Merge pull request #281 from moneroexamples/devel

Devel

README.md

@@ -72,7 +72,6 @@ The key features of the Onion Monero Blockchain Explorer are:
  - can provide total amount of all miner fees,
  - decoding encrypted payment id,
  - decoding outputs and proving txs sent to sub-address.
- - listing RandomX code for each block
 
 
 ## Development branch
@@ -244,7 +243,6 @@ xmrblocks, Onion Monero Blockchain Explorer:
   -t [ --testnet ] [=arg(=1)] (=0)      use testnet blockchain
   -s [ --stagenet ] [=arg(=1)] (=0)     use stagenet blockchain
   --enable-pusher [=arg(=1)] (=0)       enable signed transaction pusher
-  --enable-randomx [=arg(=1)] (=0)      enable generation of randomx code
   --enable-mixin-details [=arg(=1)] (=0)
                                         enable mixin details for key images,
                                         e.g., timescale, mixin of mixins, in tx

main.cpp

@@ -105,6 +105,12 @@ main(int ac, const char* av[])
     bool enable_as_hex                {*enable_as_hex_opt};
     bool enable_emission_monitor      {*enable_emission_monitor_opt};
 
+    //temprorary disable randomx
+    if (enable_randomx == true) {
+        cout << "Support for randomx code is disabled due to issues with it"<< endl;
+        enable_randomx = false;
+    }
+
     // set  monero log output level
     uint32_t log_level = 0;
     mlog_configure("", true);

src/crypto/rx-slow-hash.c

@@ -1,4 +1,4 @@
-// Copyright (c) 2019, The Monero Project
+// Copyright (c) 2019-2022, The Monero Project
 //
 // All rights reserved.
 //
@@ -34,6 +34,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <unistd.h>
+#include <limits.h>
 
 #include "randomx.h"
 #include "c_threads.h"
@@ -42,31 +43,38 @@
 
 #define RX_LOGCAT	"randomx"
 
+static CTHR_RWLOCK_TYPE main_dataset_lock = CTHR_RWLOCK_INIT;
+static CTHR_RWLOCK_TYPE main_cache_lock = CTHR_RWLOCK_INIT;
+
+static randomx_dataset *main_dataset = NULL;
+static randomx_cache *main_cache = NULL;
+static char main_seedhash[HASH_SIZE];
+static int main_seedhash_set = 0;
+
+static CTHR_RWLOCK_TYPE secondary_cache_lock = CTHR_RWLOCK_INIT;
+
+static randomx_cache *secondary_cache = NULL;
+static char secondary_seedhash[HASH_SIZE];
+static int secondary_seedhash_set = 0;
+
 #if defined(_MSC_VER)
 #define THREADV __declspec(thread)
 #else
 #define THREADV __thread
 #endif
 
-typedef struct rx_state {
+THREADV randomx_vm *main_vm_full = NULL;
-  CTHR_MUTEX_TYPE rs_mutex;
+static THREADV randomx_vm *main_vm_light = NULL;
-  char rs_hash[32];
+static THREADV randomx_vm *secondary_vm_light = NULL;
-  uint64_t  rs_height;
-  randomx_cache *rs_cache;
-} rx_state;
-
-static CTHR_MUTEX_TYPE rx_mutex = CTHR_MUTEX_INIT;
-static CTHR_MUTEX_TYPE rx_dataset_mutex = CTHR_MUTEX_INIT;
 
-static rx_state rx_s[2] = {{CTHR_MUTEX_INIT,{0},0,0},{CTHR_MUTEX_INIT,{0},0,0}};
+static THREADV uint32_t miner_thread = 0;
 
-static randomx_dataset *rx_dataset;
+static bool is_main(const char* seedhash) { return main_seedhash_set && (memcmp(seedhash, main_seedhash, HASH_SIZE) == 0); }
-static uint64_t rx_dataset_height;
+static bool is_secondary(const char* seedhash) { return secondary_seedhash_set && (memcmp(seedhash, secondary_seedhash, HASH_SIZE) == 0); }
-THREADV randomx_vm *rx_vm = NULL;
-static THREADV int rx_toggle;
 
 static void local_abort(const char *msg)
 {
+  merror(RX_LOGCAT, "%s", msg);
   fprintf(stderr, "%s\n", msg);
 #ifdef NDEBUG
   _exit(1);
@@ -75,108 +83,196 @@ static void local_abort(const char *msg)
 #endif
 }
 
-/**
+static void hash2hex(const char* hash, char* hex) {
- * @brief uses cpuid to determine if the CPU supports the AES instructions
+  const char* d = "0123456789abcdef";
- * @return true if the CPU supports AES, false otherwise
+  for (int i = 0; i < HASH_SIZE; ++i) {
- */
+    const uint8_t b = hash[i];
+    hex[i * 2 + 0] = d[b >> 4];
+    hex[i * 2 + 1] = d[b & 15];
+  }
+  hex[HASH_SIZE * 2] = '\0';
+}
 
