WebSVN – HelenOS – Path Comparison – / – /branches/dd/kernel/generic/src/mm/ Rev 4054 and /branches/dd/kernel/generic/src/mm/ Rev 4055

Ignore whitespace Rev 4054 → Rev 4055

 /branches/dd/kernel/generic/src/mm/slab.c
 ,19 → 167,19
  * Allocate frames for slab space and initialize
  *
  */
-static slab_t * slab_space_alloc(slab_cache_t *cache, int flags)
+static slab_t *slab_space_alloc(slab_cache_t *cache, int flags)
 {
     void *data;
     slab_t *slab;
     size_t fsize;
     unsigned int i;
-    unsigned int zone = 0;
+    count_t zone = 0;
     data = frame_alloc_generic(cache->order, FRAME_KA | flags, &zone);
     if (!data) {
         return NULL;
     }
-    if (! (cache->flags & SLAB_CACHE_SLINSIDE)) {
+    if (!(cache->flags & SLAB_CACHE_SLINSIDE)) {
         slab = slab_alloc(slab_extern_cache, flags);
         if (!slab) {
             frame_free(KA2PA(data));
 ,7 → 200,7
     slab->cache = cache;
     for (i = 0; i < cache->objects; i++)
-        *((int *) (slab->start + i*cache->size)) = i+1;
+        *((int *) (slab->start + i*cache->size)) = i + 1;
     atomic_inc(&cache->allocated_slabs);
     return slab;
 ,8 → 239,7
  *
  * @return Number of freed pages
  */
-static count_t slab_obj_destroy(slab_cache_t *cache, void *obj,
-                slab_t *slab)
+static count_t slab_obj_destroy(slab_cache_t *cache, void *obj, slab_t *slab)
 {
     int freed = 0;
 ,7 → 255,7
     ASSERT(slab->available < cache->objects);
     *((int *)obj) = slab->nextavail;
-    slab->nextavail = (obj - slab->start)/cache->size;
+    slab->nextavail = (obj - slab->start) / cache->size;
     slab->available++;
     /* Move it to correct list */
 ,7 → 280,7
  *
  * @return Object address or null
  */
-static void * slab_obj_create(slab_cache_t *cache, int flags)
+static void *slab_obj_create(slab_cache_t *cache, int flags)
 {
     slab_t *slab;
     void *obj;
 ,7 → 300,8
             return NULL;
         spinlock_lock(&cache->slablock);
     } else {
-        slab = list_get_instance(cache->partial_slabs.next, slab_t, link);
+        slab = list_get_instance(cache->partial_slabs.next, slab_t,
+            link);
         list_remove(&slab->link);
     }
     obj = slab->start + slab->nextavail * cache->size;
 ,8 → 332,7
  *
  * @param first If true, return first, else last mag
  */
-static slab_magazine_t * get_mag_from_cache(slab_cache_t *cache,
-                        int first)
+static slab_magazine_t *get_mag_from_cache(slab_cache_t *cache, int first)
 {
     slab_magazine_t *mag = NULL;
     link_t *cur;
 ,8 → 367,7
  *
  * @return Number of freed pages
  */
-static count_t magazine_destroy(slab_cache_t *cache,
-                slab_magazine_t *mag)
+static count_t magazine_destroy(slab_cache_t *cache, slab_magazine_t *mag)
 {
     unsigned int i;
     count_t frames = 0;
 ,7 → 387,7
  *
  * Assume cpu_magazine lock is held
  */
-static slab_magazine_t * get_full_current_mag(slab_cache_t *cache)
+static slab_magazine_t *get_full_current_mag(slab_cache_t *cache)
 {
     slab_magazine_t *cmag, *lastmag, *newmag;
 ,7 → 421,7
  *
  * @return Pointer to object or NULL if not available
  */
-static void * magazine_obj_get(slab_cache_t *cache)
+static void *magazine_obj_get(slab_cache_t *cache)
 {
     slab_magazine_t *mag;
     void *obj;
 ,7 → 456,7
  *   allocate new, exchange last & current
  *
  */
-static slab_magazine_t * make_empty_current_mag(slab_cache_t *cache)
+static slab_magazine_t *make_empty_current_mag(slab_cache_t *cache)
 {
     slab_magazine_t *cmag,*lastmag,*newmag;
 ,7 → 528,8
 static unsigned int comp_objects(slab_cache_t *cache)
 {
     if (cache->flags & SLAB_CACHE_SLINSIDE)
-        return ((PAGE_SIZE << cache->order) - sizeof(slab_t)) / cache->size;
+        return ((PAGE_SIZE << cache->order) - sizeof(slab_t)) /
+            cache->size;
     else
         return (PAGE_SIZE << cache->order) / cache->size;
 }
 ,28 → 556,25
     ASSERT(_slab_initialized >= 2);
-    cache->mag_cache = malloc(sizeof(slab_mag_cache_t)*config.cpu_count,0);
+    cache->mag_cache = malloc(sizeof(slab_mag_cache_t) * config.cpu_count,
+);
     for (i = 0; i < config.cpu_count; i++) {
-        memsetb((uintptr_t)&cache->mag_cache[i],
-            sizeof(cache->mag_cache[i]), 0);
-        spinlock_initialize(&cache->mag_cache[i].lock, "slab_maglock_cpu");
+        memsetb(&cache->mag_cache[i], sizeof(cache->mag_cache[i]), 0);
+        spinlock_initialize(&cache->mag_cache[i].lock,
+            "slab_maglock_cpu");
     }
 }
 /** Initialize allocated memory as a slab cache */
 static void
-_slab_cache_create(slab_cache_t *cache,
-           char *name,
-           size_t size,
-           size_t align,
-           int (*constructor)(void *obj, int kmflag),
-           int (*destructor)(void *obj),
-           int flags)
+_slab_cache_create(slab_cache_t *cache, char *name, size_t size, size_t align,
+    int (*constructor)(void *obj, int kmflag), int (*destructor)(void *obj),
+    int flags)
 {
     int pages;
     ipl_t ipl;
-    memsetb((uintptr_t)cache, sizeof(*cache), 0);
+    memsetb(cache, sizeof(*cache), 0);
     cache->name = name;
     if (align < sizeof(unative_t))
 ,7 → 592,7
     list_initialize(&cache->magazines);
     spinlock_initialize(&cache->slablock, "slab_lock");
     spinlock_initialize(&cache->maglock, "slab_maglock");
-    if (! (cache->flags & SLAB_CACHE_NOMAGAZINE))
+    if (!(cache->flags & SLAB_CACHE_NOMAGAZINE))
         make_magcache(cache);
     /* Compute slab sizes, object counts in slabs etc. */
 ,7 → 605,7
     if (pages == 1)
         cache->order = 0;
     else
-        cache->order = fnzb(pages-1)+1;
+        cache->order = fnzb(pages - 1) + 1;
     while (badness(cache) > SLAB_MAX_BADNESS(cache)) {
         cache->order += 1;
 ,18 → 626,16
 }
 /** Create slab cache  */
-slab_cache_t * slab_cache_create(char *name,
-                 size_t size,
-                 size_t align,
-                 int (*constructor)(void *obj, int kmflag),
-                 int (*destructor)(void *obj),
-                 int flags)
+slab_cache_t *
+slab_cache_create(char *name, size_t size, size_t align,
+    int (*constructor)(void *obj, int kmflag), int (*destructor)(void *obj),
+    int flags)
 {
     slab_cache_t *cache;
     cache = slab_alloc(&slab_cache_cache, 0);
     _slab_cache_create(cache, name, size, align, constructor, destructor,
-               flags);
+        flags);
     return cache;
 }
 ,7 → 659,7
      * endless loop
      */
     magcount = atomic_get(&cache->magazine_counter);
-    while (magcount-- && (mag=get_mag_from_cache(cache,0))) {
+    while (magcount-- && (mag=get_mag_from_cache(cache, 0))) {
         frames += magazine_destroy(cache,mag);
         if (!(flags & SLAB_RECLAIM_ALL) && frames)
             break;
 ,8 → 712,8
     _slab_reclaim(cache, SLAB_RECLAIM_ALL);
     /* All slabs must be empty */
-    if (!list_empty(&cache->full_slabs) \
-        || !list_empty(&cache->partial_slabs))
+    if (!list_empty(&cache->full_slabs) ||
+        !list_empty(&cache->partial_slabs))
         panic("Destroying cache that is not empty.");
     if (!(cache->flags & SLAB_CACHE_NOMAGAZINE))
 ,9 → 721,8
     slab_free(&slab_cache_cache, cache);
 }
-/** Allocate new object from cache - if no flags given, always returns
-    memory */
-void * slab_alloc(slab_cache_t *cache, int flags)
+/** Allocate new object from cache - if no flags given, always returns memory */
+void *slab_alloc(slab_cache_t *cache, int flags)
 {
     ipl_t ipl;
     void *result = NULL;
 ,9 → 751,8
     ipl = interrupts_disable();
-    if ((cache->flags & SLAB_CACHE_NOMAGAZINE) \
-        || magazine_obj_put(cache, obj)) {
+    if ((cache->flags & SLAB_CACHE_NOMAGAZINE) ||
+        magazine_obj_put(cache, obj)) {
         slab_obj_destroy(cache, obj, slab);
     }
 ,7 → 779,8
      * memory allocation from interrupts can deadlock.
      */
-    for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) {
+    for (cur = slab_cache_list.next; cur != &slab_cache_list;
+        cur = cur->next) {
         cache = list_get_instance(cur, slab_cache_t, link);
         frames += _slab_reclaim(cache, flags);
     }
 ,22 → 794,77
 /* Print list of slabs */
 void slab_print_list(void)
 {
-    slab_cache_t *cache;
-    link_t *cur;
-    ipl_t ipl;
-    ipl = interrupts_disable();
-    spinlock_lock(&slab_cache_lock);
-    printf("slab name        size     pages  obj/pg slabs  cached allocated ctl\n");
-    printf("---------------- -------- ------ ------ ------ ------ --------- ---\n");
-    for (cur = slab_cache_list.next; cur != &slab_cache_list; cur = cur->next) {
+    int skip = 0;
+    printf("slab name        size     pages  obj/pg slabs  cached allocated"
+        " ctl\n");
+    printf("---------------- -------- ------ ------ ------ ------ ---------"
+        " ---\n");
+    while (true) {
+        slab_cache_t *cache;
+        link_t *cur;
+        ipl_t ipl;
+        int i;
+        /*
+         * We must not hold the slab_cache_lock spinlock when printing
+         * the statistics. Otherwise we can easily deadlock if the print
+         * needs to allocate memory.
+         *
+         * Therefore, we walk through the slab cache list, skipping some
+         * amount of already processed caches during each iteration and
+         * gathering statistics about the first unprocessed cache. For
+         * the sake of printing the statistics, we realese the
+         * slab_cache_lock and reacquire it afterwards. Then the walk
+         * starts again.
+         *
+         * This limits both the efficiency and also accuracy of the
+         * obtained statistics. The efficiency is decreased because the
+         * time complexity of the algorithm is quadratic instead of
+         * linear. The accuracy is impacted because we drop the lock
+         * after processing one cache. If there is someone else
+         * manipulating the cache list, we might omit an arbitrary
+         * number of caches or process one cache multiple times.
+         * However, we don't bleed for this algorithm for it is only
+         * statistics.
+         */
+        ipl = interrupts_disable();
+        spinlock_lock(&slab_cache_lock);
+        for (i = 0, cur = slab_cache_list.next;
+            i < skip && cur != &slab_cache_list;
+            i++, cur = cur->next)
+            ;
+        if (cur == &slab_cache_list) {
+            spinlock_unlock(&slab_cache_lock);
+            interrupts_restore(ipl);
+            break;
+        }
+        skip++;
         cache = list_get_instance(cur, slab_cache_t, link);
+        char *name = cache->name;
+        uint8_t order = cache->order;
+        size_t size = cache->size;
+        unsigned int objects = cache->objects;
+        long allocated_slabs = atomic_get(&cache->allocated_slabs);
+        long cached_objs = atomic_get(&cache->cached_objs);
+        long allocated_objs = atomic_get(&cache->allocated_objs);
+        int flags = cache->flags;
-        printf("%-16s %8zd %6zd %6zd %6zd %6zd %9zd %-3s\n", cache->name, cache->size, (1 << cache->order), cache->objects, atomic_get(&cache->allocated_slabs), atomic_get(&cache->cached_objs), atomic_get(&cache->allocated_objs), cache->flags & SLAB_CACHE_SLINSIDE ? "in" : "out");
+        spinlock_unlock(&slab_cache_lock);
+        interrupts_restore(ipl);
+        printf("%-16s %8" PRIs " %6d %6u %6ld %6ld %9ld %-3s\n",
+            name, size, (1 << order), objects, allocated_slabs,
+            cached_objs, allocated_objs,
+            flags & SLAB_CACHE_SLINSIDE ? "in" : "out");
     }
-    spinlock_unlock(&slab_cache_lock);
-    interrupts_restore(ipl);
 }
 void slab_cache_init(void)
 ,32 → 872,24
     int i, size;
     /* Initialize magazine cache */
-    _slab_cache_create(&mag_cache,
-               "slab_magazine",
-               sizeof(slab_magazine_t)+SLAB_MAG_SIZE*sizeof(void*),
-               sizeof(uintptr_t),
-               NULL, NULL,
-               SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE);
+    _slab_cache_create(&mag_cache, "slab_magazine",
+        sizeof(slab_magazine_t) + SLAB_MAG_SIZE * sizeof(void*),
+        sizeof(uintptr_t), NULL, NULL, SLAB_CACHE_NOMAGAZINE |
+        SLAB_CACHE_SLINSIDE);
     /* Initialize slab_cache cache */
-    _slab_cache_create(&slab_cache_cache,
-               "slab_cache",
-               sizeof(slab_cache_cache),
-               sizeof(uintptr_t),
-               NULL, NULL,
-               SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE);
+    _slab_cache_create(&slab_cache_cache, "slab_cache",
+        sizeof(slab_cache_cache), sizeof(uintptr_t), NULL, NULL,
+        SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE);
     /* Initialize external slab cache */
-    slab_extern_cache = slab_cache_create("slab_extern",
-                          sizeof(slab_t),
-, NULL, NULL,
-                          SLAB_CACHE_SLINSIDE | SLAB_CACHE_MAGDEFERRED);
+    slab_extern_cache = slab_cache_create("slab_extern", sizeof(slab_t), 0,
+        NULL, NULL, SLAB_CACHE_SLINSIDE | SLAB_CACHE_MAGDEFERRED);
     /* Initialize structures for malloc */
-    for (i=0, size=(1<<SLAB_MIN_MALLOC_W);
-         i < (SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1);
-         i++, size <<= 1) {
-        malloc_caches[i] = slab_cache_create(malloc_names[i],
-                             size, 0,
-                             NULL,NULL, SLAB_CACHE_MAGDEFERRED);
+    for (i = 0, size = (1 << SLAB_MIN_MALLOC_W);
+        i < (SLAB_MAX_MALLOC_W - SLAB_MIN_MALLOC_W + 1);
+        i++, size <<= 1) {
+        malloc_caches[i] = slab_cache_create(malloc_names[i], size, 0,
+            NULL, NULL, SLAB_CACHE_MAGDEFERRED);
     }
 #ifdef CONFIG_DEBUG
     _slab_initialized = 1;
 ,9 → 914,11
     spinlock_lock(&slab_cache_lock);
-    for (cur=slab_cache_list.next; cur != &slab_cache_list;cur=cur->next){
+    for (cur = slab_cache_list.next; cur != &slab_cache_list;
+        cur = cur->next){
         s = list_get_instance(cur, slab_cache_t, link);
-        if ((s->flags & SLAB_CACHE_MAGDEFERRED) != SLAB_CACHE_MAGDEFERRED)
+        if ((s->flags & SLAB_CACHE_MAGDEFERRED) !=
+            SLAB_CACHE_MAGDEFERRED)
             continue;
         make_magcache(s);
         s->flags &= ~SLAB_CACHE_MAGDEFERRED;
 ,7 → 929,7
 /**************************************/
 /* kalloc/kfree functions             */
-void * malloc(unsigned int size, int flags)
+void *malloc(unsigned int size, int flags)
 {
     ASSERT(_slab_initialized);
     ASSERT(size && size <= (1 << SLAB_MAX_MALLOC_W));
 ,7 → 942,7
     return slab_alloc(malloc_caches[idx], flags);
 }
-void * realloc(void *ptr, unsigned int size, int flags)
+void *realloc(void *ptr, unsigned int size, int flags)
 {
     ASSERT(_slab_initialized);
     ASSERT(size <= (1 << SLAB_MAX_MALLOC_W));

 /branches/dd/kernel/generic/src/mm/tlb.c
 ,7 → 173,7
             tlb_invalidate_pages(asid, page, count);
             break;
         default:
-            panic("unknown type (%d)\n", type);
+            panic("Unknown type (%d).", type);
             break;
         }
         if (type == TLB_INVL_ALL)

 /branches/dd/kernel/generic/src/mm/backend_anon.c
 ,7 → 78,6
 int anon_page_fault(as_area_t *area, uintptr_t addr, pf_access_t access)
 {
     uintptr_t frame;
-    bool dirty = false;
     if (!as_area_check_access(area, access))
         return AS_PF_FAULT;
 ,7 → 105,7
              */
             for (i = 0; i < leaf->keys; i++) {
                 if (leaf->key[i] ==
-                    ALIGN_DOWN(addr, PAGE_SIZE)) {
+                    ALIGN_DOWN(addr, PAGE_SIZE) - area->base) {
                     allocate = false;
                     break;
                 }
 ,8 → 112,7
             }
             if (allocate) {
                 frame = (uintptr_t) frame_alloc(ONE_FRAME, 0);
-                memsetb(PA2KA(frame), FRAME_SIZE, 0);
-                dirty = true;
+                memsetb((void *) PA2KA(frame), FRAME_SIZE, 0);
                 /*
                  * Insert the address of the newly allocated
 ,8 → 142,7
          *   the different causes
          */
         frame = (uintptr_t) frame_alloc(ONE_FRAME, 0);
-        memsetb(PA2KA(frame), FRAME_SIZE, 0);
-        dirty = true;
+        memsetb((void *) PA2KA(frame), FRAME_SIZE, 0);
     }
     /*
 ,7 → 152,7
      */
     page_mapping_insert(AS, addr, frame, as_area_get_flags(area));
     if (!used_space_insert(area, ALIGN_DOWN(addr, PAGE_SIZE), 1))
-        panic("Could not insert used space.\n");
+        panic("Cannot insert used space.");
     return AS_PF_OK;
 }

 /branches/dd/kernel/generic/src/mm/as.c
 ,7 → 82,6
 #include <arch/mm/cache.h>
 #endif /* CONFIG_VIRT_IDX_DCACHE */
-#ifndef __OBJC__
 /**
  * Each architecture decides what functions will be used to carry out
  * address space operations such as creating or locking page tables.
