Brooklyn/userland/interface/khronos/common/khrn_client_cache.c

/*
Copyright (c) 2012, Broadcom Europe Ltd
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
    * Redistributions of source code must retain the above copyright
      notice, this list of conditions and the following disclaimer.
    * Redistributions in binary form must reproduce the above copyright
      notice, this list of conditions and the following disclaimer in the
      documentation and/or other materials provided with the distribution.
    * Neither the name of the copyright holder nor the
      names of its contributors may be used to endorse or promote products
      derived from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY
DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/

#include "interface/khronos/common/khrn_int_common.h"

#include "interface/khronos/common/khrn_client_cache.h"
#include "interface/khronos/common/khrn_client_platform.h"
#include "interface/khronos/common/khrn_client_rpc.h"

#include "interface/khronos/common/khrn_int_hash.h"
#include "interface/khronos/common/khrn_int_util.h"

#ifdef RPC_DIRECT
#include "interface/khronos/glxx/glxx_int_impl.h"
#endif

#include <assert.h>

#if defined(SIMPENROSE)
#include "tools/v3d/simpenrose/simpenrose.h"
#endif

static void link_insert(CACHE_LINK_T *link, CACHE_LINK_T *prev, CACHE_LINK_T *next)
{
   vcos_assert(prev->next == next);
   vcos_assert(next->prev == prev);

   link->prev = prev;
   link->next = next;
   prev->next = link;
   next->prev = link;
}

static void link_remove(CACHE_LINK_T *link)
{
   link->next->prev = link->prev;
   link->prev->next = link->next;
}

static void tree_init(uint8_t *tree, int depth)
{
   int i;

   tree[0] = depth + 1;

   for (i = 1; i < 1 << depth; i++)
      tree[i] = tree[i >> 1] - 1;
}

static int heap_avail(KHRN_CACHE_T *cache, int size)
{
   return cache->tree && cache->tree[1] >= size;
}

static int heap_alloc(KHRN_CACHE_T *cache, int size)
{
   int node, fixup;
   int i;

   assert(heap_avail(cache, size));

   node = 1;
   for (i = 0; i < cache->client_depth - size; i++) {
      node <<= 1;
      if (cache->tree[node + 1] >= size && (cache->tree[node] < size || cache->tree[node] > cache->tree[node + 1]))
         node++;
   }

   cache->tree[node] = 0;

   for (fixup = node; cache->tree[fixup ^ 1] < cache->tree [fixup >> 1]; fixup >>= 1)
      cache->tree[fixup >> 1] = _max(cache->tree[fixup], cache->tree[fixup ^ 1]);

   return node * (1 << (size - 1)) - (1 << (cache->client_depth - 1));
}

static void heap_free(KHRN_CACHE_T *cache, int block)
{
   int node = block + (1 << (cache->client_depth - 1));
   int reset = 1;

   while (cache->tree[node] > 0) {
      node >>= 1;
      reset++;
   }

   cache->tree[node] = reset;

   while (cache->tree[node] == cache->tree[node ^ 1]) {
      node >>= 1;
      cache->tree[node] = cache->tree[node] + 1;
   }

   while (cache->tree[node] > cache->tree[node >> 1]) {
      cache->tree[node >> 1] = cache->tree[node];
      node >>= 1;
   }
}

static uint32_t hash(const void *data, int len, int sig)
{
   int hash;

//   if (len > 256)     // TODO: turn this on later
//      len = 256;

   if (!((size_t)data & 3) && !(len & 3))
      hash = khrn_hashword((uint32_t *)data, len >> 2, 0);
   else
      hash = khrn_hashlittle(data, len, 0);

   return (hash & ~0xf) | sig;
}

int khrn_cache_init(KHRN_CACHE_T *cache)
{
   cache->tree = NULL;
   cache->data = NULL;

   cache->client_depth = 0;
   cache->server_depth = 0;

   cache->start.prev = NULL;
   cache->start.next = &cache->end;
   cache->end.prev = &cache->start;
   cache->end.next = NULL;

   return khrn_pointer_map_init(&cache->map, 64);
}

void khrn_cache_term(KHRN_CACHE_T *cache)
{
   khrn_platform_free(cache->tree);
   khrn_platform_free(cache->data);

   khrn_pointer_map_term(&cache->map);
}

static void send_create(CLIENT_THREAD_STATE_T *thread, int base)
{
   RPC_CALL1(glintCacheCreate_impl,
             thread, 
             GLINTCACHECREATE_ID,
             RPC_UINT(base));
}

static void send_delete(CLIENT_THREAD_STATE_T *thread, int base)
{
   RPC_CALL1(glintCacheDelete_impl,
             thread, 
             GLINTCACHEDELETE_ID,
             RPC_UINT(base));
}

static int send_grow(CLIENT_THREAD_STATE_T *thread)
{
   return RPC_BOOLEAN_RES(RPC_CALL0_RES(glintCacheGrow_impl,
                                        thread, 
                                        GLINTCACHEGROW_ID));
}

static void send_data(CLIENT_THREAD_STATE_T *thread, int base, const void *data, int len)
{
   int off = 0;

   while (len > 0) {
      int chunk = _min(len, MERGE_BUFFER_SIZE-CLIENT_MAKE_CURRENT_SIZE-12-8);