-static inline int force_software_aes(void)
+static inline int disabled_flags(void) {
-{
+  static int flags = -1;
-  static int use = -1;
 
-  if (use != -1)
+  if (flags != -1) {
-    return use;
+    return flags;
+  }
 
-  const char *env = getenv("MONERO_USE_SOFTWARE_AES");
+  const char *env = getenv("MONERO_RANDOMX_UMASK");
   if (!env) {
-    use = 0;
+    flags = 0;
   }
-  else if (!strcmp(env, "0") || !strcmp(env, "no")) {
+  else {
-    use = 0;
+    char* endptr;
+    long int value = strtol(env, &endptr, 0);
+    if (endptr != env && value >= 0 && value < INT_MAX) {
+      flags = value;
     }
     else {
-    use = 1;
+      flags = 0;
+    }
+  }
+
+  return flags;
 }
-  return use;
+
+static inline int enabled_flags(void) {
+  static int flags = -1;
+
+  if (flags != -1) {
+    return flags;
   }
 
-static void cpuid(int CPUInfo[4], int InfoType)
+  flags = randomx_get_flags();
+
+  return flags;
+}
+
+#define SEEDHASH_EPOCH_BLOCKS	2048	/* Must be same as BLOCKS_SYNCHRONIZING_MAX_COUNT in cryptonote_config.h */
+#define SEEDHASH_EPOCH_LAG		64
+
+static inline int is_power_of_2(uint64_t n) { return n && (n & (n-1)) == 0; }
+
+static int get_seedhash_epoch_lag(void)
 {
-#if defined(__x86_64__)
+  static unsigned int lag = (unsigned int)-1;
-    __asm __volatile__
+  if (lag != (unsigned int)-1)
-    (
+    return lag;
-    "cpuid":
+  const char *e = getenv("SEEDHASH_EPOCH_LAG");
-        "=a" (CPUInfo[0]),
+  if (e)
-        "=b" (CPUInfo[1]),
+  {
-        "=c" (CPUInfo[2]),
+    lag = atoi(e);
-        "=d" (CPUInfo[3]) :
+    if (lag > SEEDHASH_EPOCH_LAG || !is_power_of_2(lag))
-            "a" (InfoType), "c" (0)
+      lag = SEEDHASH_EPOCH_LAG;
-        );
-#endif
   }
-static inline int check_aes_hw(void)
+  else
   {
-#if defined(__x86_64__)
+    lag = SEEDHASH_EPOCH_LAG;
-    int cpuid_results[4];
+  }
-    static int supported = -1;
+  return lag;
+}
 
-    if(supported >= 0)
+static unsigned int get_seedhash_epoch_blocks(void)
-        return supported;
+{
+  static unsigned int blocks = (unsigned int)-1;
+  if (blocks != (unsigned int)-1)
+    return blocks;
+  const char *e = getenv("SEEDHASH_EPOCH_BLOCKS");
+  if (e)
+  {
+    blocks = atoi(e);
+    if (blocks < 2 || blocks > SEEDHASH_EPOCH_BLOCKS || !is_power_of_2(blocks))
+      blocks = SEEDHASH_EPOCH_BLOCKS;
+  }
+  else
+  {
+    blocks = SEEDHASH_EPOCH_BLOCKS;
+  }
+  return blocks;
+}
 
-    cpuid(cpuid_results,1);
+uint64_t me_rx_seedheight(const uint64_t height) {
-    return supported = cpuid_results[2] & (1 << 25);
+  const uint64_t seedhash_epoch_lag = get_seedhash_epoch_lag();
-#else
+  const uint64_t seedhash_epoch_blocks = get_seedhash_epoch_blocks();
-    return 0;
+  uint64_t s_height =  (height <= seedhash_epoch_blocks+seedhash_epoch_lag) ? 0 :
-#endif
+                       (height - seedhash_epoch_lag - 1) & ~(seedhash_epoch_blocks-1);
+  return s_height;
 }
 
-static volatile int use_rx_jit_flag = -1;
+void me_rx_seedheights(const uint64_t height, uint64_t *seedheight, uint64_t *nextheight) {
+  *seedheight = me_rx_seedheight(height);
+  *nextheight = me_rx_seedheight(height + get_seedhash_epoch_lag());
+}
 
-static inline int use_rx_jit(void)
+static void rx_alloc_dataset(randomx_flags flags, randomx_dataset** dataset, int ignore_env)
 {
-#if defined(__x86_64__)
+  if (*dataset) {
+    return;
+  }
 
-  if (use_rx_jit_flag != -1)
+  if (disabled_flags() & RANDOMX_FLAG_FULL_MEM) {
-    return use_rx_jit_flag;
+    static int shown = 0;
+    if (!shown) {
+      shown = 1;
+      minfo(RX_LOGCAT, "RandomX dataset is disabled by MONERO_RANDOMX_UMASK environment variable.");
+    }
+    return;
+  }
 