 ,7 → 92,6
  * Slab for as_t objects.
  */
 static slab_cache_t *as_slab;
-#endif
 /**
  * This lock serializes access to the ASID subsystem.
 ,13 → 111,11
 /** Kernel address space. */
 as_t *AS_KERNEL = NULL;
-static int area_flags_to_page_flags(int aflags);
-static as_area_t *find_area_and_lock(as_t *as, uintptr_t va);
-static bool check_area_conflicts(as_t *as, uintptr_t va, size_t size,
-    as_area_t *avoid_area);
-static void sh_info_remove_reference(share_info_t *sh_info);
+static int area_flags_to_page_flags(int);
+static as_area_t *find_area_and_lock(as_t *, uintptr_t);
+static bool check_area_conflicts(as_t *, uintptr_t, size_t, as_area_t *);
+static void sh_info_remove_reference(share_info_t *);
-#ifndef __OBJC__
 static int as_constructor(void *obj, int flags)
 {
     as_t *as = (as_t *) obj;
 ,7 → 122,7
     int rc;
     link_initialize(&as->inactive_as_with_asid_link);
-    mutex_initialize(&as->lock);
+    mutex_initialize(&as->lock, MUTEX_PASSIVE);
     rc = as_constructor_arch(as, flags);
 ,7 → 135,6
     return as_destructor_arch(as);
 }
-#endif
 /** Initialize address space subsystem. */
 void as_init(void)
 ,33 → 141,29
 {
     as_arch_init();
-#ifndef __OBJC__
     as_slab = slab_cache_create("as_slab", sizeof(as_t), 0,
         as_constructor, as_destructor, SLAB_CACHE_MAGDEFERRED);
-#endif
     AS_KERNEL = as_create(FLAG_AS_KERNEL);
     if (!AS_KERNEL)
-        panic("can't create kernel address space\n");
+        panic("Cannot create kernel address space.");
+    /* Make sure the kernel address space
+     * reference count never drops to zero.
+     */
+    atomic_set(&AS_KERNEL->refcount, 1);
 }
 /** Create address space.
  *
- * @param flags Flags that influence way in wich the address space is created.
+ * @param flags     Flags that influence the way in wich the address space
+ *          is created.
  */
 as_t *as_create(int flags)
 {
     as_t *as;
-#ifdef __OBJC__
-    as = [as_t new];
-    link_initialize(&as->inactive_as_with_asid_link);
-    mutex_initialize(&as->lock);
-    (void) as_constructor_arch(as, flags);
-#else
     as = (as_t *) slab_alloc(as_slab, 0);
-#endif
     (void) as_create_arch(as, 0);
     btree_create(&as->as_area_btree);
 ,7 → 180,7
 #else
     page_table_create(flags);
 #endif
     return as;
 }
 ,6 → 190,8
  * zero), the address space can be destroyed.
  *
  * We know that we don't hold any spinlock.
+ *
+ * @param as        Address space to be destroyed.
  */
 void as_destroy(as_t *as)
 {
 ,11 → 256,7
     interrupts_restore(ipl);
-#ifdef __OBJC__
-    [as free];
-#else
     slab_free(as_slab, as);
-#endif
 }
 /** Create address space area of common attributes.
 ,19 → 263,19
  *
  * The created address space area is added to the target address space.
  *
- * @param as Target address space.
- * @param flags Flags of the area memory.
- * @param size Size of area.
- * @param base Base address of area.
- * @param attrs Attributes of the area.
- * @param backend Address space area backend. NULL if no backend is used.
- * @param backend_data NULL or a pointer to an array holding two void *.
+ * @param as        Target address space.
+ * @param flags     Flags of the area memory.
+ * @param size      Size of area.
+ * @param base      Base address of area.
+ * @param attrs     Attributes of the area.
+ * @param backend   Address space area backend. NULL if no backend is used.
+ * @param backend_data  NULL or a pointer to an array holding two void *.
  *
- * @return Address space area on success or NULL on failure.
+ * @return      Address space area on success or NULL on failure.
  */
 as_area_t *
 as_area_create(as_t *as, int flags, size_t size, uintptr_t base, int attrs,
-           mem_backend_t *backend, mem_backend_data_t *backend_data)
+    mem_backend_t *backend, mem_backend_data_t *backend_data)
 {
     ipl_t ipl;
     as_area_t *a;
 ,7 → 301,7
     a = (as_area_t *) malloc(sizeof(as_area_t), 0);
-    mutex_initialize(&a->lock);
+    mutex_initialize(&a->lock, MUTEX_PASSIVE);
     a->as = as;
     a->flags = flags;
 ,8 → 313,7
     if (backend_data)
         a->backend_data = *backend_data;
     else
-        memsetb((uintptr_t) &a->backend_data, sizeof(a->backend_data),
-);
+        memsetb(&a->backend_data, sizeof(a->backend_data), 0);
     btree_create(&a->used_space);
 ,13 → 327,14
 /** Find address space area and change it.
  *
- * @param as Address space.
- * @param address Virtual address belonging to the area to be changed. Must be
- *     page-aligned.
- * @param size New size of the virtual memory block starting at address.
- * @param flags Flags influencing the remap operation. Currently unused.
+ * @param as        Address space.
+ * @param address   Virtual address belonging to the area to be changed.
+ *          Must be page-aligned.
+ * @param size      New size of the virtual memory block starting at
+ *          address.
+ * @param flags     Flags influencing the remap operation. Currently unused.
  *
- * @return Zero on success or a value from @ref errno.h otherwise.
+ * @return      Zero on success or a value from @ref errno.h otherwise.
  */
 int as_area_resize(as_t *as, uintptr_t address, size_t size, int flags)
 {
 ,7 → 389,7
     if (pages < area->pages) {
         bool cond;
-        uintptr_t start_free = area->base + pages*PAGE_SIZE;
+        uintptr_t start_free = area->base + pages * PAGE_SIZE;
         /*
          * Shrinking the area.
 ,7 → 399,7
         /*
          * Start TLB shootdown sequence.
          */
-        tlb_shootdown_start(TLB_INVL_PAGES, AS->asid, area->base +
+        tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base +
             pages * PAGE_SIZE, area->pages - pages);
         /*
 ,8 → 444,8
                     i = (start_free - b) >> PAGE_WIDTH;
                     if (!used_space_remove(area, start_free,
                         c - i))
-                        panic("Could not remove used "
-                            "space.\n");
+                        panic("Cannot remove used "
+                            "space.");
                 } else {
                     /*
                      * The interval of used space can be
 ,8 → 452,8
                      * completely removed.
                      */
                     if (!used_space_remove(area, b, c))
-                        panic("Could not remove used "
-                            "space.\n");
+                        panic("Cannot remove used "
+                            "space.");
                 }
                 for (; i < c; i++) {
 ,10 → 515,10
 /** Destroy address space area.
  *
- * @param as Address space.
- * @param address Address withing the area to be deleted.
+ * @param as        Address space.
+ * @param address   Address within the area to be deleted.
  *
- * @return Zero on success or a value from @ref errno.h on failure.
+ * @return      Zero on success or a value from @ref errno.h on failure.
  */
 int as_area_destroy(as_t *as, uintptr_t address)
 {
 ,18 → 615,19
  * sh_info of the source area. The process of duplicating the
  * mapping is done through the backend share function.
  *
- * @param src_as Pointer to source address space.
- * @param src_base Base address of the source address space area.
- * @param acc_size Expected size of the source area.
- * @param dst_as Pointer to destination address space.
- * @param dst_base Target base address.
+ * @param src_as    Pointer to source address space.
+ * @param src_base  Base address of the source address space area.
+ * @param acc_size  Expected size of the source area.
+ * @param dst_as    Pointer to destination address space.
+ * @param dst_base  Target base address.
  * @param dst_flags_mask Destination address space area flags mask.
  *
- * @return Zero on success or ENOENT if there is no such task or if there is no
- * such address space area, EPERM if there was a problem in accepting the area
- * or ENOMEM if there was a problem in allocating destination address space
- * area. ENOTSUP is returned if the address space area backend does not support
- * sharing.
+ * @return      Zero on success or ENOENT if there is no such task or if
+ *          there is no such address space area, EPERM if there was
+ *          a problem in accepting the area or ENOMEM if there was a
+ *          problem in allocating destination address space area.
+ *          ENOTSUP is returned if the address space area backend
+ *          does not support sharing.
  */
 int as_area_share(as_t *src_as, uintptr_t src_base, size_t acc_size,
     as_t *dst_as, uintptr_t dst_base, int dst_flags_mask)
 ,7 → 688,7
     sh_info = src_area->sh_info;
     if (!sh_info) {
         sh_info = (share_info_t *) malloc(sizeof(share_info_t), 0);
-        mutex_initialize(&sh_info->lock);
+        mutex_initialize(&sh_info->lock, MUTEX_PASSIVE);
         sh_info->refcount = 2;
         btree_create(&sh_info->pagemap);
         src_area->sh_info = sh_info;
 ,10 → 746,11
  *
  * The address space area must be locked prior to this call.
  *
- * @param area Address space area.
- * @param access Access mode.
+ * @param area      Address space area.
+ * @param access    Access mode.
  *
- * @return False if access violates area's permissions, true otherwise.
+ * @return      False if access violates area's permissions, true
+ *          otherwise.
  */
 bool as_area_check_access(as_area_t *area, pf_access_t access)
 {
 ,21 → 766,183
     return true;
 }
+/** Change adress space area flags.
+ *
+ * The idea is to have the same data, but with a different access mode.
+ * This is needed e.g. for writing code into memory and then executing it.
+ * In order for this to work properly, this may copy the data
+ * into private anonymous memory (unless it's already there).
+ *
+ * @param as      Address space.
+ * @param flags   Flags of the area memory.
+ * @param address Address within the area to be changed.
+ *
+ * @return Zero on success or a value from @ref errno.h on failure.
+ *
+ */
+int as_area_change_flags(as_t *as, int flags, uintptr_t address)
+{
+    as_area_t *area;
+    uintptr_t base;
+    link_t *cur;
+    ipl_t ipl;
+    int page_flags;
+    uintptr_t *old_frame;
+    index_t frame_idx;
+    count_t used_pages;
+    /* Flags for the new memory mapping */
+    page_flags = area_flags_to_page_flags(flags);
+    ipl = interrupts_disable();
+    mutex_lock(&as->lock);
+    area = find_area_and_lock(as, address);
+    if (!area) {
+        mutex_unlock(&as->lock);
+        interrupts_restore(ipl);
+        return ENOENT;
+    }
+    if ((area->sh_info) || (area->backend != &anon_backend)) {
+        /* Copying shared areas not supported yet */
+        /* Copying non-anonymous memory not supported yet */
+        mutex_unlock(&area->lock);
+        mutex_unlock(&as->lock);
+        interrupts_restore(ipl);
+        return ENOTSUP;
+    }
+    base = area->base;
+    /*
+     * Compute total number of used pages in the used_space B+tree
+     */
+    used_pages = 0;
+    for (cur = area->used_space.leaf_head.next;
+        cur != &area->used_space.leaf_head; cur = cur->next) {
+        btree_node_t *node;
+        unsigned int i;
+        node = list_get_instance(cur, btree_node_t, leaf_link);
+        for (i = 0; i < node->keys; i++) {
+            used_pages += (count_t) node->value[i];
+        }
+    }
+    /* An array for storing frame numbers */
+    old_frame = malloc(used_pages * sizeof(uintptr_t), 0);
+    /*
+     * Start TLB shootdown sequence.
+     */
+    tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base, area->pages);
+    /*
+     * Remove used pages from page tables and remember their frame
+     * numbers.
