afterwards  will introduce
a race condtion(noted by alc@).

The changes are :
- Revert the earlier changes in MIPS pmap.c that added UMA zone for
page table pages.
- Add a new freelist VM_FREELIST_HIGHMEM to MIPS vmparam.h for memory that
is not directly mapped (in 32bit kernel). Normal page allocations will first
try the HIGHMEM freelist and then the default(direct mapped) freelist.
- Add a new function 'vm_page_t vm_page_alloc_freelist(int flind, int
order, int req)' to vm/vm_page.c to allocate a page from a specified
freelist. The MIPS page table pages will be allocated using this function
from the freelist containing direct mapped pages.
- Move the page initialization code from vm_phys_alloc_contig() to a
new function vm_page_alloc_init(), and use this function to initialize
pages in vm_page_alloc_freelist() too.
- Split the  function vm_phys_alloc_pages(int pool, int order) to create
vm_phys_alloc_freelist_pages(int flind, int pool, int order), and use
this function from both vm_page_alloc_freelist() and vm_phys_alloc_pages().

Reviewed by:	alc

4 files changed, 154 insertions, 74 deletions

diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c
index 348ab99..44b801e 100644
--- a/sys/vm/vm_page.c
+++ b/sys/vm/vm_page.c
@@ -1355,6 +1355,95 @@ vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
 }
 
 /*
+ * Initialize a page that has been freshly dequeued from a freelist.
+ * The caller has to drop the vnode returned, if it is not NULL.
+ *
+ * To be called with vm_page_queue_free_mtx held.
+ */
+struct vnode *
+vm_page_alloc_init(vm_page_t m)
+{
+	struct vnode *drop;
+	vm_object_t m_object;
+
+	KASSERT(m->queue == PQ_NONE,
+	    ("vm_page_alloc_init: page %p has unexpected queue %d",
+	    m, m->queue));
+	KASSERT(m->wire_count == 0,
+	    ("vm_page_alloc_init: page %p is wired", m));
+	KASSERT(m->hold_count == 0,
+	    ("vm_page_alloc_init: page %p is held", m));
+	KASSERT(m->busy == 0,
+	    ("vm_page_alloc_init: page %p is busy", m));
+	KASSERT(m->dirty == 0,
+	    ("vm_page_alloc_init: page %p is dirty", m));
+	KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
+	    ("vm_page_alloc_init: page %p has unexpected memattr %d",
+	    m, pmap_page_get_memattr(m)));
+	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
+	drop = NULL;
+	if ((m->flags & PG_CACHED) != 0) {
+		m->valid = 0;
+		m_object = m->object;
+		vm_page_cache_remove(m);
+		if (m_object->type == OBJT_VNODE &&
+		    m_object->cache == NULL)
+			drop = m_object->handle;
+	} else {
+		KASSERT(VM_PAGE_IS_FREE(m),
+		    ("vm_page_alloc_init: page %p is not free", m));
+		KASSERT(m->valid == 0,
+		    ("vm_page_alloc_init: free page %p is valid", m));
+		cnt.v_free_count--;
+	}
+	if (m->flags & PG_ZERO)
+		vm_page_zero_count--;
+	/* Don't clear the PG_ZERO flag; we'll need it later. */
+	m->flags = PG_UNMANAGED | (m->flags & PG_ZERO);
+	m->oflags = 0;
+	/* Unmanaged pages don't use "act_count". */
+	return (drop);
+}
+
+/*
+ * 	vm_page_alloc_freelist:
+ * 
+ *	Allocate a page from the specified freelist with specified order.
+ *	Only the ALLOC_CLASS values in req are honored, other request flags
+ *	are ignored.
+ */
+vm_page_t
+vm_page_alloc_freelist(int flind, int order, int req)
+{
+	struct vnode *drop;
+	vm_page_t m;
+	int page_req;
+
+	m = NULL;
+	page_req = req & VM_ALLOC_CLASS_MASK;
+	mtx_lock(&vm_page_queue_free_mtx);
+	/*
+	 * Do not allocate reserved pages unless the req has asked for it.
+	 */
+	if (cnt.v_free_count + cnt.v_cache_count > cnt.v_free_reserved ||
+	    (page_req == VM_ALLOC_SYSTEM && 
+	    cnt.v_free_count + cnt.v_cache_count > cnt.v_interrupt_free_min) ||
+	    (page_req == VM_ALLOC_INTERRUPT &&
+	    cnt.v_free_count + cnt.v_cache_count > 0)) {
+		m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, order);
+	}
+	if (m == NULL) {
+		mtx_unlock(&vm_page_queue_free_mtx);
+		return (NULL);
+	}
+	drop = vm_page_alloc_init(m);
+	mtx_unlock(&vm_page_queue_free_mtx);
+	if (drop)
+		vdrop(drop);
+	return (m);
+}
+
+/*
  *	vm_wait:	(also see VM_WAIT macro)
  *
  *	Block until free pages are available for allocation
diff --git a/sys/vm/vm_page.h b/sys/vm/vm_page.h
index a877c43..86119a1 100644
--- a/sys/vm/vm_page.h
+++ b/sys/vm/vm_page.h
@@ -262,6 +262,7 @@ extern struct vpglocks pa_lock[];
  *
  */
 