-  const char *env = getenv("MONERO_USE_RX_JIT");
+  if (!ignore_env && !getenv("MONERO_RANDOMX_FULL_MEM")) {
-  if (!env) {
+    static int shown = 0;
-    use_rx_jit_flag = 1;
+    if (!shown) {
+      shown = 1;
+      minfo(RX_LOGCAT, "RandomX dataset is not enabled by default. Use MONERO_RANDOMX_FULL_MEM environment variable to enable it.");
     }
-  else if (!strcmp(env, "0") || !strcmp(env, "no")) {
+    return;
-    use_rx_jit_flag = 0;
+  }
+
+  *dataset = randomx_alloc_dataset((flags | RANDOMX_FLAG_LARGE_PAGES) & ~disabled_flags());
+  if (!*dataset) {
+    mwarning(RX_LOGCAT, "Couldn't allocate RandomX dataset using large pages");
+    *dataset = randomx_alloc_dataset(flags & ~disabled_flags());
+    if (!*dataset) {
+      merror(RX_LOGCAT, "Couldn't allocate RandomX dataset");
     }
-  else {
-    use_rx_jit_flag = 1;
   }
-  return use_rx_jit_flag;
-#else
-  return 0;
-#endif
 }
 
-#define SEEDHASH_EPOCH_BLOCKS	2048	/* Must be same as BLOCKS_SYNCHRONIZING_MAX_COUNT in cryptonote_config.h */
+static void rx_alloc_cache(randomx_flags flags, randomx_cache** cache)
-#define SEEDHASH_EPOCH_LAG		64
+{
+  if (*cache) {
+    return;
+  }
 
-void me_rx_reorg(const uint64_t split_height) {
+  *cache = randomx_alloc_cache((flags | RANDOMX_FLAG_LARGE_PAGES) & ~disabled_flags());
-  int i;
+  if (!*cache) {
-  CTHR_MUTEX_LOCK(rx_mutex);
+    mwarning(RX_LOGCAT, "Couldn't allocate RandomX cache using large pages");
-  for (i=0; i<2; i++) {
+    *cache = randomx_alloc_cache(flags & ~disabled_flags());
-    if (split_height < rx_s[i].rs_height)
+    if (!*cache) local_abort("Couldn't allocate RandomX cache");
-      rx_s[i].rs_height = 1;	/* set to an invalid seed height */
   }
-  CTHR_MUTEX_UNLOCK(rx_mutex);
 }
 
-uint64_t me_rx_seedheight(const uint64_t height) {
+static void rx_init_full_vm(randomx_flags flags, randomx_vm** vm)
-  uint64_t s_height =  (height <= SEEDHASH_EPOCH_BLOCKS+SEEDHASH_EPOCH_LAG) ? 0 :
+{
-                       (height - SEEDHASH_EPOCH_LAG - 1) & ~(SEEDHASH_EPOCH_BLOCKS-1);
+  if (*vm || !main_dataset || (disabled_flags() & RANDOMX_FLAG_FULL_MEM)) {
-  return s_height;
+    return;
   }
 
-void me_rx_seedheights(const uint64_t height, uint64_t *seedheight, uint64_t *nextheight) {
+  if ((flags & RANDOMX_FLAG_JIT) && !miner_thread) {
-  *seedheight = me_rx_seedheight(height);
+    flags |= RANDOMX_FLAG_SECURE;
-  *nextheight = me_rx_seedheight(height + SEEDHASH_EPOCH_LAG);
+  }
+
+  *vm = randomx_create_vm((flags | RANDOMX_FLAG_LARGE_PAGES | RANDOMX_FLAG_FULL_MEM) & ~disabled_flags(), NULL, main_dataset);
+  if (!*vm) {
+    mwarning(RX_LOGCAT, "Couldn't allocate RandomX full VM using large pages");
+    *vm = randomx_create_vm((flags | RANDOMX_FLAG_FULL_MEM) & ~disabled_flags(), NULL, main_dataset);
+    if (!*vm) {
+      merror(RX_LOGCAT, "Couldn't allocate RandomX full VM");
+    }
+  }
+}
+
+static void rx_init_light_vm(randomx_flags flags, randomx_vm** vm, randomx_cache* cache)
+{
+  if (*vm) {
+    randomx_vm_set_cache(*vm, cache);
+    return;
+  }
+
+  if ((flags & RANDOMX_FLAG_JIT) && !miner_thread) {
+    flags |= RANDOMX_FLAG_SECURE;
+  }
+
+  flags &= ~RANDOMX_FLAG_FULL_MEM;
+
+  *vm = randomx_create_vm((flags | RANDOMX_FLAG_LARGE_PAGES) & ~disabled_flags(), cache, NULL);
+  if (!*vm) {
+    mwarning(RX_LOGCAT, "Couldn't allocate RandomX light VM using large pages");
+    *vm = randomx_create_vm(flags & ~disabled_flags(), cache, NULL);
+    if (!*vm) local_abort("Couldn't allocate RandomX light VM");
+  }
 }
 
 typedef struct seedinfo {
@@ -185,165 +281,232 @@ typedef struct seedinfo {
   unsigned long si_count;
 } seedinfo;
 
