X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=linux_module%2Fbuddy.c;h=ed692c8ef4730851eb48192ca0d468362a3ba8de;hb=a8686374c6aa74d805b32e7675f42f7ab9a0b348;hp=548c7ff45d4f509d87a5e83dadc50e11f53c36bf;hpb=84c84a909e8c3374e2cd7768ef4240ca9d73222f;p=palacios.git

diff --git a/linux_module/buddy.c b/linux_module/buddy.c
index 548c7ff..ed692c8 100644
--- a/linux_module/buddy.c
+++ b/linux_module/buddy.c
@@ -153,9 +153,9 @@ int buddy_add_pool(struct buddy_memzone * zone,
 	return -1;
     }
 
-    mp = palacios_alloc_node_extended(sizeof(struct buddy_mempool), GFP_KERNEL, zone->node_id);
+    mp = palacios_alloc_extended(sizeof(struct buddy_mempool), GFP_KERNEL, zone->node_id);
 
-    if (IS_ERR(mp)) {
+    if (!mp) {
 	ERROR("Could not allocate mempool\n");
 	return -1;
     }
@@ -170,10 +170,17 @@ int buddy_add_pool(struct buddy_memzone * zone,
     /* Allocate a bitmap with 1 bit per minimum-sized block */
     mp->num_blocks = (1UL << pool_order) / (1UL << zone->min_order);
 
-    mp->tag_bits   = palacios_alloc_node_extended(
+    mp->tag_bits   = palacios_alloc_extended(
 				  BITS_TO_LONGS(mp->num_blocks) * sizeof(long), GFP_KERNEL, zone->node_id
 				  );
 
+    if (!(mp->tag_bits)) { 
+	ERROR("Could not allocate tag_bits\n");
+	palacios_free(mp);
+	return -1;
+    }
+	
+
     /* Initially mark all minimum-sized blocks as allocated */
     bitmap_zero(mp->tag_bits, mp->num_blocks);
 
@@ -210,6 +217,7 @@ static int __buddy_remove_mempool(struct buddy_memzone * zone,
     struct buddy_mempool * pool = NULL;
     struct block * block = NULL;
 
+
     pool = find_mempool(zone, base_addr);
 
     if (pool == NULL) {
@@ -217,16 +225,30 @@ static int __buddy_remove_mempool(struct buddy_memzone * zone,
 	return -1;
     }
 
-    if (!bitmap_empty(pool->tag_bits, pool->num_blocks)) {
-	ERROR("Trying to remove an in use memory pool\n");
-	return -1;
+    block = (struct block *)__va(pool->base_addr);
+
+    INFO("Removing Mempool %p, base=%p\n",pool,block);
+
+    // The largest order block in the memory pool must be free
+    if (!is_available(pool, block)) { 
+	if (!force) { 
+	    ERROR("Trying to remove an in use memory pool\n");
+	    *user_metadata=0;
+	    return -1;
+	} else {
+	    WARNING("Forcefully removing in use memory pool\n");
+	}
     }
 
     *user_metadata = pool->user_metadata;
     
-    block = (struct block *)__va(pool->base_addr);
-
-    list_del(&(block->link));
+    if (is_available(pool,block)) { 
+	list_del(&(block->link));
+    } else {
+	// we may not be on the free list if we are being
+	// forcibly removed before all allocations are freed
+    }
+	
     rb_erase(&(pool->tree_node), &(zone->mempools));
 
     palacios_free(pool->tag_bits);
@@ -245,6 +267,7 @@ int buddy_remove_pool(struct buddy_memzone * zone,
     unsigned long flags = 0;
     int ret = 0;
 
+
     palacios_spinlock_lock_irqsave(&(zone->lock), flags);    
     ret = __buddy_remove_mempool(zone, base_addr, force, user_metadata);
     palacios_spinlock_unlock_irqrestore(&(zone->lock), flags);
@@ -255,23 +278,26 @@ int buddy_remove_pool(struct buddy_memzone * zone,
 
 
 
+
 /**
  * Allocates a block of memory of the requested size (2^order bytes).
  *
  * Arguments:
  *       [IN] mp:    Buddy system memory allocator object.
  *       [IN] order: Block size to allocate (2^order bytes).
- *
+ *       [IN] filter_func: returns nonzero if given paddr is OK to use
+ *       [IN] filter_state: opaque argument to filter_func
  * Returns:
  *       Success: Pointer to the start of the allocated memory block.
  *       Failure: NULL
  */
 uintptr_t
-buddy_alloc(struct buddy_memzone *zone, unsigned long order)
+buddy_alloc(struct buddy_memzone *zone, unsigned long order, int (*filter_func)(void *paddr, void *filter_state), void *filter_state)
 {
     unsigned long j;
     struct buddy_mempool * mp = NULL;
     struct list_head * list = NULL;
+    struct list_head * cur = NULL;
     struct block * block = NULL;
     struct block * buddy_block = NULL;
     unsigned long flags = 0;
@@ -292,13 +318,43 @@ buddy_alloc(struct buddy_memzone *zone, unsigned long order)
 