+struct vnode;
 extern int vm_page_zero_count;
 
 extern vm_page_t vm_page_array;		/* First resident page in table */
@@ -339,6 +340,8 @@ void vm_pageq_remove(vm_page_t m);
 
 void vm_page_activate (vm_page_t);
 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
+vm_page_t vm_page_alloc_freelist(int, int, int);
+struct vnode *vm_page_alloc_init(vm_page_t);
 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
 void vm_page_cache(vm_page_t);
 void vm_page_cache_free(vm_object_t, vm_pindex_t, vm_pindex_t);
diff --git a/sys/vm/vm_phys.c b/sys/vm/vm_phys.c
index a245462..3482337 100644
--- a/sys/vm/vm_phys.c
+++ b/sys/vm/vm_phys.c
@@ -301,49 +301,67 @@ vm_phys_add_page(vm_paddr_t pa)
 vm_page_t
 vm_phys_alloc_pages(int pool, int order)
 {
+	vm_page_t m;
+	int flind;
+
+	for (flind = 0; flind < vm_nfreelists; flind++) {
+		m = vm_phys_alloc_freelist_pages(flind, pool, order);
+		if (m != NULL)
+			return (m);
+	}
+	return (NULL);
+}
+
+/*
+ * Find and dequeue a free page on the given free list, with the 
+ * specified pool and order
+ */
+vm_page_t
+vm_phys_alloc_freelist_pages(int flind, int pool, int order)
+{	
 	struct vm_freelist *fl;
 	struct vm_freelist *alt;
-	int flind, oind, pind;
+	int oind, pind;
 	vm_page_t m;
 