-static CTHR_THREAD_RTYPE rx_seedthread(void *arg) {
+static CTHR_THREAD_RTYPE me_rx_seedthread(void *arg) {
   seedinfo *si = arg;
-  randomx_init_dataset(rx_dataset, si->si_cache, si->si_start, si->si_count);
+  randomx_init_dataset(main_dataset, si->si_cache, si->si_start, si->si_count);
   CTHR_THREAD_RETURN;
 }
 
-static void rx_initdata(randomx_cache *rs_cache, const int miners, const uint64_t seedheight) {
+static void me_rx_init_dataset(size_t max_threads) {
-  if (miners > 1) {
+  if (!main_dataset) {
-    unsigned long delta = randomx_dataset_item_count() / miners;
+    return;
-    unsigned long start = 0;
-    int i;
-    seedinfo *si;
-    CTHR_THREAD_TYPE *st;
-    si = malloc(miners * sizeof(seedinfo));
-    if (si == NULL)
-      local_abort("Couldn't allocate RandomX mining threadinfo");
-    st = malloc(miners * sizeof(CTHR_THREAD_TYPE));
-    if (st == NULL) {
-      free(si);
-      local_abort("Couldn't allocate RandomX mining threadlist");
   }
-    for (i=0; i<miners-1; i++) {
+
-      si[i].si_cache = rs_cache;
+  // leave 2 CPU cores for other tasks
+  const size_t num_threads = (max_threads < 4) ? 1 : (max_threads - 2);
+  seedinfo* si = malloc(num_threads * sizeof(seedinfo));
+  if (!si) local_abort("Couldn't allocate RandomX mining threadinfo");
+
+  const uint32_t delta = randomx_dataset_item_count() / num_threads;
+  uint32_t start = 0;
+
+  const size_t n1 = num_threads - 1;
+  for (size_t i = 0; i < n1; ++i) {
+    si[i].si_cache = main_cache;
     si[i].si_start = start;
     si[i].si_count = delta;
     start += delta;
   }
-    si[i].si_cache = rs_cache;
+
-    si[i].si_start = start;
+  si[n1].si_cache = main_cache;
-    si[i].si_count = randomx_dataset_item_count() - start;
+  si[n1].si_start = start;
-    for (i=1; i<miners; i++) {
+  si[n1].si_count = randomx_dataset_item_count() - start;
-      CTHR_THREAD_CREATE(st[i], rx_seedthread, &si[i]);
+
+  CTHR_THREAD_TYPE *st = malloc(num_threads * sizeof(CTHR_THREAD_TYPE));
+  if (!st) local_abort("Couldn't allocate RandomX mining threadlist");
+
+  CTHR_RWLOCK_LOCK_READ(main_cache_lock);
+  for (size_t i = 0; i < n1; ++i) {
+    if (!CTHR_THREAD_CREATE(st[i], me_rx_seedthread, &si[i])) {
+      local_abort("Couldn't start RandomX seed thread");
     }
-    randomx_init_dataset(rx_dataset, rs_cache, 0, si[0].si_count);
-    for (i=1; i<miners; i++) {
-      CTHR_THREAD_JOIN(st[i]);
   }
+  me_rx_seedthread(&si[n1]);
+  for (size_t i = 0; i < n1; ++i) CTHR_THREAD_JOIN(st[i]);
+  CTHR_RWLOCK_UNLOCK_READ(main_cache_lock);
+
   free(st);
   free(si);
+
+  minfo(RX_LOGCAT, "RandomX dataset initialized");
+}
+
+typedef struct thread_info {
+  char seedhash[HASH_SIZE];
+  size_t max_threads;
+} thread_info;
+
+static CTHR_THREAD_RTYPE me_rx_set_main_seedhash_thread(void *arg) {
+  thread_info* info = arg;
+
+  CTHR_RWLOCK_LOCK_WRITE(main_dataset_lock);
+  CTHR_RWLOCK_LOCK_WRITE(main_cache_lock);
+
+  // Double check that seedhash wasn't already updated
+  if (is_main(info->seedhash)) {
+    CTHR_RWLOCK_UNLOCK_WRITE(main_cache_lock);
+    CTHR_RWLOCK_UNLOCK_WRITE(main_dataset_lock);
+    free(info);
+    CTHR_THREAD_RETURN;
+  }
+  memcpy(main_seedhash, info->seedhash, HASH_SIZE);
+  main_seedhash_set = 1;
+
+  char buf[HASH_SIZE * 2 + 1];
+  hash2hex(main_seedhash, buf);
+  minfo(RX_LOGCAT, "RandomX new main seed hash is %s", buf);
+
+  const randomx_flags flags = enabled_flags() & ~disabled_flags();
+  rx_alloc_dataset(flags, &main_dataset, 0);
+  rx_alloc_cache(flags, &main_cache);
+
+  randomx_init_cache(main_cache, info->seedhash, HASH_SIZE);
+  minfo(RX_LOGCAT, "RandomX main cache initialized");
+
+  CTHR_RWLOCK_UNLOCK_WRITE(main_cache_lock);
+
+  // From this point, rx_slow_hash can calculate hashes in light mode, but dataset is not initialized yet
+  me_rx_init_dataset(info->max_threads);
+
+  CTHR_RWLOCK_UNLOCK_WRITE(main_dataset_lock);
+