 	INFO("Order iter=%lu\n", j);
 
-	/* Try to allocate the first block in the order j list */
-	list = &zone->avail[j];
+	block=NULL;
 
-	if (list_empty(list)) 
+	list = &(zone->avail[j]);
+	
+	if (list_empty(list)) {
 	    continue;
+	}
+
+	list_for_each(cur, list) {
+	    block = list_entry(cur, struct block, link);
+
+	    if (!filter_func) {
+		// without a filter, we just want the first one
+		break;
+	    } else {
+
+		void *block_pa = (void*)__pa(block);
+
+		if (filter_func(block_pa,filter_state)) { 
+		    // this block will work
+		    break;
+		} else {
+		    // this block won't work
+		    block=NULL;
+		    continue;
+		}
+
+	    }
+	}
+	
+	if (!block) { 
+	    // uh oh, no block, look to next order
+	    continue;
+	}
+
+	// have appropriate block, will allocate
 
-	block = list_entry(list->next, struct block, link);
 	list_del(&(block->link));
 
 	mp = block->mp;
@@ -307,11 +363,6 @@ buddy_alloc(struct buddy_memzone *zone, unsigned long order)
 
 	INFO("pool=%p, block=%p, order=%lu, j=%lu\n", mp, block, order, j);
 
-	/*
-	palacios_spinlock_unlock_irqrestore(&(zone->lock), flags);
-	return 0;
-	*/
-
 	/* Trim if a higher order block than necessary was allocated */
 	while (j > order) {
 	    --j;
@@ -338,7 +389,7 @@ buddy_alloc(struct buddy_memzone *zone, unsigned long order)
 /**
  * Returns a block of memory to the buddy system memory allocator.
  */
-void
+int
 buddy_free(
 	//!    Buddy system memory allocator object.
 	struct buddy_memzone *	zone,
@@ -373,9 +424,9 @@ buddy_free(
     pool = find_mempool(zone, addr);
 
     if ((pool == NULL) || (order > pool->pool_order)) {
-	WARNING("Attempted to free an invalid page address (%p)\n", (void *)addr);
+	WARNING("Attempted to free an invalid page address (%p) - pool=%p order=%lu\n", (void *)addr,pool,order);
 	palacios_spinlock_unlock_irqrestore(&(zone->lock), flags);
-	return;
+	return -1;
     }
 
 
@@ -385,7 +436,7 @@ buddy_free(
     if (is_available(pool, block)) {
 	ERROR("Error: Freeing an available block\n");
 	palacios_spinlock_unlock_irqrestore(&(zone->lock), flags);
-	return;
+	return -1;
     }
 
     pool->num_free_blocks += (1UL << (order - zone->min_order));
@@ -416,6 +467,8 @@ buddy_free(
     list_add(&(block->link), &(zone->avail[order]));
 
     palacios_spinlock_unlock_irqrestore(&(zone->lock), flags);
+
+    return 0;
 }
 
 
@@ -481,9 +534,9 @@ zone_mem_show(struct seq_file * s, void * v) {
 
 
 static int zone_proc_open(struct inode * inode, struct file * filp) {
-    struct proc_dir_entry * proc_entry = PDE(inode);
-    INFO("proc_entry at %p, data at %p\n", proc_entry, proc_entry->data);
-    return single_open(filp, zone_mem_show, proc_entry->data);
+    struct proc_dir_entry * proc_entry = PAL_PDE(inode);
+    INFO("proc_entry at %p, data at %p\n", proc_entry, PAL_PROC_GETDATA(inode));
+    return single_open(filp, zone_mem_show, PAL_PROC_GETDATA(inode));
 }
 
 
@@ -496,28 +549,77 @@ static struct file_operations zone_proc_ops = {
 };
 
 
-extern struct proc_dir_entry * palacios_proc_dir;
-
-void buddy_deinit(struct buddy_memzone * zone) {
+void buddy_deinit(struct buddy_memzone * zone, int (*free_callback)(void *user_metadata)) {
     unsigned long flags;
-
+    struct rb_node *node;
+    struct buddy_mempool **pools;
+    unsigned long long base_addr;
+    void *meta;
+    int i;
+    unsigned long num_in_tree;
+
+    pools = (struct buddy_mempool **) palacios_alloc(sizeof(struct buddy_mempool *)*zone->num_pools);
+    if (!pools) { 
+	ERROR("Cannot allocate space for doing deinit of memory zone\n");
+	return ;
+    }
+    
+    // We will lock only to build up the memory pool list
+    // when we free memory, we need to be able to support free callbacks
+    // that could block.  This does leave a race with adds, allocs, and frees, however
+    // In Palacios, we expect a deinit will only really happen on the module unload
+    // so this should not present a problem
     palacios_spinlock_lock_irqsave(&(zone->lock), flags);
 
