4 files changed, 42 insertions, 24 deletions
diff --git a/src/crypto/CMakeLists.txt b/src/crypto/CMakeLists.txt
index 9d83caca8..1e037a07d 100644
--- a/src/crypto/CMakeLists.txt
+++ b/src/crypto/CMakeLists.txt
@@ -89,3 +89,14 @@ if (ARM)
       PROPERTY COMPILE_DEFINITIONS "NO_OPTIMIZED_MULTIPLY_ON_ARM")
   endif()
 endif()
+
+# Because of the way Qt works on android with JNI, the code does not live in the main android thread
+# So this code runs with a 1 MB default stack size. 
+# This will force the use of the heap for the allocation of the scratchpad
+if (ANDROID)
+  if( BUILD_GUI_DEPS )
+    add_definitions(-DFORCE_USE_HEAP=1)
+  endif()
+endif()
+
+
diff --git a/src/crypto/oaes_lib.c b/src/crypto/oaes_lib.c
index f054a16f4..0afec6212 100644
--- a/src/crypto/oaes_lib.c
+++ b/src/crypto/oaes_lib.c
@@ -39,8 +39,8 @@
  #include <malloc.h>
 #endif
 
-// FreeBSD, and OpenBSD also don't need timeb.h
-#if !defined(__FreeBSD__) && !defined(__OpenBSD__)
+// ANDROID, FreeBSD, and OpenBSD also don't need timeb.h
+#if !defined(__FreeBSD__) && !defined(__OpenBSD__) && !defined(__ANDROID__)
  #include <sys/timeb.h>
 #else
  #include <sys/time.h>
@@ -499,7 +499,7 @@ static void oaes_get_seed( char buf[RANDSIZ + 1] )
 #else
 static uint32_t oaes_get_seed(void)
 {
-        #if !defined(__FreeBSD__) && !defined(__OpenBSD__)
+        #if !defined(__FreeBSD__) && !defined(__OpenBSD__) && !defined(__ANDROID__)
 	struct timeb timer;
 	struct tm *gmTimer;
 	char * _test = NULL;
diff --git a/src/crypto/slow-hash.c b/src/crypto/slow-hash.c
index 66d9ca5d9..43b9619f3 100644
--- a/src/crypto/slow-hash.c
+++ b/src/crypto/slow-hash.c
@@ -1052,7 +1052,6 @@ STATIC INLINE void xor_blocks(uint8_t* a, const uint8_t* b)
 
 void cn_slow_hash(const void *data, size_t length, char *hash)
 {
-    uint8_t long_state[MEMORY];
     uint8_t text[INIT_SIZE_BYTE];
     uint8_t a[AES_BLOCK_SIZE];
     uint8_t b[AES_BLOCK_SIZE];
@@ -1070,6 +1069,13 @@ void cn_slow_hash(const void *data, size_t length, char *hash)
         hash_extra_blake, hash_extra_groestl, hash_extra_jh, hash_extra_skein
     };
 
+#ifndef FORCE_USE_HEAP
+    uint8_t long_state[MEMORY];
+#else
+    uint8_t *long_state = NULL;
+    long_state = (uint8_t *)malloc(MEMORY);
+#endif
+
     hash_process(&state.hs, data, length);
     memcpy(text, state.init, INIT_SIZE_BYTE);
 
@@ -1129,6 +1135,9 @@ void cn_slow_hash(const void *data, size_t length, char *hash)
     memcpy(state.init, text, INIT_SIZE_BYTE);
     hash_permutation(&state.hs);
     extra_hashes[state.hs.b[0] & 3](&state, 200, hash);
+#ifdef FORCE_USE_HEAP
+    free(long_state);
+#endif
 }
 #endif /* !aarch64 || !crypto */
 
diff --git a/src/crypto/tree-hash.c b/src/crypto/tree-hash.c
index d73f0d959..5cdaa8c94 100644
--- a/src/crypto/tree-hash.c
+++ b/src/crypto/tree-hash.c
@@ -40,27 +40,28 @@
  #include <stdlib.h>
 #endif
 
-/// Quick check if this is power of two (use on unsigned types; in this case for size_t only)
-bool ispowerof2_size_t(size_t x) {
-	return x && !(x & (x - 1));
-}
-
 /*** 
 * Round to power of two, for count>=3 and for count being not too large (as reasonable for tree hash calculations)
 */
 size_t tree_hash_cnt(size_t count) {
-	assert( count >= 3); // cases for 0,1,2 are handled elsewhere
-	// Round down the count size: fun(2**n)= 2**(n-1) to round down to power of two
-	size_t tmp = count - 1;
-	size_t jj = 1;
-	for (jj=1 ; tmp != 0 ; ++jj) {
-		tmp /= 2; // dividing by 2 until to get how many powers of 2 fits size_to tmp
-	}
-	size_t cnt = 1 << (jj-2); // cnt is the count, but rounded down to power of two
-	// printf("count=%zu cnt=%zu jj=%zu tmp=%zu \n" , count,cnt,jj,tmp);
-	assert( cnt > 0 );	assert( cnt >= count/2 ); 	assert( cnt <= count );
-	assert( ispowerof2_size_t( cnt ));
-	return cnt;
+	// This algo has some bad history but all we are doing is 1 << floor(log2(count))
+	// There are _many_ ways to do log2, for some reason the one selected was the most obscure one,
+	// and fixing it made it even more obscure.
+	//
+	// Iterative method implemented below aims for clarity over speed, if performance is needed
+	// then my advice is to use the BSR instruction on x86
+	//
+	// All the paranoid asserts have been removed since it is trivial to mathematically prove that
+	// the return will always be a power of 2.
+	// Problem space has been defined as 3 <= count <= 2^28. Of course quarter of a billion transactions
+	// is not a sane upper limit for a block, so there will be tighter limits in other parts of the code
+
+	assert( count >= 3 ); // cases for 0,1,2 are handled elsewhere
+	assert( count <= 0x10000000 ); // sanity limit to 2^28, MSB=1 will cause an inf loop
+
+	size_t pow = 2;
+	while(pow < count) pow <<= 1;
+	return pow >> 1;
 }
 
 void tree_hash(const char (*hashes)[HASH_SIZE], size_t count, char *root_hash) {
@@ -86,9 +87,6 @@ void tree_hash(const char (*hashes)[HASH_SIZE], size_t count, char *root_hash) {
     size_t i, j;
 
     size_t cnt = tree_hash_cnt( count );
-    size_t max_size_t = (size_t) -1; // max allowed value of size_t 
-    assert( cnt < max_size_t/2 ); // reasonable size to avoid any overflows. /2 is extra; Anyway should be limited much stronger by logical code 
-    // as we have sane limits on transactions counts in blockchain rules
 
     char (*ints)[HASH_SIZE];
     size_t ints_size = cnt * HASH_SIZE;