+  free(info);
+  CTHR_THREAD_RETURN;
+}
+
+void me_rx_set_main_seedhash(const char *seedhash, size_t max_dataset_init_threads) {
+  // Early out if seedhash didn't change
+  if (is_main(seedhash)) {
+    return;
+  }
+
+  // Update main cache and dataset in the background
+  thread_info* info = malloc(sizeof(thread_info));
+  if (!info) local_abort("Couldn't allocate RandomX mining threadinfo");
+
+  memcpy(info->seedhash, seedhash, HASH_SIZE);
+  info->max_threads = max_dataset_init_threads;
+
+  CTHR_THREAD_TYPE t;
+  if (!CTHR_THREAD_CREATE(t, me_rx_set_main_seedhash_thread, info)) {
+    local_abort("Couldn't start RandomX seed thread");
+  }
+}
+
+void me_rx_slow_hash(const char *seedhash, const void *data, size_t length, char *result_hash) {
+  const randomx_flags flags = enabled_flags() & ~disabled_flags();
+  int success = 0;
+
+  // Fast path (seedhash == main_seedhash)
+  // Multiple threads can run in parallel in fast or light mode, 1-2 ms or 10-15 ms per hash per thread
+  if (is_main(seedhash)) {
+    // If CTHR_RWLOCK_TRYLOCK_READ fails it means dataset is being initialized now, so use the light mode
+    if (main_dataset && CTHR_RWLOCK_TRYLOCK_READ(main_dataset_lock)) {
+      // Double check that main_seedhash didn't change
+      if (is_main(seedhash)) {
+        rx_init_full_vm(flags, &main_vm_full);
+        if (main_vm_full) {
+          randomx_calculate_hash(main_vm_full, data, length, result_hash);
+          success = 1;
+        }
+      }
+      CTHR_RWLOCK_UNLOCK_READ(main_dataset_lock);
     } else {
-    randomx_init_dataset(rx_dataset, rs_cache, 0, randomx_dataset_item_count());
+      CTHR_RWLOCK_LOCK_READ(main_cache_lock);
-  }
+      // Double check that main_seedhash didn't change
-  rx_dataset_height = seedheight;
+      if (is_main(seedhash)) {
-}
+        rx_init_light_vm(flags, &main_vm_light, main_cache);
-
+        randomx_calculate_hash(main_vm_light, data, length, result_hash);
-void me_rx_slow_hash(const uint64_t mainheight, const uint64_t seedheight, const char *seedhash, const void *data, size_t length,
+        success = 1;
-  char *hash, int miners, int is_alt) {
-  uint64_t s_height = me_rx_seedheight(mainheight);
-  int changed = 0;
-  int toggle = is_alt ? s_height : seedheight;
-  randomx_flags flags = RANDOMX_FLAG_DEFAULT;
-  rx_state *rx_sp;
-  randomx_cache *cache;
-
-  toggle = (toggle & SEEDHASH_EPOCH_BLOCKS) != 0;
-  CTHR_MUTEX_LOCK(rx_mutex);
-
-  /* if alt block but with same seed as mainchain, no need for alt cache */
-  if (is_alt && s_height == seedheight && !memcmp(rx_s[toggle].rs_hash, seedhash, sizeof(rx_s[toggle].rs_hash)))
-    is_alt = 0;
-
-  /* RPC could request an earlier block on mainchain */
-  if (!is_alt && s_height > seedheight)
-    is_alt = 1;
-
-  toggle ^= (is_alt != 0);
-  if (toggle != rx_toggle)
-    changed = 1;
-  rx_toggle = toggle;
-
-  rx_sp = &rx_s[toggle];
-  CTHR_MUTEX_LOCK(rx_sp->rs_mutex);
-  CTHR_MUTEX_UNLOCK(rx_mutex);
-
-  cache = rx_sp->rs_cache;
-  if (cache == NULL) {
-    if (use_rx_jit())
-      flags |= RANDOMX_FLAG_JIT;
-    if (cache == NULL) {
-      cache = randomx_alloc_cache(flags | RANDOMX_FLAG_LARGE_PAGES);
-      if (cache == NULL) {
-        mdebug(RX_LOGCAT, "Couldn't use largePages for RandomX cache");
-        cache = randomx_alloc_cache(flags);
-      }
-      if (cache == NULL)
-        local_abort("Couldn't allocate RandomX cache");
-    }
-  }
-  if (rx_sp->rs_height != seedheight || rx_sp->rs_cache == NULL || memcmp(seedhash, rx_sp->rs_hash, sizeof(rx_sp->rs_hash))) {
-    randomx_init_cache(cache, seedhash, 32);
-    rx_sp->rs_cache = cache;
-    rx_sp->rs_height = seedheight;
-    memcpy(rx_sp->rs_hash, seedhash, sizeof(rx_sp->rs_hash));
-    changed = 1;
-  }
-  if (rx_vm == NULL) {
-    randomx_flags flags = RANDOMX_FLAG_DEFAULT;
-    if (use_rx_jit()) {
-      flags |= RANDOMX_FLAG_JIT;
-      if (!miners)