+     */
+    frame_idx = 0;
+    for (cur = area->used_space.leaf_head.next;
+        cur != &area->used_space.leaf_head; cur = cur->next) {
+        btree_node_t *node;
+        unsigned int i;
+        node = list_get_instance(cur, btree_node_t, leaf_link);
+        for (i = 0; i < node->keys; i++) {
+            uintptr_t b = node->key[i];
+            count_t j;
+            pte_t *pte;
+            for (j = 0; j < (count_t) node->value[i]; j++) {
+                page_table_lock(as, false);
+                pte = page_mapping_find(as, b + j * PAGE_SIZE);
+                ASSERT(pte && PTE_VALID(pte) &&
+                    PTE_PRESENT(pte));
+                old_frame[frame_idx++] = PTE_GET_FRAME(pte);
+                /* Remove old mapping */
+                page_mapping_remove(as, b + j * PAGE_SIZE);
+                page_table_unlock(as, false);
+            }
+        }
+    }
+    /*
+     * Finish TLB shootdown sequence.
+     */
+    tlb_invalidate_pages(as->asid, area->base, area->pages);
+    /*
+     * Invalidate potential software translation caches (e.g. TSB on
+     * sparc64).
+     */
+    as_invalidate_translation_cache(as, area->base, area->pages);
+    tlb_shootdown_finalize();
+    /*
+     * Set the new flags.
+     */
+    area->flags = flags;
+    /*
+     * Map pages back in with new flags. This step is kept separate
+     * so that the memory area could not be accesed with both the old and
+     * the new flags at once.
+     */
+    frame_idx = 0;
+    for (cur = area->used_space.leaf_head.next;
+        cur != &area->used_space.leaf_head; cur = cur->next) {
+        btree_node_t *node;
+        unsigned int i;
+        node = list_get_instance(cur, btree_node_t, leaf_link);
+        for (i = 0; i < node->keys; i++) {
+            uintptr_t b = node->key[i];
+            count_t j;
+            for (j = 0; j < (count_t) node->value[i]; j++) {
+                page_table_lock(as, false);
+                /* Insert the new mapping */
+                page_mapping_insert(as, b + j * PAGE_SIZE,
+                    old_frame[frame_idx++], page_flags);
+                page_table_unlock(as, false);
+            }
+        }
+    }
+    free(old_frame);
+    mutex_unlock(&area->lock);
+    mutex_unlock(&as->lock);
+    interrupts_restore(ipl);
+    return 0;
+}
 /** Handle page fault within the current address space.
  *
- * This is the high-level page fault handler. It decides
- * whether the page fault can be resolved by any backend
- * and if so, it invokes the backend to resolve the page
- * fault.
+ * This is the high-level page fault handler. It decides whether the page fault
+ * can be resolved by any backend and if so, it invokes the backend to resolve
+ * the page fault.
  *
  * Interrupts are assumed disabled.
  *
- * @param page Faulting page.
- * @param access Access mode that caused the fault (i.e. read/write/exec).
- * @param istate Pointer to interrupted state.
+ * @param page      Faulting page.
+ * @param access    Access mode that caused the page fault (i.e.
+ *          read/write/exec).
+ * @param istate    Pointer to the interrupted state.
  *
- * @return AS_PF_FAULT on page fault, AS_PF_OK on success or AS_PF_DEFER if the
- *     fault was caused by copy_to_uspace() or copy_from_uspace().
+ * @return      AS_PF_FAULT on page fault, AS_PF_OK on success or
+ *          AS_PF_DEFER if the fault was caused by copy_to_uspace()
+ *          or copy_from_uspace().
  */
 int as_page_fault(uintptr_t page, pf_access_t access, istate_t *istate)
 {
 ,9 → 988,8
     page_table_lock(AS, false);
     /*
-     * To avoid race condition between two page faults
-     * on the same address, we need to make sure
-     * the mapping has not been already inserted.
+     * To avoid race condition between two page faults on the same address,
+     * we need to make sure the mapping has not been already inserted.
      */
     if ((pte = page_mapping_find(AS, page))) {
         if (PTE_PRESENT(pte)) {
 ,8 → 1043,8
  *
  * When this function is enetered, no spinlocks may be held.
  *
- * @param old Old address space or NULL.
- * @param new New address space.
+ * @param old       Old address space or NULL.
+ * @param new       New address space.
  */
 void as_switch(as_t *old_as, as_t *new_as)
 {
 ,9 → 1115,9
 /** Convert address space area flags to page flags.
  *
- * @param aflags Flags of some address space area.
+ * @param aflags    Flags of some address space area.
  *
- * @return Flags to be passed to page_mapping_insert().
+ * @return      Flags to be passed to page_mapping_insert().
  */
 int area_flags_to_page_flags(int aflags)
 {
 ,9 → 1145,9
  * The address space area must be locked.
  * Interrupts must be disabled.
  *
- * @param a Address space area.
+ * @param a     Address space area.
  *
- * @return Flags to be used in page_mapping_insert().
+ * @return      Flags to be used in page_mapping_insert().
  */
 int as_area_get_flags(as_area_t *a)
 {
 ,23 → 1156,20
 /** Create page table.
  *
- * Depending on architecture, create either address space
- * private or global page table.
+ * Depending on architecture, create either address space private or global page
+ * table.
  *
- * @param flags Flags saying whether the page table is for kernel address space.
+ * @param flags     Flags saying whether the page table is for the kernel
+ *          address space.
  *
- * @return First entry of the page table.
+ * @return      First entry of the page table.
  */
 pte_t *page_table_create(int flags)
 {
-#ifdef __OBJC__
-    return [as_t page_table_create: flags];
-#else
     ASSERT(as_operations);
     ASSERT(as_operations->page_table_create);
     return as_operations->page_table_create(flags);
-#endif
 }
 /** Destroy page table.
 ,18 → 1176,14
  *
  * Destroy page table in architecture specific way.
  *
- * @param page_table Physical address of PTL0.
+ * @param page_table    Physical address of PTL0.
  */
 void page_table_destroy(pte_t *page_table)
 {
-#ifdef __OBJC__
-    return [as_t page_table_destroy: page_table];
-#else
     ASSERT(as_operations);
     ASSERT(as_operations->page_table_destroy);
     as_operations->page_table_destroy(page_table);
-#endif
 }
 /** Lock page table.
 ,36 → 1195,28
  * prior to this call. Address space can be locked prior to this
  * call in which case the lock argument is false.
  *
- * @param as Address space.
- * @param lock If false, do not attempt to lock as->lock.
+ * @param as        Address space.
+ * @param lock      If false, do not attempt to lock as->lock.
  */
 void page_table_lock(as_t *as, bool lock)
 {
-#ifdef __OBJC__
-    [as page_table_lock: lock];
-#else
     ASSERT(as_operations);
     ASSERT(as_operations->page_table_lock);
     as_operations->page_table_lock(as, lock);
-#endif
 }
 /** Unlock page table.
  *
- * @param as Address space.
- * @param unlock If false, do not attempt to unlock as->lock.
+ * @param as        Address space.
+ * @param unlock    If false, do not attempt to unlock as->lock.
  */
 void page_table_unlock(as_t *as, bool unlock)
 {
-#ifdef __OBJC__
-    [as page_table_unlock: unlock];
-#else
     ASSERT(as_operations);
     ASSERT(as_operations->page_table_unlock);
     as_operations->page_table_unlock(as, unlock);
-#endif
 }
 ,11 → 1224,11
  *
  * The address space must be locked and interrupts must be disabled.
  *
- * @param as Address space.
- * @param va Virtual address.
+ * @param as        Address space.
+ * @param va        Virtual address.
  *
- * @return Locked address space area containing va on success or NULL on
- *     failure.
+ * @return      Locked address space area containing va on success or
+ *          NULL on failure.
  */
 as_area_t *find_area_and_lock(as_t *as, uintptr_t va)
 {
 ,15 → 1280,15
  *
  * The address space must be locked and interrupts must be disabled.
  *
- * @param as Address space.
- * @param va Starting virtual address of the area being tested.
- * @param size Size of the area being tested.
- * @param avoid_area Do not touch this area.
+ * @param as        Address space.
+ * @param va        Starting virtual address of the area being tested.
+ * @param size      Size of the area being tested.
+ * @param avoid_area    Do not touch this area.
  *
- * @return True if there is no conflict, false otherwise.
+ * @return      True if there is no conflict, false otherwise.
  */
-bool check_area_conflicts(as_t *as, uintptr_t va, size_t size,
-              as_area_t *avoid_area)
+bool
+check_area_conflicts(as_t *as, uintptr_t va, size_t size, as_area_t *avoid_area)
 {
     as_area_t *a;
     btree_node_t *leaf, *node;
 ,7 → 1377,7
     ipl = interrupts_disable();
     src_area = find_area_and_lock(AS, base);
-    if (src_area){
+    if (src_area) {
         size = src_area->pages * PAGE_SIZE;
         mutex_unlock(&src_area->lock);
     } else {
 ,11 → 1391,11
  *
  * The address space area must be already locked.
  *
- * @param a Address space area.
- * @param page First page to be marked.
- * @param count Number of page to be marked.
+ * @param a     Address space area.
+ * @param page      First page to be marked.
+ * @param count     Number of page to be marked.
  *
- * @return 0 on failure and 1 on success.
+ * @return      Zero on failure and non-zero on success.
  */
 int used_space_insert(as_area_t *a, uintptr_t page, count_t count)
 {
 ,8 → 1665,8
         }
     }
-    panic("Inconsistency detected while adding %d pages of used space at "
-        "%p.\n", count, page);
+    panic("Inconsistency detected while adding %" PRIc " pages of used "
+        "space at %p.", count, page);
 }
 /** Mark portion of address space area as unused.
 ,11 → 1673,11
  *
  * The address space area must be already locked.
  *
- * @param a Address space area.
- * @param page First page to be marked.
- * @param count Number of page to be marked.
+ * @param a     Address space area.
+ * @param page      First page to be marked.
+ * @param count     Number of page to be marked.
  *
- * @return 0 on failure and 1 on success.
+ * @return      Zero on failure and non-zero on success.
  */
 int used_space_remove(as_area_t *a, uintptr_t page, count_t count)
 {
 ,8 → 1844,8
     }
 error:
-    panic("Inconsistency detected while removing %d pages of used space "
-        "from %p.\n", count, page);
+    panic("Inconsistency detected while removing %" PRIc " pages of used "
+        "space from %p.", count, page);
 }
 /** Remove reference to address space area share info.
 ,7 → 1852,7
  *
  * If the reference count drops to 0, the sh_info is deallocated.
  *
- * @param sh_info Pointer to address space area share info.
+ * @param sh_info   Pointer to address space area share info.
  */
 void sh_info_remove_reference(share_info_t *sh_info)
 {
 ,6 → 1907,12
     return (unative_t) as_area_resize(AS, address, size, 0);
 }
+/** Wrapper for as_area_change_flags(). */
+unative_t sys_as_area_change_flags(uintptr_t address, int flags)
+{
+    return (unative_t) as_area_change_flags(AS, flags, address);
+}
 /** Wrapper for as_area_destroy(). */
 unative_t sys_as_area_destroy(uintptr_t address)
 {
 ,7 → 1921,7
 /** Print out information about address space.
  *
- * @param as Address space.
+ * @param as        Address space.
  */
 void as_print(as_t *as)
 {
 ,9 → 1943,9
             as_area_t *area = node->value[i];
             mutex_lock(&area->lock);
-            printf("as_area: %p, base=%p, pages=%d (%p - %p)\n",
-                area, area->base, area->pages, area->base,
-                area->base + area->pages*PAGE_SIZE);
+            printf("as_area: %p, base=%p, pages=%" PRIc
+                " (%p - %p)\n", area, area->base, area->pages,
+                area->base, area->base + FRAMES2SIZE(area->pages));
             mutex_unlock(&area->lock);
         }
     }

 /branches/dd/kernel/generic/src/mm/buddy.c
 ,8 → 35,7
  * @brief   Buddy allocator framework.
  *
  * This file contains buddy system allocator framework.
- * Specialized functions are needed for this abstract framework
- * to be useful.
+ * Specialized functions are needed for this abstract framework to be useful.
  */
 #include <mm/buddy.h>
 ,29 → 43,29
 #include <arch/types.h>
 #include <debug.h>
 #include <print.h>
+#include <macros.h>
-/** Return size needed for the buddy configuration data */
-size_t buddy_conf_size(int max_order)
+/** Return size needed for the buddy configuration data. */
+size_t buddy_conf_size(size_t max_order)
 {
     return sizeof(buddy_system_t) + (max_order + 1) * sizeof(link_t);
 }
-/** Create buddy system
+/** Create buddy system.
  *
  * Allocate memory for and initialize new buddy system.
  *
- * @param b Preallocated buddy system control data.
- * @param max_order The biggest allocable size will be 2^max_order.
- * @param op Operations for new buddy system.
- * @param data Pointer to be used by implementation.
+ * @param b     Preallocated buddy system control data.
+ * @param max_order The biggest allocable size will be 2^max_order.
+ * @param op        Operations for new buddy system.
+ * @param data      Pointer to be used by implementation.
  *
- * @return New buddy system.
+ * @return      New buddy system.
  */
-void buddy_system_create(buddy_system_t *b,
-             uint8_t max_order,
-             buddy_system_operations_t *op,
-             void *data)
+void
+buddy_system_create(buddy_system_t *b, uint8_t max_order,
+    buddy_system_operations_t *op, void *data)
 {
     int i;
 ,7 → 79,7
     ASSERT(op->mark_busy);
     /*
-     * Use memory after our own structure
+     * Use memory after our own structure.
      */
     b->order = (link_t *) (&b[1]);
 ,14 → 91,15
     b->data = data;
 }
-/** Check if buddy system can allocate block
+/** Check if buddy system can allocate block.
  *
- * @param b Buddy system pointer
- * @param i Size of the block (2^i)
+ * @param b     Buddy system pointer.
+ * @param i     Size of the block (2^i).
  *
- * @return True if block can be allocated
+ * @return      True if block can be allocated.
  */
-bool buddy_system_can_alloc(buddy_system_t *b, uint8_t i) {
+bool buddy_system_can_alloc(buddy_system_t *b, uint8_t i)
+{
     uint8_t k;
     /*
 ,12 → 106,13
      * If requested block is greater then maximal block
      * we know immediatly that we cannot satisfy the request.
      */
-    if (i > b->max_order) return false;
+    if (i > b->max_order)
+        return false;
     /*
      * Check if any bigger or equal order has free elements
      */
-    for (k=i; k <= b->max_order; k++) {
+    for (k = i; k <= b->max_order; k++) {
         if (!list_empty(&b->order[k])) {
             return true;
         }
 ,12 → 119,11
     }
     return false;
 }
-/** Allocate PARTICULAR block from buddy system
+/** Allocate PARTICULAR block from buddy system.
  *
- * @ return Block of data or NULL if no such block was found
+ * @return      Block of data or NULL if no such block was found.
  */
 link_t *buddy_system_alloc_block(buddy_system_t *b, link_t *block)
 {
 ,7 → 134,7
     ASSERT(left);
     list_remove(left);
     while (1) {
-        if (! b->op->get_order(b,left)) {
+        if (!b->op->get_order(b, left)) {
             b->op->mark_busy(b, left);
             return left;
         }
 ,8 → 142,8
         order = b->op->get_order(b, left);
         right = b->op->bisect(b, left);
-        b->op->set_order(b, left, order-1);
-        b->op->set_order(b, right, order-1);
+        b->op->set_order(b, left, order - 1);
+        b->op->set_order(b, right, order - 1);
         tmp = b->op->find_block(b, block, BUDDY_SYSTEM_INNER_BLOCK);
 ,10 → 160,10
 /** Allocate block from buddy system.