      RPC_CALL3_IN_CTRL(glintCacheData_impl,
                        thread, 
                        GLINTCACHEDATA_ID,
                        RPC_UINT(base + off),
                        RPC_SIZEI(chunk),
                        (char *)data + off,
                        chunk);

      off += chunk;
      len -= chunk;
   }
}

static void discard(CLIENT_THREAD_STATE_T *thread, KHRN_CACHE_T *cache, CACHE_ENTRY_T *entry)
{
   heap_free(cache, (int)((uint8_t *)entry - cache->data) >> CACHE_LOG2_BLOCK_SIZE);

   khrn_pointer_map_delete(&cache->map, entry->key);

   link_remove(&entry->link);

   send_delete(thread, (int)((uint8_t *)entry - cache->data));
}

static void *relocate(void *data, void *user)
{
   return (uint8_t *)data - ((uint8_t **)user)[0] + ((uint8_t **)user)[1];
}

static void callback(KHRN_POINTER_MAP_T *map, uint32_t key, void *value, void *user)
{
   CACHE_ENTRY_T *entry = (CACHE_ENTRY_T *)value;

   entry->link.prev = (CACHE_LINK_T *)relocate(entry->link.prev, user);
   entry->link.next = (CACHE_LINK_T *)relocate(entry->link.next, user);

   // Coverity has rightly pointed out that the allocations done in the next codeline can fail
   // verify will only assert the code in debug mode. Use in release mode stays the same as before...
   verify(khrn_pointer_map_insert(map, key, relocate(value, user)));
}

static int grow(CLIENT_THREAD_STATE_T *thread, KHRN_CACHE_T *cache)
{
   /*
      try to grow the server cache
   */

   uint8_t *tree;
   uint8_t *data;
   int i;

   if (cache->server_depth == cache->client_depth) {
      if (cache->server_depth < CACHE_MAX_DEPTH && send_grow(thread))
         cache->server_depth++;
      else
         return 0;
   }

   tree = (uint8_t *)khrn_platform_malloc(1 << (cache->client_depth + 1), "KHRN_CACHE_T.tree");
   data = (uint8_t *)khrn_platform_malloc(1 << (cache->client_depth + CACHE_LOG2_BLOCK_SIZE), "KHRN_CACHE_T.data");

   if (!tree || !data) {
      khrn_platform_free(tree);
      khrn_platform_free(data);
      return 0;
   }

   /*
      set up new tree structure
   */

   tree_init(tree, cache->client_depth + 1);

   if (cache->client_depth) {
      for (i = 1; i < 1 << cache->client_depth; i++)
         tree[i ^ 3 << _msb(i)] = cache->tree[i];

      tree[1] = tree[3] + (tree[2] == tree[3]);
   }

   /*
      relocate pointermap and linked list pointers
   */
   {
      uint8_t *user[2];
	  user[0] = cache->data;
	  user[1] = data;

      khrn_pointer_map_iterate(&cache->map, callback, user);

      cache->start.next->prev = &cache->start;
      if (cache->start.next != &cache->end)
         cache->start.next = (CACHE_LINK_T *)relocate(cache->start.next, user);

      cache->end.prev->next = &cache->end;
      if (cache->end.prev != &cache->start)
         cache->end.prev = (CACHE_LINK_T *)relocate(cache->end.prev, user);
   }

   /*
      set up new data block
   */

   if (cache->data)
      platform_memcpy(data, cache->data, 1 << (cache->client_depth + CACHE_LOG2_BLOCK_SIZE - 1));

   /*
      free old blocks, update structure
   */

   khrn_platform_free(cache->tree);
   khrn_platform_free(cache->data);

   cache->tree = tree;
   cache->data = data;

   cache->client_depth++;

   return 1;
}

#ifdef SIMPENROSE_RECORD_OUTPUT
static bool xxx_first = true;
#endif
int khrn_cache_lookup(CLIENT_THREAD_STATE_T *thread, KHRN_CACHE_T *cache, const void *data, int len, int sig)
{
   int key = hash(data, len, sig);

   CACHE_ENTRY_T *entry = (CACHE_ENTRY_T *)khrn_pointer_map_lookup(&cache->map, key);

#ifdef SIMPENROSE_RECORD_OUTPUT
   if (xxx_first)
   {
      /* Cannot grow cache while things are locked for recording. So grow it now as much as we think we'll need */
      uint32_t i;
      xxx_first = false;
      for (i = 0; i < 15; i++)
         grow(thread, cache);
   }
#endif

   if (entry && entry->len >= len && !memcmp(entry->data, data, len)) {
      /*
         move link to end of discard queue
      */

      link_remove(&entry->link);
      link_insert(&entry->link, cache->end.prev, &cache->end);
   } else {
      int size = _max(_msb(len + sizeof(CACHE_ENTRY_T) - 1) + 2 - CACHE_LOG2_BLOCK_SIZE, 1);
      int block;

      CACHE_LINK_T *link;

      if (entry)
         discard(thread, cache, entry);

      while (!heap_avail(cache, size) && grow(thread, cache));

      for (link = cache->start.next; link != &cache->end && !heap_avail(cache, size); link = link->next)
         discard(thread, cache, (CACHE_ENTRY_T *)link);

      if (!heap_avail(cache, size))
         return -1;

      block = heap_alloc(cache, size);

      entry = (CACHE_ENTRY_T *)(cache->data + (block << CACHE_LOG2_BLOCK_SIZE));
      entry->len = len;
      entry->key = key;
      platform_memcpy(entry->data, data, len);

      if (!khrn_pointer_map_insert(&cache->map, key, entry)) {
         heap_free(cache, block);
         return -1;
      }

      link_insert(&entry->link, cache->end.prev, &cache->end);

      send_create(thread, (int)((uint8_t *)entry - cache->data));
      send_data(thread, (int)(entry->data - cache->data), data, len);
   }

   return (int)((uint8_t *)entry - cache->data);
}

int khrn_cache_get_entries(KHRN_CACHE_T *cache)
{
   return cache->map.entries;
}