-          flags |= RANDOMX_FLAG_SECURE;
       }
-    if(!force_software_aes() && check_aes_hw())
+      CTHR_RWLOCK_UNLOCK_READ(main_cache_lock);
-      flags |= RANDOMX_FLAG_HARD_AES;
+    }
-    if (miners) {
+  }
-      CTHR_MUTEX_LOCK(rx_dataset_mutex);
+
-      if (rx_dataset == NULL) {
+  if (success) {
-        rx_dataset = randomx_alloc_dataset(RANDOMX_FLAG_LARGE_PAGES);
+    return;
-        if (rx_dataset == NULL) {
+  }
-          mdebug(RX_LOGCAT, "Couldn't use largePages for RandomX dataset");
+
-          rx_dataset = randomx_alloc_dataset(RANDOMX_FLAG_DEFAULT);
+  char buf[HASH_SIZE * 2 + 1];
-        }
+
-        if (rx_dataset != NULL)
+  // Slow path (seedhash != main_seedhash, but seedhash == secondary_seedhash)
-          rx_initdata(rx_sp->rs_cache, miners, seedheight);
+  // Multiple threads can run in parallel in light mode, 10-15 ms per hash per thread
-      }
+  if (!secondary_cache) {
-      if (rx_dataset != NULL)
+    CTHR_RWLOCK_LOCK_WRITE(secondary_cache_lock);
-        flags |= RANDOMX_FLAG_FULL_MEM;
+    if (!secondary_cache) {
-      else {
+      hash2hex(seedhash, buf);
-        miners = 0;
+      minfo(RX_LOGCAT, "RandomX new secondary seed hash is %s", buf);
-        mwarning(RX_LOGCAT, "Couldn't allocate RandomX dataset for miner");
+
+      rx_alloc_cache(flags, &secondary_cache);
+      randomx_init_cache(secondary_cache, seedhash, HASH_SIZE);
+      minfo(RX_LOGCAT, "RandomX secondary cache updated");
+      memcpy(secondary_seedhash, seedhash, HASH_SIZE);
+      secondary_seedhash_set = 1;
     }
-      CTHR_MUTEX_UNLOCK(rx_dataset_mutex);
+    CTHR_RWLOCK_UNLOCK_WRITE(secondary_cache_lock);
   }
-    rx_vm = randomx_create_vm(flags | RANDOMX_FLAG_LARGE_PAGES, rx_sp->rs_cache, rx_dataset);
+
-    if(rx_vm == NULL) { //large pages failed
+  CTHR_RWLOCK_LOCK_READ(secondary_cache_lock);
-      mdebug(RX_LOGCAT, "Couldn't use largePages for RandomX VM");
+  if (is_secondary(seedhash)) {
-      rx_vm = randomx_create_vm(flags, rx_sp->rs_cache, rx_dataset);
+    rx_init_light_vm(flags, &secondary_vm_light, secondary_cache);
+    randomx_calculate_hash(secondary_vm_light, data, length, result_hash);
+    success = 1;
   }
-    if(rx_vm == NULL) {//fallback if everything fails
+  CTHR_RWLOCK_UNLOCK_READ(secondary_cache_lock);
-      flags = RANDOMX_FLAG_DEFAULT | (miners ? RANDOMX_FLAG_FULL_MEM : 0);
+
-      rx_vm = randomx_create_vm(flags, rx_sp->rs_cache, rx_dataset);
+  if (success) {
+    return;
   }
-    if (rx_vm == NULL)
+
-      local_abort("Couldn't allocate RandomX VM");
+  // Slowest path (seedhash != main_seedhash, seedhash != secondary_seedhash)
-  } else if (miners) {
+  // Only one thread runs at a time and updates secondary_seedhash if needed, up to 200-500 ms per hash
-    CTHR_MUTEX_LOCK(rx_dataset_mutex);
+  CTHR_RWLOCK_LOCK_WRITE(secondary_cache_lock);
-    if (rx_dataset != NULL && rx_dataset_height != seedheight)
+  if (!is_secondary(seedhash)) {
-      rx_initdata(cache, miners, seedheight);
+    hash2hex(seedhash, buf);
-    CTHR_MUTEX_UNLOCK(rx_dataset_mutex);
+    minfo(RX_LOGCAT, "RandomX new secondary seed hash is %s", buf);
-  } else if (changed) {
+
-    randomx_vm_set_cache(rx_vm, rx_sp->rs_cache);
+    randomx_init_cache(secondary_cache, seedhash, HASH_SIZE);
+    minfo(RX_LOGCAT, "RandomX secondary cache updated");
+    memcpy(secondary_seedhash, seedhash, HASH_SIZE);
+    secondary_seedhash_set = 1;
   }
-  /* mainchain users can run in parallel */
+  rx_init_light_vm(flags, &secondary_vm_light, secondary_cache);
-  if (!is_alt)
+  randomx_calculate_hash(secondary_vm_light, data, length, result_hash);
-    CTHR_MUTEX_UNLOCK(rx_sp->rs_mutex);
+  CTHR_RWLOCK_UNLOCK_WRITE(secondary_cache_lock);
-  randomx_calculate_hash(rx_vm, data, length, hash);
-  /* altchain slot users always get fully serialized */
-  if (is_alt)
-    CTHR_MUTEX_UNLOCK(rx_sp->rs_mutex);
 }
 