  *
- * @param b Buddy system pointer.
- * @param i Returned block will be 2^i big.
+ * @param b     Buddy system pointer.
+ * @param i     Returned block will be 2^i big.
  *
- * @return Block of data represented by link_t.
+ * @return      Block of data represented by link_t.
  */
 link_t *buddy_system_alloc(buddy_system_t *b, uint8_t i)
 {
 ,13 → 217,12
     buddy_system_free(b, hlp);
     return res;
 }
 /** Return block to buddy system.
  *
- * @param b Buddy system pointer.
- * @param block Block to return.
+ * @param b     Buddy system pointer.
+ * @param block     Block to return.
  */
 void buddy_system_free(buddy_system_t *b, link_t *block)
 {
 ,7 → 266,8
             b->op->set_order(b, hlp, i + 1);
             /*
-             * Recursively add the coalesced block to the list of order i + 1.
+             * Recursively add the coalesced block to the list of
+             * order i + 1.
              */
             buddy_system_free(b, hlp);
             return;
 ,46 → 278,7
      * Insert block into the list of order i.
      */
     list_append(block, &b->order[i]);
 }
-/** Prints out structure of buddy system
- *
- * @param b Pointer to buddy system
- * @param elem_size Element size
- */
-void buddy_system_structure_print(buddy_system_t *b, size_t elem_size) {
-    index_t i;
-    count_t cnt, elem_count = 0, block_count = 0;
-    link_t * cur;
-    printf("Order\tBlocks\tSize    \tBlock size\tElems per block\n");
-    printf("-----\t------\t--------\t----------\t---------------\n");
-    for (i=0;i <= b->max_order; i++) {
-        cnt = 0;
-        if (!list_empty(&b->order[i])) {
-            for (cur = b->order[i].next; cur != &b->order[i]; cur = cur->next)
-                cnt++;
-        }
-        printf("#%zd\t%5zd\t%7zdK\t%8zdK\t%6zd\t", i, cnt, (cnt * (1 << i) * elem_size) >> 10, ((1 << i) * elem_size) >> 10, 1 << i);
-        if (!list_empty(&b->order[i])) {
-            for (cur = b->order[i].next; cur != &b->order[i]; cur = cur->next) {
-                b->op->print_id(b, cur);
-                printf(" ");
-            }
-        }
-        printf("\n");
-        block_count += cnt;
-        elem_count += (1 << i) * cnt;
-    }
-    printf("-----\t------\t--------\t----------\t---------------\n");
-    printf("Buddy system contains %zd free elements (%zd blocks)\n" , elem_count, block_count);
-}
 /** @}
  */

 /branches/dd/kernel/generic/src/mm/backend_phys.c
 ,7 → 77,7
     page_mapping_insert(AS, addr, base + (addr - area->base),
         as_area_get_flags(area));
         if (!used_space_insert(area, ALIGN_DOWN(addr, PAGE_SIZE), 1))
-                panic("Could not insert used space.\n");
+                panic("Cannot insert used space.");
     return AS_PF_OK;
 }

 /branches/dd/kernel/generic/src/mm/frame.c
 ,6 → 1,7
 /*
  * Copyright (c) 2001-2005 Jakub Jermar
  * Copyright (c) 2005 Sergey Bondari
+ * Copyright (c) 2009 Martin Decky
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
 ,7 → 34,7
 /**
  * @file
- * @brief   Physical frame allocator.
+ * @brief Physical frame allocator.
  *
  * This file contains the physical frame allocator and memory zone management.
  * The frame allocator is built on top of the buddy allocator.
 ,16 → 42,6
  * @see buddy.c
  */
-/*
- * Locking order
- *
- * In order to access particular zone, the process must first lock
- * the zones.lock, then lock the zone and then unlock the zones.lock.
- * This insures, that we can fiddle with the zones in runtime without
- * affecting the processes.
- *
- */
 #include <arch/types.h>
 #include <mm/frame.h>
 #include <mm/as.h>
 ,7 → 48,8
 #include <panic.h>
 #include <debug.h>
 #include <adt/list.h>
-#include <synch/spinlock.h>
+#include <synch/mutex.h>
+#include <synch/condvar.h>
 #include <arch/asm.h>
 #include <arch.h>
 #include <print.h>
 ,43 → 59,17
 #include <macros.h>
 #include <config.h>
-typedef struct {
-    count_t refcount;   /**< tracking of shared frames  */
-    uint8_t buddy_order;    /**< buddy system block order */
-    link_t buddy_link;  /**< link to the next free block inside one
-                     order */
-    void *parent;           /**< If allocated by slab, this points there */
-} frame_t;
+zones_t zones;
-typedef struct {
-    SPINLOCK_DECLARE(lock); /**< this lock protects everything below */
-    pfn_t base;     /**< frame_no of the first frame in the frames
-                     array */
-    count_t count;          /**< Size of zone */
-    frame_t *frames;    /**< array of frame_t structures in this
-                     zone */
-    count_t free_count; /**< number of free frame_t structures */
-    count_t busy_count; /**< number of busy frame_t structures */
-    buddy_system_t *buddy_system; /**< buddy system for the zone */
-    int flags;
-} zone_t;
 /*
- * The zoneinfo.lock must be locked when accessing zoneinfo structure.
- * Some of the attributes in zone_t structures are 'read-only'
+ * Synchronization primitives used to sleep when there is no memory
+ * available.
  */
+mutex_t mem_avail_mtx;
+condvar_t mem_avail_cv;
+count_t mem_avail_req = 0;  /**< Number of frames requested. */
+count_t mem_avail_gen = 0;  /**< Generation counter. */
-typedef struct {
-    SPINLOCK_DECLARE(lock);
-    unsigned int count;
-    zone_t *info[ZONES_MAX];
-} zones_t;
-static zones_t zones;
 /********************/
 /* Helper functions */
 /********************/
 ,22 → 84,20
     return (index_t) (frame - zone->frames) + zone->base;
 }
-static inline int frame_index_valid(zone_t *zone, index_t index)
+static inline bool frame_index_valid(zone_t *zone, index_t index)
 {
     return (index < zone->count);
 }
-/** Compute pfn_t from frame_t pointer & zone pointer */
-static index_t make_frame_index(zone_t *zone, frame_t *frame)
+static inline index_t make_frame_index(zone_t *zone, frame_t *frame)
 {
     return (frame - zone->frames);
 }
-/** Initialize frame structure
+/** Initialize frame structure.
  *
- * Initialize frame structure.
+ * @param frame Frame structure to be initialized.
  *
- * @param frame Frame structure to be initialized.
  */
 static void frame_initialize(frame_t *frame)
 {
 ,142 → 105,145
     frame->buddy_order = 0;
 }
-/**********************/
-/* Zoneinfo functions */
-/**********************/
+/*******************/
+/* Zones functions */
+/*******************/
-/**
- * Insert-sort zone into zones list
+/** Insert-sort zone into zones list.
  *
- * @param newzone New zone to be inserted into zone list
- * @return zone number on success, -1 on error
+ * Assume interrupts are disabled and zones lock is
+ * locked.
+ *
+ * @param base  Base frame of the newly inserted zone.
+ * @param count Number of frames of the newly inserted zone.
+ *
+ * @return Zone number on success, -1 on error.
+ *
  */
-static int zones_add_zone(zone_t *newzone)
+static count_t zones_insert_zone(pfn_t base, count_t count)
 {
-    unsigned int i, j;
-    ipl_t ipl;
-    zone_t *z;
-    ipl = interrupts_disable();
-    spinlock_lock(&zones.lock);
-    /* Try to merge */
     if (zones.count + 1 == ZONES_MAX) {
         printf("Maximum zone count %u exceeded!\n", ZONES_MAX);
-        spinlock_unlock(&zones.lock);
-        interrupts_restore(ipl);
-        return -1;
+        return (count_t) -1;
     }
+    count_t i;
     for (i = 0; i < zones.count; i++) {
-        /* Check for overflow */
-        z = zones.info[i];
-        if (overlaps(newzone->base, newzone->count, z->base, z->count)) {
+        /* Check for overlap */
+        if (overlaps(base, count,
+            zones.info[i].base, zones.info[i].count)) {
             printf("Zones overlap!\n");
-            return -1;
+            return (count_t) -1;
         }
-        if (newzone->base < z->base)
+        if (base < zones.info[i].base)
             break;
     }
     /* Move other zones up */
-    for (j = i; j < zones.count; j++)
-        zones.info[j + 1] = zones.info[j];
+    count_t j;
+    for (j = zones.count; j > i; j--) {
+        zones.info[j] = zones.info[j - 1];
+        zones.info[j].buddy_system->data =
+            (void *) &zones.info[j - 1];
+    }
-    zones.info[i] = newzone;
     zones.count++;
-    spinlock_unlock(&zones.lock);
-    interrupts_restore(ipl);
     return i;
 }
-/**
- * Try to find a zone where can we find the frame
+/** Get total available frames.
  *
- * Assume interrupts are disabled.
+ * Assume interrupts are disabled and zones lock is
+ * locked.
  *
- * @param frame Frame number contained in zone
- * @param pzone If not null, it is used as zone hint. Zone index
- *              is filled into the variable on success.
- * @return Pointer to locked zone containing frame
+ * @return Total number of available frames.
+ *
  */
-static zone_t * find_zone_and_lock(pfn_t frame, unsigned int *pzone)
+#ifdef CONFIG_DEBUG
+static count_t total_frames_free(void)
 {
-    unsigned int i;
-    unsigned int hint = pzone ? *pzone : 0;
-    zone_t *z;
+    count_t total = 0;
+    count_t i;
+    for (i = 0; i < zones.count; i++)
+        total += zones.info[i].free_count;
-    spinlock_lock(&zones.lock);
+    return total;
+}
+#endif
+/** Find a zone with a given frames.
+ *
+ * Assume interrupts are disabled and zones lock is
+ * locked.
+ *
+ * @param frame Frame number contained in zone.
+ * @param count Number of frames to look for.
+ * @param hint  Used as zone hint.
+ *
+ * @return Zone index or -1 if not found.
+ *
+ */
+count_t find_zone(pfn_t frame, count_t count, count_t hint)
+{
     if (hint >= zones.count)
         hint = 0;
-    i = hint;
+    count_t i = hint;
     do {
-        z = zones.info[i];
-        spinlock_lock(&z->lock);
-        if (z->base <= frame && z->base + z->count > frame) {
-            /* Unlock the global lock */
-            spinlock_unlock(&zones.lock);
-            if (pzone)
-                *pzone = i;
-            return z;
-        }
-        spinlock_unlock(&z->lock);
+        if ((zones.info[i].base <= frame)
+            && (zones.info[i].base + zones.info[i].count >= frame + count))
+            return i;
         i++;
         if (i >= zones.count)
             i = 0;
-    } while(i != hint);
-    spinlock_unlock(&zones.lock);
-    return NULL;
+    } while (i != hint);
+    return (count_t) -1;
 }
 /** @return True if zone can allocate specified order */
-static int zone_can_alloc(zone_t *z, uint8_t order)
+static bool zone_can_alloc(zone_t *zone, uint8_t order)
 {
-    return buddy_system_can_alloc(z->buddy_system, order);
+    return (zone_flags_available(zone->flags)
+        && buddy_system_can_alloc(zone->buddy_system, order));
 }
-/** Find and lock zone that can allocate order frames.
+/** Find a zone that can allocate order frames.
  *
- * Assume interrupts are disabled.
+ * Assume interrupts are disabled and zones lock is
+ * locked.
  *
- * @param order Size (2^order) of free space we are trying to find
- * @param pzone Pointer to preferred zone or NULL, on return contains zone
- *      number
+ * @param order Size (2^order) of free space we are trying to find.
+ * @param flags Required flags of the target zone.
+ * @param hind  Preferred zone.
+ *
  */
-static zone_t * find_free_zone_and_lock(uint8_t order, unsigned int *pzone)
+static count_t find_free_zone(uint8_t order, zone_flags_t flags, count_t hint)
 {
-    unsigned int i;
-    zone_t *z;
-    unsigned int hint = pzone ? *pzone : 0;
-    spinlock_lock(&zones.lock);
     if (hint >= zones.count)
         hint = 0;
-    i = hint;
+    count_t i = hint;
     do {
-        z = zones.info[i];
+        /*
+         * Check whether the zone meets the search criteria.
+         */
+        if ((zones.info[i].flags & flags) == flags) {
+            /*
+             * Check if the zone has 2^order frames area available.
+             */
+            if (zone_can_alloc(&zones.info[i], order))
+                return i;
+        }
-        spinlock_lock(&z->lock);
-        /* Check if the zone has 2^order frames area available  */
-        if (zone_can_alloc(z, order)) {
-            spinlock_unlock(&zones.lock);
-            if (pzone)
-                *pzone = i;
-            return z;
-        }
-        spinlock_unlock(&z->lock);
-        if (++i >= zones.count)
+        i++;
+        if (i >= zones.count)
             i = 0;
-    } while(i != hint);
-    spinlock_unlock(&zones.lock);
-    return NULL;
+    } while (i != hint);
+    return (count_t) -1;
 }
 /**************************/
 ,167 → 250,146
 /* Buddy system functions */
 /**************************/
-/** Buddy system find_block implementation
+/** Buddy system find_block implementation.
  *
  * Find block that is parent of current list.
  * That means go to lower addresses, until such block is found
  *
- * @param order - Order of parent must be different then this parameter!!
+ * @param order Order of parent must be different then this
+ *              parameter!!
+ *
  */
-static link_t *zone_buddy_find_block(buddy_system_t *b, link_t *child,
+static link_t *zone_buddy_find_block(buddy_system_t *buddy, link_t *child,
     uint8_t order)
 {
-    frame_t *frame;
-    zone_t *zone;
-    index_t index;
+    frame_t *frame = list_get_instance(child, frame_t, buddy_link);
+    zone_t *zone = (zone_t *) buddy->data;
-    frame = list_get_instance(child, frame_t, buddy_link);
-    zone = (zone_t *) b->data;
-    index = frame_index(zone, frame);
+    index_t index = frame_index(zone, frame);
     do {
-        if (zone->frames[index].buddy_order != order) {
+        if (zone->frames[index].buddy_order != order)
             return &zone->frames[index].buddy_link;
-        }
-    } while(index-- > 0);
+    } while (index-- > 0);
     return NULL;
 }
-static void zone_buddy_print_id(buddy_system_t *b, link_t *block)
-{
-    frame_t *frame;
-    zone_t *zone;
-    index_t index;
-    frame = list_get_instance(block, frame_t, buddy_link);
-    zone = (zone_t *) b->data;
-    index = frame_index(zone, frame);
-    printf("%zd", index);
-}
-/** Buddy system find_buddy implementation
+/** Buddy system find_buddy implementation.
  *
- * @param b Buddy system.
- * @param block Block for which buddy should be found
+ * @param buddy Buddy system.
+ * @param block Block for which buddy should be found.
  *
- * @return Buddy for given block if found
+ * @return Buddy for given block if found.