-    // for each pool, free it
-#warning We really need to free the memory pools here
+    // because it does not appear possible to erase while iterating
+    // over the rb tree, we do the following contorted mess
+    // get the pools 
+    for (num_in_tree=0, node=rb_first(&(zone->mempools));
+	 node && num_in_tree<zone->num_pools;
+	 node=rb_next(node), num_in_tree++) { 
+	
+	pools[num_in_tree]=rb_entry(node,struct buddy_mempool, tree_node);
+    }
 
     palacios_spinlock_unlock_irqrestore(&(zone->lock), flags);
-    
+
+    if (num_in_tree != zone->num_pools) { 
+	WARNING("Odd, the number of pools in the tree is %lu, but the zone reports %lu\n",
+		num_in_tree, zone->num_pools);
+    }
+
+    // now we'll free the memory
+    // note that buddy_remove_mempool also removes them
+    // from the rb tree, and frees them
+    for (i=0;i<num_in_tree;i++) { 
+	base_addr = pools[i]->base_addr;
+	
+	if (buddy_remove_pool(zone, base_addr, 1, &meta)) {
+	    WARNING("Cannot remove memory pool at %p during zone deinit...\n",(void*)(base_addr));
+	    continue;
+	}
+	
+	// pool and node are now gone... 
+
+	// invoke the callback to free the actual memory, if any
+	if (free_callback) { 
+	    free_callback(meta);
+	}
+    }
+
+
+    // get rid of /proc entry
     {
 	char proc_file_name[128];
 
 	memset(proc_file_name, 0, 128);
-	snprintf(proc_file_name, 128, "v3-mem%d", zone->node_id);
+	snprintf(proc_file_name, 128, "v3-mem%u", zone->node_id);
 
-	remove_proc_entry(proc_file_name, palacios_proc_dir);
+	remove_proc_entry(proc_file_name, palacios_get_procdir());
     }
 
 
+    palacios_free(pools);
     palacios_free(zone->avail);
     palacios_free(zone);
 
@@ -553,6 +655,8 @@ buddy_init(
 {
     struct buddy_memzone * zone = NULL;
     unsigned long i;
+	struct proc_dir_entry * zone_entry = NULL;
+	char proc_file_name[128];
 
     DEBUG("Initializing Memory zone with up to %lu bit blocks on Node %d\n", max_order, node_id);
 
@@ -565,11 +669,11 @@ buddy_init(
     if (min_order > max_order)
 	return NULL;
 
-    zone = palacios_alloc_node_extended(sizeof(struct buddy_memzone), GFP_KERNEL, node_id);
+    zone = palacios_alloc_extended(sizeof(struct buddy_memzone), GFP_KERNEL, node_id);
 	
     INFO("Allocated zone at %p\n", zone);
 
-    if (IS_ERR(zone)) {
+    if (!zone) {
 	ERROR("Could not allocate memzone\n");
 	return NULL;
     }
@@ -581,7 +685,13 @@ buddy_init(
     zone->node_id = node_id;
 
     /* Allocate a list for every order up to the maximum allowed order */
-    zone->avail = palacios_alloc_node_extended((max_order + 1) * sizeof(struct list_head), GFP_KERNEL, zone->node_id);
+    zone->avail = palacios_alloc_extended((max_order + 1) * sizeof(struct list_head), GFP_KERNEL, zone->node_id);
+
+    if (!(zone->avail)) { 
+	ERROR("Unable to allocate space for zone list\n");
+	palacios_free(zone);
+	return NULL;
+    }
 
     INFO("Allocated free lists at %p\n", zone->avail);
 
@@ -597,22 +707,17 @@ buddy_init(
 
     INFO("Allocated zone at %p\n", zone);
 
-    {
-	struct proc_dir_entry * zone_entry = NULL;
-	char proc_file_name[128];
 
 	memset(proc_file_name, 0, 128);
-	snprintf(proc_file_name, 128, "v3-mem%d", zone->node_id);
+	snprintf(proc_file_name, 128, "v3-mem%u", zone->node_id);
 
-	zone_entry = create_proc_entry(proc_file_name, 0444, palacios_proc_dir);
-	if (zone_entry) {
-	    zone_entry->proc_fops = &zone_proc_ops;
-	    zone_entry->data = zone;
+    PAL_PROC_CREATE_DATA(zone_entry, proc_file_name, 0444, palacios_get_procdir(), &zone_proc_ops, zone);
+    if (zone_entry) {
+	    INFO("Successfully created /proc/v3vee/v3-mem%d\n", zone->node_id);
 	} else {
-	    ERROR("Error creating memory zone proc file\n");
+	    ERROR("Cannot create /proc/v3vee/v3-mem%d\n", zone->node_id);
 	}
 
-    }
 
     return zone;
 }