+	KASSERT(flind < VM_NFREELIST,
+	    ("vm_phys_alloc_freelist_pages: freelist %d is out of range", flind));
 	KASSERT(pool < VM_NFREEPOOL,
-	    ("vm_phys_alloc_pages: pool %d is out of range", pool));
+	    ("vm_phys_alloc_freelist_pages: pool %d is out of range", pool));
 	KASSERT(order < VM_NFREEORDER,
-	    ("vm_phys_alloc_pages: order %d is out of range", order));
+	    ("vm_phys_alloc_freelist_pages: order %d is out of range", order));
 	mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
-	for (flind = 0; flind < vm_nfreelists; flind++) {
-		fl = vm_phys_free_queues[flind][pool];
-		for (oind = order; oind < VM_NFREEORDER; oind++) {
-			m = TAILQ_FIRST(&fl[oind].pl);
+	fl = vm_phys_free_queues[flind][pool];
+	for (oind = order; oind < VM_NFREEORDER; oind++) {
+		m = TAILQ_FIRST(&fl[oind].pl);
+		if (m != NULL) {
+			TAILQ_REMOVE(&fl[oind].pl, m, pageq);
+			fl[oind].lcnt--;
+			m->order = VM_NFREEORDER;
+			vm_phys_split_pages(m, oind, fl, order);
+			return (m);
+		}
+	}
+
+	/*
+	 * The given pool was empty.  Find the largest
+	 * contiguous, power-of-two-sized set of pages in any
+	 * pool.  Transfer these pages to the given pool, and
+	 * use them to satisfy the allocation.
+	 */
+	for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
+		for (pind = 0; pind < VM_NFREEPOOL; pind++) {
+			alt = vm_phys_free_queues[flind][pind];
+			m = TAILQ_FIRST(&alt[oind].pl);
 			if (m != NULL) {
-				TAILQ_REMOVE(&fl[oind].pl, m, pageq);
-				fl[oind].lcnt--;
+				TAILQ_REMOVE(&alt[oind].pl, m, pageq);
+				alt[oind].lcnt--;
 				m->order = VM_NFREEORDER;
+				vm_phys_set_pool(pool, m, oind);
 				vm_phys_split_pages(m, oind, fl, order);
 				return (m);
 			}
 		}
-
-		/*
-		 * The given pool was empty.  Find the largest
-		 * contiguous, power-of-two-sized set of pages in any
-		 * pool.  Transfer these pages to the given pool, and
-		 * use them to satisfy the allocation.
-		 */
-		for (oind = VM_NFREEORDER - 1; oind >= order; oind--) {
-			for (pind = 0; pind < VM_NFREEPOOL; pind++) {
-				alt = vm_phys_free_queues[flind][pind];
-				m = TAILQ_FIRST(&alt[oind].pl);
-				if (m != NULL) {
-					TAILQ_REMOVE(&alt[oind].pl, m, pageq);
-					alt[oind].lcnt--;
-					m->order = VM_NFREEORDER;
-					vm_phys_set_pool(pool, m, oind);
-					vm_phys_split_pages(m, oind, fl, order);
-					return (m);
-				}
-			}
-		}
 	}
 	return (NULL);
 }
@@ -592,7 +610,7 @@ vm_phys_alloc_contig(unsigned long npages, vm_paddr_t low, vm_paddr_t high,
 {
 	struct vm_freelist *fl;
 	struct vm_phys_seg *seg;
-	vm_object_t m_object;
+	struct vnode *vp;
 	vm_paddr_t pa, pa_last, size;
 	vm_page_t deferred_vdrop_list, m, m_ret;
 	int flind, i, oind, order, pind;
@@ -687,50 +705,19 @@ done:
 	vm_phys_split_pages(m_ret, oind, fl, order);
 	for (i = 0; i < npages; i++) {
 		m = &m_ret[i];
-		KASSERT(m->queue == PQ_NONE,
-		    ("vm_phys_alloc_contig: page %p has unexpected queue %d",
-		    m, m->queue));
-		KASSERT(m->wire_count == 0,
-		    ("vm_phys_alloc_contig: page %p is wired", m));
-		KASSERT(m->hold_count == 0,
-		    ("vm_phys_alloc_contig: page %p is held", m));
-		KASSERT(m->busy == 0,
-		    ("vm_phys_alloc_contig: page %p is busy", m));
-		KASSERT(m->dirty == 0,
-		    ("vm_phys_alloc_contig: page %p is dirty", m));
-		KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
-		    ("vm_phys_alloc_contig: page %p has unexpected memattr %d",
-		    m, pmap_page_get_memattr(m)));
-		if ((m->flags & PG_CACHED) != 0) {
-			m->valid = 0;
-			m_object = m->object;
-			vm_page_cache_remove(m);
-			if (m_object->type == OBJT_VNODE &&
-			    m_object->cache == NULL) {
-				/*
-				 * Enqueue the vnode for deferred vdrop().
-				 *
-				 * Unmanaged pages don't use "pageq", so it
-				 * can be safely abused to construct a short-
-				 * lived queue of vnodes.
-				 */
-				m->pageq.tqe_prev = m_object->handle;
-				m->pageq.tqe_next = deferred_vdrop_list;
-				deferred_vdrop_list = m;
-			}
-		} else {
-			KASSERT(VM_PAGE_IS_FREE(m),
-			    ("vm_phys_alloc_contig: page %p is not free", m));
-			KASSERT(m->valid == 0,
-			    ("vm_phys_alloc_contig: free page %p is valid", m));
-			cnt.v_free_count--;
+		vp = vm_page_alloc_init(m);
+		if (vp != NULL) {
+			/*
+			 * Enqueue the vnode for deferred vdrop().
+			 *
+			 * Unmanaged pages don't use "pageq", so it
+			 * can be safely abused to construct a short-
+			 * lived queue of vnodes.
+			 */
+			m->pageq.tqe_prev = (void *)vp;
+			m->pageq.tqe_next = deferred_vdrop_list;
+			deferred_vdrop_list = m;
 		}
-		if (m->flags & PG_ZERO)
-			vm_page_zero_count--;
-		/* Don't clear the PG_ZERO flag; we'll need it later. */
-		m->flags = PG_UNMANAGED | (m->flags & PG_ZERO);
-		m->oflags = 0;
-		/* Unmanaged pages don't use "act_count". */
 	}
 	for (; i < roundup2(npages, 1 << imin(oind, order)); i++) {
 		m = &m_ret[i];
diff --git a/sys/vm/vm_phys.h b/sys/vm/vm_phys.h
index 0e012c3..0dbd96a 100644
--- a/sys/vm/vm_phys.h
+++ b/sys/vm/vm_phys.h
@@ -44,6 +44,7 @@ void vm_phys_add_page(vm_paddr_t pa);
 vm_page_t vm_phys_alloc_contig(unsigned long npages,
     vm_paddr_t low, vm_paddr_t high,
     unsigned long alignment, unsigned long boundary);
+vm_page_t vm_phys_alloc_freelist_pages(int flind, int pool, int order);
 vm_page_t vm_phys_alloc_pages(int pool, int order);
 vm_paddr_t vm_phys_bootstrap_alloc(vm_size_t size, unsigned long alignment);
 void vm_phys_free_pages(vm_page_t m, int order);