-void me_rx_slow_hash_allocate_state(void) {
+void me_rx_set_miner_thread(uint32_t value, size_t max_dataset_init_threads) {
+  miner_thread = value;
+
+  // If dataset is not allocated yet, try to allocate and initialize it
+  CTHR_RWLOCK_LOCK_WRITE(main_dataset_lock);
+  if (main_dataset) {
+    CTHR_RWLOCK_UNLOCK_WRITE(main_dataset_lock);
+    return;
   }
 
-void me_rx_slow_hash_free_state(void) {
+  const randomx_flags flags = enabled_flags() & ~disabled_flags();
-  if (rx_vm != NULL) {
+  rx_alloc_dataset(flags, &main_dataset, 1);
-    randomx_destroy_vm(rx_vm);
+  me_rx_init_dataset(max_dataset_init_threads);
-    rx_vm = NULL;
+
+  CTHR_RWLOCK_UNLOCK_WRITE(main_dataset_lock);
 }
+
+uint32_t me_rx_get_miner_thread() {
+  return miner_thread;
 }
 
+void me_rx_slow_hash_allocate_state() {}
 
+static void me_rx_destroy_vm(randomx_vm** vm) {
+  if (*vm) {
+    randomx_destroy_vm(*vm);
+    *vm = NULL;
+  }
+}
+
+void me_rx_slow_hash_free_state() {
+  me_rx_destroy_vm(&main_vm_full);
+  me_rx_destroy_vm(&main_vm_light);
+  me_rx_destroy_vm(&secondary_vm_light);
+}

src/page.h

@@ -42,7 +42,7 @@ extern "C" void me_rx_slow_hash(const uint64_t mainheight, const uint64_t seedhe
                              char *hash, int miners, int is_alt);
 //extern "C" void me_rx_reorg(const uint64_t split_height);
 
-extern  __thread randomx_vm *rx_vm;
+extern  __thread randomx_vm *main_vm_full;
 
 #include <algorithm>
 #include <limits>
@@ -5695,17 +5695,29 @@ json_networkinfo()
         return j_response;
     }
 
-    uint64_t fee_estimated {0};
+    uint64_t per_kb_fee_estimated {0};
 
     // get dynamic fee estimate from last 10 blocks
-    if (!get_dynamic_per_kb_fee_estimate(fee_estimated))
+    if (!get_dynamic_per_kb_fee_estimate(per_kb_fee_estimated))
     {
         j_response["status"]  = "error";
-        j_response["message"] = "Cant get dynamic fee esimate";
+        j_response["message"] = "Cant get per kb dynamic fee esimate";
         return j_response;
     }
 
-    j_info["fee_per_kb"] = fee_estimated;
+    uint64_t fee_estimated {0};
+
+    // get dynamic fee estimate from last 10 blocks
+    //@todo: make this work
+//    if (!get_base_fee_estimate(fee_estimated))
+//    {
+//        j_response["status"]  = "error";
+//        j_response["message"] = "Cant get dynamic fee esimate";
+//        return j_response;
+//    }
+
+    j_info["fee_per_kb"] = per_kb_fee_estimated;
+    j_info["fee_estimate"] = fee_estimated;
 
     j_info["tx_pool_size"]        = MempoolStatus::mempool_no.load();
     j_info["tx_pool_size_kbytes"] = MempoolStatus::mempool_size.load();
@@ -6836,6 +6848,25 @@ get_dynamic_per_kb_fee_estimate(uint64_t& fee_estimated)
     return true;
 }
 