+ *
  */
-static link_t *zone_buddy_find_buddy(buddy_system_t *b, link_t *block)
+static link_t *zone_buddy_find_buddy(buddy_system_t *buddy, link_t *block)
 {
-    frame_t *frame;
-    zone_t *zone;
-    index_t index;
-    bool is_left, is_right;
-    frame = list_get_instance(block, frame_t, buddy_link);
-    zone = (zone_t *) b->data;
+    frame_t *frame = list_get_instance(block, frame_t, buddy_link);
+    zone_t *zone = (zone_t *) buddy->data;
     ASSERT(IS_BUDDY_ORDER_OK(frame_index_abs(zone, frame),
         frame->buddy_order));
-    is_left = IS_BUDDY_LEFT_BLOCK_ABS(zone, frame);
-    is_right = IS_BUDDY_RIGHT_BLOCK_ABS(zone, frame);
-    ASSERT(is_left ^ is_right);
+    bool is_left = IS_BUDDY_LEFT_BLOCK_ABS(zone, frame);
+    index_t index;
     if (is_left) {
-        index = (frame_index(zone, frame)) + (1 << frame->buddy_order);
-    } else {    /* if (is_right) */
-        index = (frame_index(zone, frame)) - (1 << frame->buddy_order);
+        index = (frame_index(zone, frame)) +
+            (1 << frame->buddy_order);
+    } else {    /* is_right */
+        index = (frame_index(zone, frame)) -
+            (1 << frame->buddy_order);
     }
     if (frame_index_valid(zone, index)) {
-        if (zone->frames[index].buddy_order == frame->buddy_order &&
-            zone->frames[index].refcount == 0) {
+        if ((zone->frames[index].buddy_order == frame->buddy_order) &&
+            (zone->frames[index].refcount == 0)) {
             return &zone->frames[index].buddy_link;
         }
     }
-    return NULL;
+    return NULL;
 }
-/** Buddy system bisect implementation
+/** Buddy system bisect implementation.
  *
- * @param b Buddy system.
- * @param block Block to bisect
+ * @param buddy Buddy system.
+ * @param block Block to bisect.
  *
- * @return right block
+ * @return Right block.
+ *
  */
-static link_t * zone_buddy_bisect(buddy_system_t *b, link_t *block) {
-    frame_t *frame_l, *frame_r;
-    frame_l = list_get_instance(block, frame_t, buddy_link);
-    frame_r = (frame_l + (1 << (frame_l->buddy_order - 1)));
+static link_t *zone_buddy_bisect(buddy_system_t *buddy, link_t *block)
+{
+    frame_t *frame_l = list_get_instance(block, frame_t, buddy_link);
+    frame_t *frame_r = (frame_l + (1 << (frame_l->buddy_order - 1)));
     return &frame_r->buddy_link;
 }
-/** Buddy system coalesce implementation
+/** Buddy system coalesce implementation.
  *
- * @param b Buddy system.
- * @param block_1 First block
- * @param block_2 First block's buddy
+ * @param buddy   Buddy system.
+ * @param block_1 First block.
+ * @param block_2 First block's buddy.
  *
- * @return Coalesced block (actually block that represents lower address)
+ * @return Coalesced block (actually block that represents lower
+ *         address).
+ *
  */
-static link_t *zone_buddy_coalesce(buddy_system_t *b, link_t *block_1,
-    link_t *block_2)
+static link_t *zone_buddy_coalesce(buddy_system_t *buddy, link_t *block_1,
+    link_t *block_2)
 {
-    frame_t *frame1, *frame2;
+    frame_t *frame1 = list_get_instance(block_1, frame_t, buddy_link);
+    frame_t *frame2 = list_get_instance(block_2, frame_t, buddy_link);
-    frame1 = list_get_instance(block_1, frame_t, buddy_link);
-    frame2 = list_get_instance(block_2, frame_t, buddy_link);
-    return frame1 < frame2 ? block_1 : block_2;
+    return ((frame1 < frame2) ? block_1 : block_2);
 }
-/** Buddy system set_order implementation
+/** Buddy system set_order implementation.
  *
- * @param b Buddy system.
- * @param block Buddy system block
- * @param order Order to set
+ * @param buddy Buddy system.
+ * @param block Buddy system block.
+ * @param order Order to set.
+ *
  */
-static void zone_buddy_set_order(buddy_system_t *b, link_t *block,
-    uint8_t order) {
-    frame_t *frame;
-    frame = list_get_instance(block, frame_t, buddy_link);
-    frame->buddy_order = order;
+static void zone_buddy_set_order(buddy_system_t *buddy, link_t *block,
+    uint8_t order)
+{
+    list_get_instance(block, frame_t, buddy_link)->buddy_order = order;
 }
-/** Buddy system get_order implementation
+/** Buddy system get_order implementation.
  *
- * @param b Buddy system.
- * @param block Buddy system block
+ * @param buddy Buddy system.
+ * @param block Buddy system block.
  *
- * @return Order of block
+ * @return Order of block.
+ *
  */
-static uint8_t zone_buddy_get_order(buddy_system_t *b, link_t *block) {
-    frame_t *frame;
-    frame = list_get_instance(block, frame_t, buddy_link);
-    return frame->buddy_order;
+static uint8_t zone_buddy_get_order(buddy_system_t *buddy, link_t *block)
+{
+    return list_get_instance(block, frame_t, buddy_link)->buddy_order;
 }
-/** Buddy system mark_busy implementation
+/** Buddy system mark_busy implementation.
  *
- * @param b Buddy system
- * @param block Buddy system block
+ * @param buddy Buddy system.
+ * @param block Buddy system block.
  *
  */
-static void zone_buddy_mark_busy(buddy_system_t *b, link_t * block) {
-    frame_t * frame;
-    frame = list_get_instance(block, frame_t, buddy_link);
-    frame->refcount = 1;
+static void zone_buddy_mark_busy(buddy_system_t *buddy, link_t * block)
+{
+    list_get_instance(block, frame_t, buddy_link)->refcount = 1;
 }
-/** Buddy system mark_available implementation
+/** Buddy system mark_available implementation.
  *
- * @param b Buddy system
- * @param block Buddy system block
- *
+ * @param buddy Buddy system.
+ * @param block Buddy system block.
  */
-static void zone_buddy_mark_available(buddy_system_t *b, link_t *block) {
-    frame_t *frame;
-    frame = list_get_instance(block, frame_t, buddy_link);
-    frame->refcount = 0;
+static void zone_buddy_mark_available(buddy_system_t *buddy, link_t *block)
+{
+    list_get_instance(block, frame_t, buddy_link)->refcount = 0;
 }
 static buddy_system_operations_t zone_buddy_system_operations = {
 ,8 → 400,7
     .get_order = zone_buddy_get_order,
     .mark_busy = zone_buddy_mark_busy,
     .mark_available = zone_buddy_mark_available,
-    .find_block = zone_buddy_find_block,
-    .print_id = zone_buddy_print_id
+    .find_block = zone_buddy_find_block
 };
 /******************/
 ,62 → 407,59
 /* Zone functions */
 /******************/
-/** Allocate frame in particular zone
+/** Allocate frame in particular zone.
  *
- * Assume zone is locked
+ * Assume zone is locked and is available for allocation.
  * Panics if allocation is impossible.
  *
  * @param zone  Zone to allocate from.
  * @param order Allocate exactly 2^order frames.
  *
- * @return Frame index in zone
+ * @return Frame index in zone.
  *
  */
 static pfn_t zone_frame_alloc(zone_t *zone, uint8_t order)
 {
-    pfn_t v;
-    link_t *tmp;
-    frame_t *frame;
+    ASSERT(zone_flags_available(zone->flags));
     /* Allocate frames from zone buddy system */
-    tmp = buddy_system_alloc(zone->buddy_system, order);
+    link_t *link = buddy_system_alloc(zone->buddy_system, order);
-    ASSERT(tmp);
+    ASSERT(link);
     /* Update zone information. */
     zone->free_count -= (1 << order);
     zone->busy_count += (1 << order);
     /* Frame will be actually a first frame of the block. */
-    frame = list_get_instance(tmp, frame_t, buddy_link);
+    frame_t *frame = list_get_instance(link, frame_t, buddy_link);
-    /* get frame address */
-    v = make_frame_index(zone, frame);
-    return v;
+    /* Get frame address */
+    return make_frame_index(zone, frame);
 }
-/** Free frame from zone
+/** Free frame from zone.
  *
- * Assume zone is locked
+ * Assume zone is locked and is available for deallocation.
  *
- * @param zone Pointer to zone from which the frame is to be freed
- * @param frame_idx Frame index relative to zone
+ * @param zone      Pointer to zone from which the frame is to be freed.
+ * @param frame_idx Frame index relative to zone.
+ *
  */
 static void zone_frame_free(zone_t *zone, index_t frame_idx)
 {
-    frame_t *frame;
-    uint8_t order;
-    frame = &zone->frames[frame_idx];
+    ASSERT(zone_flags_available(zone->flags));
-    /* remember frame order */
-    order = frame->buddy_order;
+    frame_t *frame = &zone->frames[frame_idx];
+    /* Remember frame order */
+    uint8_t order = frame->buddy_order;
     ASSERT(frame->refcount);
     if (!--frame->refcount) {
         buddy_system_free(zone->buddy_system, &frame->buddy_link);
         /* Update zone information. */
         zone->free_count += (1 << order);
         zone->busy_count -= (1 << order);
 ,521 → 466,643
     }
 }
-/** Return frame from zone */
-static frame_t * zone_get_frame(zone_t *zone, index_t frame_idx)
+/** Return frame from zone. */
+static frame_t *zone_get_frame(zone_t *zone, index_t frame_idx)
 {
     ASSERT(frame_idx < zone->count);
     return &zone->frames[frame_idx];
 }
-/** Mark frame in zone unavailable to allocation */
+/** Mark frame in zone unavailable to allocation. */
 static void zone_mark_unavailable(zone_t *zone, index_t frame_idx)
 {
-    frame_t *frame;
-    link_t *link;
-    frame = zone_get_frame(zone, frame_idx);
+    ASSERT(zone_flags_available(zone->flags));
+    frame_t *frame = zone_get_frame(zone, frame_idx);
     if (frame->refcount)
         return;
+    link_t *link __attribute__ ((unused));
     link = buddy_system_alloc_block(zone->buddy_system,
         &frame->buddy_link);
     ASSERT(link);
     zone->free_count--;
 }
-/**
- * Join 2 zones
+/** Merge two zones.
  *
- * Expect zone_t *z to point to space at least zone_conf_size large
+ * Expect buddy to point to space at least zone_conf_size large.
+ * Assume z1 & z2 are locked and compatible and zones lock is
+ * locked.
  *
- * Assume z1 & z2 are locked
+ * @param z1     First zone to merge.
+ * @param z2     Second zone to merge.
+ * @param old_z1 Original date of the first zone.
+ * @param buddy  Merged zone buddy.
  *
- * @param z Target zone structure pointer
- * @param z1 Zone to merge
- * @param z2 Zone to merge
  */
-static void _zone_merge(zone_t *z, zone_t *z1, zone_t *z2)
+static void zone_merge_internal(count_t z1, count_t z2, zone_t *old_z1, buddy_system_t *buddy)
 {
-    uint8_t max_order;
-    unsigned int i;
-    int z2idx;
-    pfn_t frame_idx;
-    frame_t *frame;
-    ASSERT(!overlaps(z1->base, z1->count, z2->base, z2->count));
-    ASSERT(z1->base < z2->base);
-    spinlock_initialize(&z->lock, "zone_lock");
-    z->base = z1->base;
-    z->count = z2->base + z2->count - z1->base;
-    z->flags = z1->flags & z2->flags;
-    z->free_count = z1->free_count + z2->free_count;
-    z->busy_count = z1->busy_count + z2->busy_count;
+    ASSERT(zone_flags_available(zones.info[z1].flags));
+    ASSERT(zone_flags_available(zones.info[z2].flags));
+    ASSERT(zones.info[z1].flags == zones.info[z2].flags);
+    ASSERT(zones.info[z1].base < zones.info[z2].base);
+    ASSERT(!overlaps(zones.info[z1].base, zones.info[z1].count,
+        zones.info[z2].base, zones.info[z2].count));
-    max_order = fnzb(z->count);
-    z->buddy_system = (buddy_system_t *) &z[1];
-    buddy_system_create(z->buddy_system, max_order,
-        &zone_buddy_system_operations, (void *) z);
-    z->frames = (frame_t *)((uint8_t *) z->buddy_system +
-        buddy_conf_size(max_order));
-    for (i = 0; i < z->count; i++) {
-        /* This marks all frames busy */
-        frame_initialize(&z->frames[i]);
-    }
+    /* Difference between zone bases */
+    pfn_t base_diff = zones.info[z2].base - zones.info[z1].base;
+    zones.info[z1].count = base_diff + zones.info[z2].count;
+    zones.info[z1].free_count += zones.info[z2].free_count;
+    zones.info[z1].busy_count += zones.info[z2].busy_count;
+    zones.info[z1].buddy_system = buddy;
+    uint8_t order = fnzb(zones.info[z1].count);
+    buddy_system_create(zones.info[z1].buddy_system, order,
+        &zone_buddy_system_operations, (void *) &zones.info[z1]);
+    zones.info[z1].frames =
+        (frame_t *) ((uint8_t *) zones.info[z1].buddy_system
+        + buddy_conf_size(order));
+    /* This marks all frames busy */
+    count_t i;
+    for (i = 0; i < zones.info[z1].count; i++)
+        frame_initialize(&zones.info[z1].frames[i]);
     /* Copy frames from both zones to preserve full frame orders,
-     * parents etc. Set all free frames with refcount=0 to 1, because
-     * we add all free frames to buddy allocator later again, clear
-     * order to 0. Don't set busy frames with refcount=0, as they
+     * parents etc. Set all free frames with refcount = 0 to 1, because
+     * we add all free frames to buddy allocator later again, clearing
+     * order to 0. Don't set busy frames with refcount = 0, as they
      * will not be reallocated during merge and it would make later
      * problems with allocation/free.
      */
-    for (i = 0; i < z1->count; i++)
-        z->frames[i] = z1->frames[i];
-    for (i = 0; i < z2->count; i++) {
-        z2idx = i + (z2->base - z1->base);
-        z->frames[z2idx] = z2->frames[i];
-    }
+    for (i = 0; i < old_z1->count; i++)
+        zones.info[z1].frames[i] = old_z1->frames[i];
+    for (i = 0; i < zones.info[z2].count; i++)
+        zones.info[z1].frames[base_diff + i]
+            = zones.info[z2].frames[i];
     i = 0;
-    while (i < z->count) {
-        if (z->frames[i].refcount) {
-            /* skip busy frames */
-            i += 1 << z->frames[i].buddy_order;
-        } else { /* Free frames, set refcount=1 */
-            /* All free frames have refcount=0, we need not
-             * to check the order */
-            z->frames[i].refcount = 1;
-            z->frames[i].buddy_order = 0;
+    while (i < zones.info[z1].count) {
+        if (zones.info[z1].frames[i].refcount) {
+            /* Skip busy frames */
+            i += 1 << zones.info[z1].frames[i].buddy_order;
+        } else {
+            /* Free frames, set refcount = 1
+             * (all free frames have refcount == 0, we need not
+             * to check the order)
+             */
+            zones.info[z1].frames[i].refcount = 1;
+            zones.info[z1].frames[i].buddy_order = 0;
             i++;
         }
     }
-    /* Add free blocks from the 2 original zones */
-    while (zone_can_alloc(z1, 0)) {
-        frame_idx = zone_frame_alloc(z1, 0);
-        frame = &z->frames[frame_idx];
+    /* Add free blocks from the original zone z1 */
+    while (zone_can_alloc(old_z1, 0)) {
+        /* Allocate from the original zone */
+        pfn_t frame_idx = zone_frame_alloc(old_z1, 0);
+        /* Free the frame from the merged zone */
+        frame_t *frame = &zones.info[z1].frames[frame_idx];
         frame->refcount = 0;
-        buddy_system_free(z->buddy_system, &frame->buddy_link);
+        buddy_system_free(zones.info[z1].buddy_system, &frame->buddy_link);
     }
-    while (zone_can_alloc(z2, 0)) {
-        frame_idx = zone_frame_alloc(z2, 0);
-        frame = &z->frames[frame_idx + (z2->base - z1->base)];
+    /* Add free blocks from the original zone z2 */
+    while (zone_can_alloc(&zones.info[z2], 0)) {
+        /* Allocate from the original zone */
+        pfn_t frame_idx = zone_frame_alloc(&zones.info[z2], 0);
+        /* Free the frame from the merged zone */
+        frame_t *frame = &zones.info[z1].frames[base_diff + frame_idx];
         frame->refcount = 0;
-        buddy_system_free(z->buddy_system, &frame->buddy_link);
+        buddy_system_free(zones.info[z1].buddy_system, &frame->buddy_link);
     }
 }
-/** Return old configuration frames into the zone
+/** Return old configuration frames into the zone.
  *
- * We have several cases
- * - the conf. data is outside of zone -> exit, shall we call frame_free??
- * - the conf. data was created by zone_create or
- *   updated with reduce_region -> free every frame
+ * We have two cases:
+ * - The configuration data is outside the zone
+ *   -> do nothing (perhaps call frame_free?)
+ * - The configuration data was created by zone_create
+ *   or updated by reduce_region -> free every frame
  *
- * @param newzone The actual zone where freeing should occur
- * @param oldzone Pointer to old zone configuration data that should
- *                be freed from new zone
+ * @param znum  The actual zone where freeing should occur.
+ * @param pfn   Old zone configuration frame.
+ * @param count Old zone frame count.
+ *
  */
-static void return_config_frames(zone_t *newzone, zone_t *oldzone)
+static void return_config_frames(count_t znum, pfn_t pfn, count_t count)
 {
-    pfn_t pfn;
-    frame_t *frame;
-    count_t cframes;
-    unsigned int i;
-    pfn = ADDR2PFN((uintptr_t)KA2PA(oldzone));
-    cframes = SIZE2FRAMES(zone_conf_size(oldzone->count));
+    ASSERT(zone_flags_available(zones.info[znum].flags));
-    if (pfn < newzone->base || pfn >= newzone->base + newzone->count)
+    count_t cframes = SIZE2FRAMES(zone_conf_size(count));
+    if ((pfn < zones.info[znum].base)
+        || (pfn >= zones.info[znum].base + zones.info[znum].count))
         return;
+    frame_t *frame __attribute__ ((unused));
-    frame = &newzone->frames[pfn - newzone->base];
+    frame = &zones.info[znum].frames[pfn - zones.info[znum].base];
     ASSERT(!frame->buddy_order);
+    count_t i;
     for (i = 0; i < cframes; i++) {
-        newzone->busy_count++;
-        zone_frame_free(newzone, pfn+i-newzone->base);
+        zones.info[znum].busy_count++;
+        zone_frame_free(&zones.info[znum],
+            pfn - zones.info[znum].base + i);
     }
 }
-/** Reduce allocated block to count of order 0 frames
+/** Reduce allocated block to count of order 0 frames.
  *
- * The allocated block need 2^order frames of space. Reduce all frames
- * in block to order 0 and free the unneeded frames. This means, that
- * when freeing the previously allocated block starting with frame_idx,
+ * The allocated block needs 2^order frames. Reduce all frames
+ * in the block to order 0 and free the unneeded frames. This means that
+ * when freeing the previously allocated block starting with frame_idx,
  * you have to free every frame.
  *
- * @param zone
- * @param frame_idx Index to block
- * @param count Allocated space in block
+ * @param znum      Zone.
+ * @param frame_idx Index the first frame of the block.
+ * @param count     Allocated frames in block.
+ *
  */
-static void zone_reduce_region(zone_t *zone, pfn_t frame_idx, count_t count)
+static void zone_reduce_region(count_t znum, pfn_t frame_idx, count_t count)
 {
-    count_t i;
-    uint8_t order;
-    frame_t *frame;
+    ASSERT(zone_flags_available(zones.info[znum].flags));
+    ASSERT(frame_idx + count < zones.info[znum].count);
-    ASSERT(frame_idx + count < zone->count);
-    order = zone->frames[frame_idx].buddy_order;
+    uint8_t order = zones.info[znum].frames[frame_idx].buddy_order;
     ASSERT((count_t) (1 << order) >= count);
     /* Reduce all blocks to order 0 */
+    count_t i;
     for (i = 0; i < (count_t) (1 << order); i++) {
-        frame = &zone->frames[i + frame_idx];
+        frame_t *frame = &zones.info[znum].frames[i + frame_idx];
         frame->buddy_order = 0;
         if (!frame->refcount)
             frame->refcount = 1;
         ASSERT(frame->refcount == 1);
     }
     /* Free unneeded frames */
-    for (i = count; i < (count_t) (1 << order); i++) {
-        zone_frame_free(zone, i + frame_idx);
-    }
+    for (i = count; i < (count_t) (1 << order); i++)
+        zone_frame_free(&zones.info[znum], i + frame_idx);
 }
-/** Merge zones z1 and z2
+/** Merge zones z1 and z2.
  *
- * - the zones must be 2 zones with no zone existing in between,
- *   which means that z2 = z1+1
+ * The merged zones must be 2 zones with no zone existing in between
+ * (which means that z2 = z1 + 1). Both zones must be available zones
+ * with the same flags.
  *
- * - When you create a new zone, the frame allocator configuration does
- *   not to be 2^order size. Once the allocator is running it is no longer
- *   possible, merged configuration data occupies more space :-/
+ * When you create a new zone, the frame allocator configuration does
+ * not to be 2^order size. Once the allocator is running it is no longer
+ * possible, merged configuration data occupies more space :-/
+ *
+ * The function uses
+ *
  */
-void zone_merge(unsigned int z1, unsigned int z2)
+bool zone_merge(count_t z1, count_t z2)
 {
-    ipl_t ipl;
-    zone_t *zone1, *zone2, *newzone;
-    unsigned int cframes;
-    uint8_t order;
-    unsigned int i;
-    pfn_t pfn;
-    ipl = interrupts_disable();
+    ipl_t ipl = interrupts_disable();
     spinlock_lock(&zones.lock);
-    if ((z1 >= zones.count) || (z2 >= zones.count))
+    bool ret = true;
+    /* We can join only 2 zones with none existing inbetween,
+     * the zones have to be available and with the same
+     * set of flags
+     */
+    if ((z1 >= zones.count) || (z2 >= zones.count)
+        || (z2 - z1 != 1)
+        || (!zone_flags_available(zones.info[z1].flags))
+        || (!zone_flags_available(zones.info[z2].flags))
+        || (zones.info[z1].flags != zones.info[z2].flags)) {
+        ret = false;
         goto errout;
-    /* We can join only 2 zones with none existing inbetween */
-    if (z2-z1 != 1)
-        goto errout;
-    zone1 = zones.info[z1];
-    zone2 = zones.info[z2];
-    spinlock_lock(&zone1->lock);
-    spinlock_lock(&zone2->lock);
-    cframes = SIZE2FRAMES(zone_conf_size(zone2->base + zone2->count -
-        zone1->base));
+    }
+    pfn_t cframes = SIZE2FRAMES(zone_conf_size(
+        zones.info[z2].base - zones.info[z1].base
+        + zones.info[z2].count));
+    uint8_t order;
     if (cframes == 1)
         order = 0;
-    else
+    else
         order = fnzb(cframes - 1) + 1;
-    /* Allocate zonedata inside one of the zones */
-    if (zone_can_alloc(zone1, order))
-        pfn = zone1->base + zone_frame_alloc(zone1, order);
-    else if (zone_can_alloc(zone2, order))
-        pfn = zone2->base + zone_frame_alloc(zone2, order);
-    else
-        goto errout2;
-    newzone = (zone_t *) PA2KA(PFN2ADDR(pfn));
-    _zone_merge(newzone, zone1, zone2);
+    /* Allocate merged zone data inside one of the zones */
+    pfn_t pfn;
+    if (zone_can_alloc(&zones.info[z1], order)) {
+        pfn = zones.info[z1].base + zone_frame_alloc(&zones.info[z1], order);
+    } else if (zone_can_alloc(&zones.info[z2], order)) {
+        pfn = zones.info[z2].base + zone_frame_alloc(&zones.info[z2], order);
+    } else {
+        ret = false;
+        goto errout;
+    }
+    /* Preserve original data from z1 */
+    zone_t old_z1 = zones.info[z1];
+    old_z1.buddy_system->data = (void *) &old_z1;
+    /* Do zone merging */
+    buddy_system_t *buddy = (buddy_system_t *) PA2KA(PFN2ADDR(pfn));
+    zone_merge_internal(z1, z2, &old_z1, buddy);
     /* Free unneeded config frames */
-    zone_reduce_region(newzone, pfn - newzone->base,  cframes);
+    zone_reduce_region(z1, pfn - zones.info[z1].base, cframes);
     /* Subtract zone information from busy frames */
-    newzone->busy_count -= cframes;
-    /* Replace existing zones in zoneinfo list */
-    zones.info[z1] = newzone;
-    for (i = z2 + 1; i < zones.count; i++)
+    zones.info[z1].busy_count -= cframes;
+    /* Free old zone information */
+    return_config_frames(z1,
+        ADDR2PFN(KA2PA((uintptr_t) old_z1.frames)), old_z1.count);
+    return_config_frames(z1,
+        ADDR2PFN(KA2PA((uintptr_t) zones.info[z2].frames)),
+        zones.info[z2].count);
+    /* Move zones down */
+    count_t i;
+    for (i = z2 + 1; i < zones.count; i++) {
         zones.info[i - 1] = zones.info[i];
+        zones.info[i - 1].buddy_system->data =
+            (void *) &zones.info[i - 1];
+    }
     zones.count--;
-    /* Free old zone information */
-    return_config_frames(newzone, zone1);
-    return_config_frames(newzone, zone2);
-errout2:
-    /* Nobody is allowed to enter to zone, so we are safe
-     * to touch the spinlocks last time */
-    spinlock_unlock(&zone1->lock);
-    spinlock_unlock(&zone2->lock);
 errout:
     spinlock_unlock(&zones.lock);
     interrupts_restore(ipl);
+    return ret;
 }
-/**
- * Merge all zones into one big zone
+/** Merge all mergeable zones into one big zone.
  *
- * It is reasonable to do this on systems whose bios reports parts in chunks,
- * so that we could have 1 zone (it's faster).
+ * It is reasonable to do this on systems where
+ * BIOS reports parts in chunks, so that we could
+ * have 1 zone (it's faster).
+ *
  */
 void zone_merge_all(void)
 {
-    int count = zones.count;
-    while (zones.count > 1 && --count) {
-        zone_merge(0,1);
-        break;
+    count_t i = 0;
+    while (i < zones.count) {
+        if (!zone_merge(i, i + 1))
+            i++;
     }
 }
-/** Create frame zone
+/** Create new frame zone.
  *
- * Create new frame zone.
- *
+ * @param zone  Zone to construct.
+ * @param buddy Address of buddy system configuration information.
  * @param start Physical address of the first frame within the zone.
- * @param count Count of frames in zone
- * @param z Address of configuration information of zone
+ * @param count Count of frames in zone.
  * @param flags Zone flags.
  *
  * @return Initialized zone.
+ *
  */
-static void zone_construct(pfn_t start, count_t count, zone_t *z, int flags)
+static void zone_construct(zone_t *zone, buddy_system_t *buddy, pfn_t start, count_t count, zone_flags_t flags)
 {
-    unsigned int i;
-    uint8_t max_order;
-    spinlock_initialize(&z->lock, "zone_lock");
-    z->base = start;
-    z->count = count;
-    z->flags = flags;
-    z->free_count = count;
-    z->busy_count = 0;
-    /*
-     * Compute order for buddy system, initialize
-     */
-    max_order = fnzb(count);
-    z->buddy_system = (buddy_system_t *)&z[1];
+    zone->base = start;
+    zone->count = count;
+    zone->flags = flags;
+    zone->free_count = count;
+    zone->busy_count = 0;
+    zone->buddy_system = buddy;
-    buddy_system_create(z->buddy_system, max_order,
-                &zone_buddy_system_operations,
-                (void *) z);
-    /* Allocate frames _after_ the conframe */
-    /* Check sizes */
-    z->frames = (frame_t *)((uint8_t *) z->buddy_system +
-        buddy_conf_size(max_order));
-    for (i = 0; i < count; i++) {
-        frame_initialize(&z->frames[i]);
-    }
-    /* Stuffing frames */
-    for (i = 0; i < count; i++) {
-        z->frames[i].refcount = 0;
-        buddy_system_free(z->buddy_system, &z->frames[i].buddy_link);
-    }
+    if (zone_flags_available(flags)) {
+        /*
+         * Compute order for buddy system and initialize
+         */
+        uint8_t order = fnzb(count);
+        buddy_system_create(zone->buddy_system, order,
+            &zone_buddy_system_operations, (void *) zone);
+        /* Allocate frames _after_ the confframe */
+        /* Check sizes */
+        zone->frames = (frame_t *) ((uint8_t *) zone->buddy_system +
+            buddy_conf_size(order));
+        count_t i;
+        for (i = 0; i < count; i++)
+            frame_initialize(&zone->frames[i]);
+        /* Stuffing frames */
+        for (i = 0; i < count; i++) {
+            zone->frames[i].refcount = 0;
+            buddy_system_free(zone->buddy_system, &zone->frames[i].buddy_link);
+        }
+    } else
+        zone->frames = NULL;
 }
-/** Compute configuration data size for zone
+/** Compute configuration data size for zone.
  *
- * @param count Size of zone in frames
- * @return Size of zone configuration info (in bytes)
+ * @param count Size of zone in frames.
+ *
+ * @return Size of zone configuration info (in bytes).
+ *
  */
 uintptr_t zone_conf_size(count_t count)
 {
-    int size = sizeof(zone_t) + count*sizeof(frame_t);
-    int max_order;
-    max_order = fnzb(count);
-    size += buddy_conf_size(max_order);
-    return size;
+    return (count * sizeof(frame_t) + buddy_conf_size(fnzb(count)));
 }
-/** Create and add zone to system
+/** Create and add zone to system.
  *
- * @param start First frame number (absolute)
- * @param count Size of zone in frames
+ * @param start     First frame number (absolute).
+ * @param count     Size of zone in frames.
  * @param confframe Where configuration frames are supposed to be.
- *                  Automatically checks, that we will not disturb the
- *                  kernel and possibly init.
- *                  If confframe is given _outside_ this zone, it is expected,
- *                  that the area is already marked BUSY and big enough
- *                  to contain zone_conf_size() amount of data.
- *                  If the confframe is inside the area, the zone free frame
- *                  information is modified not to include it.
+ *                  Automatically checks, that we will not disturb the
+ *                  kernel and possibly init. If confframe is given
+ *                  _outside_ this zone, it is expected, that the area is
+ *                  already marked BUSY and big enough to contain
+ *                  zone_conf_size() amount of data. If the confframe is
+ *                  inside the area, the zone free frame information is
+ *                  modified not to include it.