+bool
+get_base_fee_estimate(uint64_t& fee_estimated)
+{
+
+    string error_msg;
+
+    if (!rpc.get_base_fee_estimate(
+            FEE_ESTIMATE_GRACE_BLOCKS,
+            fee_estimated))
+    {
+        cerr << "rpc.get_base_fee_estimate failed" << endl;
+        return false;
+    }
+
+    (void) error_msg;
+
+    return true;
+}
+
 bool
 are_absolute_offsets_good(
         std::vector<uint64_t> const& absolute_offsets,
@@ -6947,18 +6978,18 @@ get_randomx_code(uint64_t blk_height,
 
     std::lock_guard<std::mutex> lk {mtx};
 
-    if (!rx_vm)
+    if (!main_vm_full)
     {
 
         crypto::hash block_hash;
 
-        // this will create rx_vm instance if one
+        // this will create main_vm_full instance if one
         // does not exist
         me_get_block_longhash(core_storage, blk, block_hash, blk_height, 0);
 
-        if (!rx_vm)
+        if (!main_vm_full)
         {
-            cerr << "rx_vm is still null!";
+            cerr << "main_vm_full is still null!";
             return {};
         }
     }
@@ -6969,32 +7000,32 @@ get_randomx_code(uint64_t blk_height,
     alignas(16) uint64_t tempHash[8];
     blake2b(tempHash, sizeof(tempHash), bd.data(), bd.size(), nullptr, 0); 
 
-    rx_vm->initScratchpad(&tempHash);
+    main_vm_full->initScratchpad(&tempHash);
-    rx_vm->resetRoundingMode();
+    main_vm_full->resetRoundingMode();
 
     for (int chain = 0; chain < RANDOMX_PROGRAM_COUNT - 1; ++chain) 
     {
-        rx_vm->run(&tempHash);
+        main_vm_full->run(&tempHash);
 
         blake2b(tempHash, sizeof(tempHash), 
-                rx_vm->getRegisterFile(), 
+                main_vm_full->getRegisterFile(),
                 sizeof(randomx::RegisterFile), nullptr, 0); 
 
         rx_code.push_back({});
 
-        rx_code.back().prog = rx_vm->getProgram();
+        rx_code.back().prog = main_vm_full->getProgram();
-    	rx_code.back().reg_file = *(rx_vm->getRegisterFile());
+        rx_code.back().reg_file = *(main_vm_full->getRegisterFile());
     }   
 
-    rx_vm->run(&tempHash);
+    main_vm_full->run(&tempHash);
 
     rx_code.push_back({});
 
-    rx_code.back().prog = rx_vm->getProgram();
+    rx_code.back().prog = main_vm_full->getProgram();
-    rx_code.back().reg_file = *(rx_vm->getRegisterFile());
+    rx_code.back().reg_file = *(main_vm_full->getRegisterFile());
 
     //crypto::hash res2;
-    //rx_vm->getFinalResult(res2.data, RANDOMX_HASH_SIZE);
+    //main_vm_full->getFinalResult(res2.data, RANDOMX_HASH_SIZE);
     //cout << "pow2: " << pod_to_hex(res2) << endl;
 
     return rx_code;

src/rpccalls.cpp

@@ -40,6 +40,33 @@ rpccalls::connect_to_monero_daemon()
     return m_http_client.connect(timeout_time_ms);
 }
 
+
+bool
+rpccalls::get_base_fee_estimate(uint64_t grace_blocks,
+                                uint64_t& fee_estimate) {
+
+    cryptonote::COMMAND_RPC_GET_BASE_FEE_ESTIMATE::request req;
+    cryptonote::COMMAND_RPC_GET_BASE_FEE_ESTIMATE::response res;
+
+    req.grace_blocks = grace_blocks;
+
+    if (!connect_to_monero_daemon())
+    {
+        cerr << "get_base_fee_estimate: not connected to daemon" << endl;
+        return false;
+    }
+
+    bool r = epee::net_utils::invoke_http_json(
+            "/get_fee_estimate",
+            req, res, m_http_client, timeout_time_ms);
+
+    fee_estimate = res.fee;
+
+    return r;
+
+
+}
+
 uint64_t
 rpccalls::get_current_height()
 {

src/rpccalls.h

@@ -105,6 +105,9 @@ public:
     bool
     get_network_info(COMMAND_RPC_GET_INFO::response& info);
 
+    bool
+    get_base_fee_estimate(uint64_t grace_blocks, uint64_t& fee_estimate);
+
     bool
     get_hardfork_info( COMMAND_RPC_HARD_FORK_INFO::response& res);
 

src/tools.cpp

@@ -627,8 +627,6 @@ sum_fees_in_txs(const vector<transaction>& txs)
     return fees_sum;
 }
 
-
-
 vector<output_tuple_with_tag>
 get_ouputs(const transaction& tx)
 {