  *
- * @return Zone number or -1 on error
+ * @return Zone number or -1 on error.
+ *
  */
-int zone_create(pfn_t start, count_t count, pfn_t confframe, int flags)
+count_t zone_create(pfn_t start, count_t count, pfn_t confframe, zone_flags_t flags)
 {
-    zone_t *z;
-    uintptr_t addr;
-    count_t confcount;
-    unsigned int i;
-    int znum;
-    /* Theoretically we could have here 0, practically make sure
-     * nobody tries to do that. If some platform requires, remove
-     * the assert
-     */
-    ASSERT(confframe);
-    /* If conframe is supposed to be inside our zone, then make sure
-     * it does not span kernel & init
-     */
-    confcount = SIZE2FRAMES(zone_conf_size(count));
-    if (confframe >= start && confframe < start+count) {
-        for (;confframe < start + count; confframe++) {
-            addr = PFN2ADDR(confframe);
-            if (overlaps(addr, PFN2ADDR(confcount),
-                KA2PA(config.base), config.kernel_size))
-                continue;
-            if (overlaps(addr, PFN2ADDR(confcount),
-                KA2PA(config.stack_base), config.stack_size))
-                continue;
-            bool overlap = false;
-            count_t i;
-            for (i = 0; i < init.cnt; i++)
+    ipl_t ipl = interrupts_disable();
+    spinlock_lock(&zones.lock);
+    if (zone_flags_available(flags)) {  /* Create available zone */
+        /* Theoretically we could have NULL here, practically make sure
+         * nobody tries to do that. If some platform requires, remove
+         * the assert
+         */
+        ASSERT(confframe != NULL);
+        /* If confframe is supposed to be inside our zone, then make sure
+         * it does not span kernel & init
+         */
+        count_t confcount = SIZE2FRAMES(zone_conf_size(count));
+        if ((confframe >= start) && (confframe < start + count)) {
+            for (; confframe < start + count; confframe++) {
+                uintptr_t addr = PFN2ADDR(confframe);
                 if (overlaps(addr, PFN2ADDR(confcount),
-                    KA2PA(init.tasks[i].addr),
-                    init.tasks[i].size)) {
-                    overlap = true;
-                    break;
-                }
-            if (overlap)
-                continue;
+                    KA2PA(config.base), config.kernel_size))
+                    continue;
+                if (overlaps(addr, PFN2ADDR(confcount),
+                    KA2PA(config.stack_base), config.stack_size))
+                    continue;
+                bool overlap = false;
+                count_t i;
+                for (i = 0; i < init.cnt; i++)
+                    if (overlaps(addr, PFN2ADDR(confcount),
+                        KA2PA(init.tasks[i].addr),
+                        init.tasks[i].size)) {
+                        overlap = true;
+                        break;
+                    }
+                if (overlap)
+                    continue;
+                break;
+            }
-            break;
+            if (confframe >= start + count)
+                panic("Cannot find configuration data for zone.");
         }
-        if (confframe >= start + count)
-            panic("Cannot find configuration data for zone.");
+        count_t znum = zones_insert_zone(start, count);
+        if (znum == (count_t) -1) {
+            spinlock_unlock(&zones.lock);
+            interrupts_restore(ipl);
+            return (count_t) -1;
+        }
+        buddy_system_t *buddy = (buddy_system_t *) PA2KA(PFN2ADDR(confframe));
+        zone_construct(&zones.info[znum], buddy, start, count, flags);
+        /* If confdata in zone, mark as unavailable */
+        if ((confframe >= start) && (confframe < start + count)) {
+            count_t i;
+            for (i = confframe; i < confframe + confcount; i++)
+                zone_mark_unavailable(&zones.info[znum],
+                    i - zones.info[znum].base);
+        }
+        spinlock_unlock(&zones.lock);
+        interrupts_restore(ipl);
+        return znum;
     }
-    z = (zone_t *) PA2KA(PFN2ADDR(confframe));
-    zone_construct(start, count, z, flags);
-    znum = zones_add_zone(z);
-    if (znum == -1)
-        return -1;
-    /* If confdata in zone, mark as unavailable */
-    if (confframe >= start && confframe < start + count)
-        for (i = confframe; i < confframe + confcount; i++) {
-            zone_mark_unavailable(z, i - z->base);
-        }
+    /* Non-available zone */
+    count_t znum = zones_insert_zone(start, count);
+    if (znum == (count_t) -1) {
+        spinlock_unlock(&zones.lock);
+        interrupts_restore(ipl);
+        return (count_t) -1;
+    }
+    zone_construct(&zones.info[znum], NULL, start, count, flags);
+    spinlock_unlock(&zones.lock);
+    interrupts_restore(ipl);
     return znum;
 }
-/***************************************/
+/*******************/
 /* Frame functions */
+/*******************/
-/** Set parent of frame */
-void frame_set_parent(pfn_t pfn, void *data, unsigned int hint)
+/** Set parent of frame. */
+void frame_set_parent(pfn_t pfn, void *data, count_t hint)
 {
-    zone_t *zone = find_zone_and_lock(pfn, &hint);
-    ASSERT(zone);
-    zone_get_frame(zone, pfn-zone->base)->parent = data;
-    spinlock_unlock(&zone->lock);
+    ipl_t ipl = interrupts_disable();
+    spinlock_lock(&zones.lock);
+    count_t znum = find_zone(pfn, 1, hint);
+    ASSERT(znum != (count_t) -1);
+    zone_get_frame(&zones.info[znum],
+        pfn - zones.info[znum].base)->parent = data;
+    spinlock_unlock(&zones.lock);
+    interrupts_restore(ipl);
 }
-void *frame_get_parent(pfn_t pfn, unsigned int hint)
+void *frame_get_parent(pfn_t pfn, count_t hint)
 {
-    zone_t *zone = find_zone_and_lock(pfn, &hint);
-    void *res;
-    ASSERT(zone);
-    res = zone_get_frame(zone, pfn - zone->base)->parent;
+    ipl_t ipl = interrupts_disable();
+    spinlock_lock(&zones.lock);
-    spinlock_unlock(&zone->lock);
+    count_t znum = find_zone(pfn, 1, hint);
+    ASSERT(znum != (count_t) -1);
+    void *res = zone_get_frame(&zones.info[znum],
+        pfn - zones.info[znum].base)->parent;
+    spinlock_unlock(&zones.lock);
+    interrupts_restore(ipl);
     return res;
 }
 /** Allocate power-of-two frames of physical memory.
  *
- * @param order  Allocate exactly 2^order frames.
- * @param flags  Flags for host zone selection and address processing.
- * @param pzone  Preferred zone
+ * @param order Allocate exactly 2^order frames.
+ * @param flags Flags for host zone selection and address processing.
+ * @param pzone Preferred zone.
  *
  * @return Physical address of the allocated frame.
  *
  */
-void * frame_alloc_generic(uint8_t order, int flags, unsigned int *pzone)
+void *frame_alloc_generic(uint8_t order, frame_flags_t flags, count_t *pzone)
 {
+    count_t size = ((count_t) 1) << order;
     ipl_t ipl;
-    int freed;
-    pfn_t v;
-    zone_t *zone;
+    count_t hint = pzone ? (*pzone) : 0;
 loop:
     ipl = interrupts_disable();
+    spinlock_lock(&zones.lock);
     /*
      * First, find suitable frame zone.
      */
-    zone = find_free_zone_and_lock(order, pzone);
+    count_t znum = find_free_zone(order,
+        FRAME_TO_ZONE_FLAGS(flags), hint);
     /* If no memory, reclaim some slab memory,
        if it does not help, reclaim all */
-    if (!zone && !(flags & FRAME_NO_RECLAIM)) {
-        freed = slab_reclaim(0);
-        if (freed)
-            zone = find_free_zone_and_lock(order, pzone);
-        if (!zone) {
+    if ((znum == (count_t) -1) && (!(flags & FRAME_NO_RECLAIM))) {
+        count_t freed = slab_reclaim(0);
+        if (freed > 0)
+            znum = find_free_zone(order,
+                FRAME_TO_ZONE_FLAGS(flags), hint);
+        if (znum == (count_t) -1) {
             freed = slab_reclaim(SLAB_RECLAIM_ALL);
-            if (freed)
-                zone = find_free_zone_and_lock(order, pzone);
+            if (freed > 0)
+                znum = find_free_zone(order,
+                    FRAME_TO_ZONE_FLAGS(flags), hint);
         }
     }
-    if (!zone) {
+    if (znum == (count_t) -1) {
+        if (flags & FRAME_ATOMIC) {
+            spinlock_unlock(&zones.lock);
+            interrupts_restore(ipl);
+            return NULL;
+        }
+#ifdef CONFIG_DEBUG
+        count_t avail = total_frames_free();
+#endif
+        spinlock_unlock(&zones.lock);
+        interrupts_restore(ipl);
         /*
-         * TODO: Sleep until frames are available again.
+         * Sleep until some frames are available again.
          */
-        interrupts_restore(ipl);
-        if (flags & FRAME_ATOMIC)
-            return 0;
-        panic("Sleep not implemented.\n");
+#ifdef CONFIG_DEBUG
+        printf("Thread %" PRIu64 " waiting for %" PRIc " frames, "
+            "%" PRIc " available.\n", THREAD->tid, size, avail);
+#endif
+        mutex_lock(&mem_avail_mtx);
+        if (mem_avail_req > 0)
+            mem_avail_req = min(mem_avail_req, size);
+        else
+            mem_avail_req = size;
+        count_t gen = mem_avail_gen;
+        while (gen == mem_avail_gen)
+            condvar_wait(&mem_avail_cv, &mem_avail_mtx);
+        mutex_unlock(&mem_avail_mtx);
+#ifdef CONFIG_DEBUG
+        printf("Thread %" PRIu64 " woken up.\n", THREAD->tid);
+#endif
         goto loop;
     }
-    v = zone_frame_alloc(zone, order);
-    v += zone->base;
-    spinlock_unlock(&zone->lock);
+    pfn_t pfn = zone_frame_alloc(&zones.info[znum], order)
+        + zones.info[znum].base;
+    spinlock_unlock(&zones.lock);
     interrupts_restore(ipl);
+    if (pzone)
+        *pzone = znum;
     if (flags & FRAME_KA)
-        return (void *)PA2KA(PFN2ADDR(v));
-    return (void *)PFN2ADDR(v);
+        return (void *) PA2KA(PFN2ADDR(pfn));
+    return (void *) PFN2ADDR(pfn);
 }
 /** Free a frame.
  *
  * Find respective frame structure for supplied physical frame address.
- * Decrement frame reference count.
- * If it drops to zero, move the frame structure to free list.
+ * Decrement frame reference count. If it drops to zero, move the frame
+ * structure to free list.
  *
- * @param Frame Physical Address of of the frame to be freed.
+ * @param frame Physical Address of of the frame to be freed.
+ *
  */
 void frame_free(uintptr_t frame)
 {
-    ipl_t ipl;
-    zone_t *zone;
-    pfn_t pfn = ADDR2PFN(frame);
-    ipl = interrupts_disable();
+    ipl_t ipl = interrupts_disable();
+    spinlock_lock(&zones.lock);
     /*
      * First, find host frame zone for addr.
      */
-    zone = find_zone_and_lock(pfn,NULL);
-    ASSERT(zone);
+    pfn_t pfn = ADDR2PFN(frame);
+    count_t znum = find_zone(pfn, 1, NULL);
-    zone_frame_free(zone, pfn-zone->base);
+    ASSERT(znum != (count_t) -1);
-    spinlock_unlock(&zone->lock);
+    zone_frame_free(&zones.info[znum], pfn - zones.info[znum].base);
+    spinlock_unlock(&zones.lock);
     interrupts_restore(ipl);
+    /*
+     * Signal that some memory has been freed.
+     */
+    mutex_lock(&mem_avail_mtx);
+    if (mem_avail_req > 0)
+        mem_avail_req--;
+    if (mem_avail_req == 0) {
+        mem_avail_gen++;
+        condvar_broadcast(&mem_avail_cv);
+    }
+    mutex_unlock(&mem_avail_mtx);
 }
 /** Add reference to frame.
 ,55 → 1111,56
  * increment frame reference count.
  *
  * @param pfn Frame number of the frame to be freed.
+ *
  */
 void frame_reference_add(pfn_t pfn)
 {
-    ipl_t ipl;
-    zone_t *zone;
-    frame_t *frame;
-    ipl = interrupts_disable();
+    ipl_t ipl = interrupts_disable();
+    spinlock_lock(&zones.lock);
     /*
      * First, find host frame zone for addr.
      */
-    zone = find_zone_and_lock(pfn,NULL);
-    ASSERT(zone);
+    count_t znum = find_zone(pfn, 1, NULL);
-    frame = &zone->frames[pfn-zone->base];
-    frame->refcount++;
+    ASSERT(znum != (count_t) -1);
-    spinlock_unlock(&zone->lock);
+    zones.info[znum].frames[pfn - zones.info[znum].base].refcount++;
+    spinlock_unlock(&zones.lock);
     interrupts_restore(ipl);
 }
-/** Mark given range unavailable in frame zones */
+/** Mark given range unavailable in frame zones. */
 void frame_mark_unavailable(pfn_t start, count_t count)
 {
-    unsigned int i;
-    zone_t *zone;
-    unsigned int prefzone = 0;
+    ipl_t ipl = interrupts_disable();
+    spinlock_lock(&zones.lock);
+    count_t i;
     for (i = 0; i < count; i++) {
-        zone = find_zone_and_lock(start + i, &prefzone);
-        if (!zone) /* PFN not found */
+        count_t znum = find_zone(start + i, 1, 0);
+        if (znum == (count_t) -1)  /* PFN not found */
             continue;
-        zone_mark_unavailable(zone, start + i - zone->base);
-        spinlock_unlock(&zone->lock);
+        zone_mark_unavailable(&zones.info[znum],
+            start + i - zones.info[znum].base);
     }
+    spinlock_unlock(&zones.lock);
+    interrupts_restore(ipl);
 }
-/** Initialize physical memory management
- *
- * Initialize physical memory managemnt.
- */
+/** Initialize physical memory management. */
 void frame_init(void)
 {
     if (config.cpu_active == 1) {
         zones.count = 0;
         spinlock_initialize(&zones.lock, "zones.lock");
+        mutex_initialize(&mem_avail_mtx, MUTEX_ACTIVE);
+        condvar_initialize(&mem_avail_cv);
     }
     /* Tell the architecture to create some memory */
     frame_arch_init();
     if (config.cpu_active == 1) {
 ,36 → 1175,28
             frame_mark_unavailable(pfn,
                 SIZE2FRAMES(init.tasks[i].size));
         }
         if (ballocs.size)
             frame_mark_unavailable(ADDR2PFN(KA2PA(ballocs.base)),
                 SIZE2FRAMES(ballocs.size));
         /* Black list first frame, as allocating NULL would
-         * fail in some places */
+         * fail in some places
+         */
         frame_mark_unavailable(0, 1);
     }
 }
-/** Return total size of all zones
- *
- */
-uint64_t zone_total_size(void) {
-    zone_t *zone = NULL;
-    unsigned int i;
-    ipl_t ipl;
-    uint64_t total = 0;
-    ipl = interrupts_disable();
+/** Return total size of all zones. */
+uint64_t zone_total_size(void)
+{
+    ipl_t ipl = interrupts_disable();
     spinlock_lock(&zones.lock);
-    for (i = 0; i < zones.count; i++) {
-        zone = zones.info[i];
-        spinlock_lock(&zone->lock);
-        total += (uint64_t) FRAMES2SIZE(zone->count);
-        spinlock_unlock(&zone->lock);
-    }
+    uint64_t total = 0;
+    count_t i;
+    for (i = 0; i < zones.count; i++)
+        total += (uint64_t) FRAMES2SIZE(zones.info[i].count);
     spinlock_unlock(&zones.lock);
     interrupts_restore(ipl);
 ,86 → 1204,128
     return total;
 }
+/** Prints list of zones. */
+void zone_print_list(void)
+{
+#ifdef __32_BITS__
+    printf("#  base address frames       flags    free frames  busy frames\n");
+    printf("-- ------------ ------------ -------- ------------ ------------\n");
+#endif
-/** Prints list of zones
- *
- */
-void zone_print_list(void) {
-    zone_t *zone = NULL;
-    unsigned int i;
-    ipl_t ipl;
-    ipl = interrupts_disable();
-    spinlock_lock(&zones.lock);
+#ifdef __64_BITS__
+    printf("#  base address          frames      flags    free frames  busy frames\n");
+    printf("-- -------------------- ------------ -------- ------------ ------------\n");
+#endif
-    if (sizeof(void *) == 4) {
-        printf("#  base address free frames  busy frames\n");
-        printf("-- ------------ ------------ ------------\n");
-    } else {
-        printf("#  base address         free frames  busy frames\n");
-        printf("-- -------------------- ------------ ------------\n");
-    }
+    /*
+     * Because printing may require allocation of memory, we may not hold
+     * the frame allocator locks when printing zone statistics.  Therefore,
+     * we simply gather the statistics under the protection of the locks and
+     * print the statistics when the locks have been released.
+     *
+     * When someone adds/removes zones while we are printing the statistics,
+     * we may end up with inaccurate output (e.g. a zone being skipped from
+     * the listing).
+     */
-    for (i = 0; i < zones.count; i++) {
-        zone = zones.info[i];
-        spinlock_lock(&zone->lock);
+    count_t i;
+    for (i = 0;; i++) {
+        ipl_t ipl = interrupts_disable();
+        spinlock_lock(&zones.lock);
-        if (sizeof(void *) == 4)
-            printf("%-2d   %#10zx %12zd %12zd\n", i, PFN2ADDR(zone->base),
-                zone->free_count, zone->busy_count);
-        else
-            printf("%-2d   %#18zx %12zd %12zd\n", i, PFN2ADDR(zone->base),
-                zone->free_count, zone->busy_count);
+        if (i >= zones.count) {
+            spinlock_unlock(&zones.lock);
+            interrupts_restore(ipl);
+            break;
+        }
-        spinlock_unlock(&zone->lock);
+        uintptr_t base = PFN2ADDR(zones.info[i].base);
+        count_t count = zones.info[i].count;
+        zone_flags_t flags = zones.info[i].flags;
+        count_t free_count = zones.info[i].free_count;
+        count_t busy_count = zones.info[i].busy_count;
+        spinlock_unlock(&zones.lock);
+        interrupts_restore(ipl);
+        bool available = zone_flags_available(flags);
+        printf("%-2" PRIc, i);
+#ifdef __32_BITS__
+        printf("   %10p", base);
+#endif
+#ifdef __64_BITS__
+        printf("   %18p", base);
+#endif
+        printf(" %12" PRIc " %c%c%c      ", count,
+            available ? 'A' : ' ',
+            (flags & ZONE_RESERVED) ? 'R' : ' ',
+            (flags & ZONE_FIRMWARE) ? 'F' : ' ');
+        if (available)
+            printf("%12" PRIc " %12" PRIc,
+                free_count, busy_count);
+        printf("\n");
     }
-    spinlock_unlock(&zones.lock);
-    interrupts_restore(ipl);
 }
 /** Prints zone details.
  *
  * @param num Zone base address or zone number.
+ *
  */
-void zone_print_one(unsigned int num) {
-    zone_t *zone = NULL;
-    ipl_t ipl;
-    unsigned int i;
-    ipl = interrupts_disable();
+void zone_print_one(count_t num)
+{
+    ipl_t ipl = interrupts_disable();
     spinlock_lock(&zones.lock);
+    count_t znum = (count_t) -1;
+    count_t i;
     for (i = 0; i < zones.count; i++) {
-        if ((i == num) || (PFN2ADDR(zones.info[i]->base) == num)) {
-            zone = zones.info[i];
+        if ((i == num) || (PFN2ADDR(zones.info[i].base) == num)) {
+            znum = i;
             break;
         }
     }
-    if (!zone) {
+    if (znum == (count_t) -1) {
+        spinlock_unlock(&zones.lock);
+        interrupts_restore(ipl);
         printf("Zone not found.\n");
-        goto out;
+        return;
     }
-    spinlock_lock(&zone->lock);
-    printf("Memory zone information\n");
-    printf("Zone base address: %#.*p\n", sizeof(uintptr_t) * 2,
-        PFN2ADDR(zone->base));
-    printf("Zone size: %zd frames (%zd KB)\n", zone->count,
-        SIZE2KB(FRAMES2SIZE(zone->count)));
-    printf("Allocated space: %zd frames (%zd KB)\n", zone->busy_count,
-        SIZE2KB(FRAMES2SIZE(zone->busy_count)));
-    printf("Available space: %zd frames (%zd KB)\n", zone->free_count,
-        SIZE2KB(FRAMES2SIZE(zone->free_count)));
-    buddy_system_structure_print(zone->buddy_system, FRAME_SIZE);
-    spinlock_unlock(&zone->lock);
+    uintptr_t base = PFN2ADDR(zones.info[i].base);
+    zone_flags_t flags = zones.info[i].flags;
+    count_t count = zones.info[i].count;
+    count_t free_count = zones.info[i].free_count;
+    count_t busy_count = zones.info[i].busy_count;
-out:
     spinlock_unlock(&zones.lock);
     interrupts_restore(ipl);
+    bool available = zone_flags_available(flags);
+    printf("Zone number:       %" PRIc "\n", znum);
+    printf("Zone base address: %p\n", base);
+    printf("Zone size:         %" PRIc " frames (%" PRIs " KiB)\n", count,
+        SIZE2KB(FRAMES2SIZE(count)));
+    printf("Zone flags:        %c%c%c\n",
+        available ? 'A' : ' ',
+        (flags & ZONE_RESERVED) ? 'R' : ' ',
+        (flags & ZONE_FIRMWARE) ? 'F' : ' ');
+    if (available) {
+        printf("Allocated space:   %" PRIc " frames (%" PRIs " KiB)\n",
+            busy_count, SIZE2KB(FRAMES2SIZE(busy_count)));
+        printf("Available space:   %" PRIc " frames (%" PRIs " KiB)\n",
+            free_count, SIZE2KB(FRAMES2SIZE(free_count)));
+    }
 }
 /** @}
  */

 /branches/dd/kernel/generic/src/mm/page.c
 ,11 → 40,28
  * They however, define the single interface.
  */
+/*
+ * Note on memory prefetching and updating memory mappings, also described in:
+ * AMD x86-64 Architecture Programmer's Manual, Volume 2, System Programming,
+ * 7.2.1 Special Coherency Considerations.
+ *
+ * The processor which modifies a page table mapping can access prefetched data
+ * from the old mapping.  In order to prevent this, we place a memory barrier
+ * after a mapping is updated.
+ *
+ * We assume that the other processors are either not using the mapping yet
+ * (i.e. during the bootstrap) or are executing the TLB shootdown code.  While
+ * we don't care much about the former case, the processors in the latter case
+ * will do an implicit serialization by virtue of running the TLB shootdown
+ * interrupt handler.
+ */
 #include <mm/page.h>
 #include <arch/mm/page.h>
 #include <arch/mm/asid.h>
 #include <mm/as.h>
 #include <mm/frame.h>
+#include <arch/barrier.h>
 #include <arch/types.h>
 #include <arch/asm.h>
 #include <memstr.h>
 ,8 → 82,8
  * considering possible crossings
  * of page boundaries.
  *
- * @param s Address of the structure.
- * @param size Size of the structure.
+ * @param s     Address of the structure.
+ * @param size      Size of the structure.
  */
 void map_structure(uintptr_t s, size_t size)
 {
 ,8 → 93,11
     cnt = length / PAGE_SIZE + (length % PAGE_SIZE > 0);
     for (i = 0; i < cnt; i++)
-        page_mapping_insert(AS_KERNEL, s + i * PAGE_SIZE, s + i * PAGE_SIZE, PAGE_NOT_CACHEABLE | PAGE_WRITE);
+        page_mapping_insert(AS_KERNEL, s + i * PAGE_SIZE,
+            s + i * PAGE_SIZE, PAGE_NOT_CACHEABLE | PAGE_WRITE);
+    /* Repel prefetched accesses to the old mapping. */
+    memory_barrier();
 }
 /** Insert mapping of page to frame.
 ,10 → 107,11
  *
  * The page table must be locked and interrupts must be disabled.
  *
- * @param as Address space to wich page belongs.
- * @param page Virtual address of the page to be mapped.
- * @param frame Physical address of memory frame to which the mapping is done.
- * @param flags Flags to be used for mapping.
+ * @param as        Address space to wich page belongs.
+ * @param page      Virtual address of the page to be mapped.
+ * @param frame     Physical address of memory frame to which the mapping is
+ *          done.
+ * @param flags     Flags to be used for mapping.
  */
 void page_mapping_insert(as_t *as, uintptr_t page, uintptr_t frame, int flags)
 {
 ,6 → 119,9
     ASSERT(page_mapping_operations->mapping_insert);
     page_mapping_operations->mapping_insert(as, page, frame, flags);
+    /* Repel prefetched accesses to the old mapping. */
+    memory_barrier();
 }
 /** Remove mapping of page.
 ,8 → 132,8
  *
  * The page table must be locked and interrupts must be disabled.
  *
- * @param as Address space to wich page belongs.
- * @param page Virtual address of the page to be demapped.
+ * @param as        Address space to wich page belongs.
+ * @param page      Virtual address of the page to be demapped.
  */
 void page_mapping_remove(as_t *as, uintptr_t page)
 {
 ,6 → 141,9
     ASSERT(page_mapping_operations->mapping_remove);
     page_mapping_operations->mapping_remove(as, page);
+    /* Repel prefetched accesses to the old mapping. */
+    memory_barrier();
 }
 /** Find mapping for virtual page
 ,10 → 152,11
  *
  * The page table must be locked and interrupts must be disabled.
  *
- * @param as Address space to wich page belongs.
- * @param page Virtual page.
+ * @param as        Address space to wich page belongs.
+ * @param page      Virtual page.
  *
- * @return NULL if there is no such mapping; requested mapping otherwise.
+ * @return      NULL if there is no such mapping; requested mapping
+ *          otherwise.
  */
 pte_t *page_mapping_find(as_t *as, uintptr_t page)
 {

 /branches/dd/kernel/generic/src/mm/backend_elf.c
 ,6 → 48,7
 #include <memstr.h>
 #include <macros.h>
 #include <arch.h>
+#include <arch/barrier.h>
 #ifdef CONFIG_VIRT_IDX_DCACHE
 #include <arch/mm/cache.h>
 ,12 → 68,13
  *
  * The address space area and page tables must be already locked.
  *
- * @param area Pointer to the address space area.
- * @param addr Faulting virtual address.
- * @param access Access mode that caused the fault (i.e. read/write/exec).
+ * @param area      Pointer to the address space area.
+ * @param addr      Faulting virtual address.
+ * @param access    Access mode that caused the fault (i.e.
+ *          read/write/exec).
  *
- * @return AS_PF_FAULT on failure (i.e. page fault) or AS_PF_OK on success (i.e.
- *     serviced).
+ * @return      AS_PF_FAULT on failure (i.e. page fault) or AS_PF_OK
+ *          on success (i.e. serviced).
  */
 int elf_page_fault(as_area_t *area, uintptr_t addr, pf_access_t access)
 {
 ,7 → 118,7
              */
             for (i = 0; i < leaf->keys; i++) {
-                if (leaf->key[i] == page) {
+                if (leaf->key[i] == page - area->base) {
                     found = true;
                     break;
                 }
 ,7 → 129,7
             page_mapping_insert(AS, addr, frame,
                 as_area_get_flags(area));
             if (!used_space_insert(area, page, 1))
-                panic("Could not insert used space.\n");
+                panic("Cannot insert used space.");
             mutex_unlock(&area->sh_info->lock);
             return AS_PF_OK;
         }
 ,6 → 152,10
             frame = (uintptr_t)frame_alloc(ONE_FRAME, 0);
             memcpy((void *) PA2KA(frame),
                 (void *) (base + i * FRAME_SIZE), FRAME_SIZE);
+            if (entry->p_flags & PF_X) {
+                smc_coherence_block((void *) PA2KA(frame),
+                    FRAME_SIZE);
+            }
             dirty = true;
         } else {
             frame = KA2PA(base + i * FRAME_SIZE);
 ,7 → 168,7
          * and cleared.
          */
         frame = (uintptr_t)frame_alloc(ONE_FRAME, 0);
-        memsetb(PA2KA(frame), FRAME_SIZE, 0);
+        memsetb((void *) PA2KA(frame), FRAME_SIZE, 0);
         dirty = true;
     } else {
         size_t pad_lo, pad_hi;
 ,8 → 193,13
         memcpy((void *) (PA2KA(frame) + pad_lo),
             (void *) (base + i * FRAME_SIZE + pad_lo),
             FRAME_SIZE - pad_lo - pad_hi);
-        memsetb(PA2KA(frame), pad_lo, 0);
-        memsetb(PA2KA(frame) + FRAME_SIZE - pad_hi, pad_hi, 0);
+        if (entry->p_flags & PF_X) {
+            smc_coherence_block((void *) (PA2KA(frame) + pad_lo),
+                FRAME_SIZE - pad_lo - pad_hi);
+        }
+        memsetb((void *) PA2KA(frame), pad_lo, 0);
+        memsetb((void *) (PA2KA(frame) + FRAME_SIZE - pad_hi), pad_hi,
+);
         dirty = true;
     }
 ,7 → 214,7
     page_mapping_insert(AS, addr, frame, as_area_get_flags(area));
     if (!used_space_insert(area, page, 1))
-        panic("Could not insert used space.\n");
+        panic("Cannot insert used space.");
     return AS_PF_OK;
 }
 ,9 → 223,10
  *
  * The address space area and page tables must be already locked.
  *
- * @param area Pointer to the address space area.
- * @param page Page that is mapped to frame. Must be aligned to PAGE_SIZE.
- * @param frame Frame to be released.
+ * @param area      Pointer to the address space area.
+ * @param page      Page that is mapped to frame. Must be aligned to
+ *          PAGE_SIZE.
+ * @param frame     Frame to be released.
  *
  */
 void elf_frame_free(as_area_t *area, uintptr_t page, uintptr_t frame)
 ,7 → 269,7
  *
  * The address space and address space area must be locked prior to the call.
  *
- * @param area Address space area.
+ * @param area      Address space area.
  */
 void elf_share(as_area_t *area)
 {

Subversion Repositories HelenOS

Compare Revisions

Ignore whitespace Rev 4054 → Rev 4055