diff --git a/components/modules/spi_master.c b/components/modules/spi_master.c
index e004f6c5..31628938 100644
--- a/components/modules/spi_master.c
+++ b/components/modules/spi_master.c
@@ -6,7 +6,7 @@
 #include "lextra.h"
 
 #include "driver/spi_master.h"
-#include "esp_heap_alloc_caps.h"
+#include "esp_heap_caps.h"
 
 #include "esp_log.h"
 
@@ -97,8 +97,8 @@ static int lspi_device_transfer( lua_State *L )
     const char * const options[] = {"std", "dio", "qio"};
     const uint32_t options_flags[] = {0, SPI_TRANS_MODE_DIO, SPI_TRANS_MODE_QIO};
 
-    CONFIG_TRANS_FROM_FIELD(command);
-    CONFIG_TRANS_FROM_FIELD(address);
+    CONFIG_TRANS_FROM_FIELD(cmd);
+    CONFIG_TRANS_FROM_FIELD(addr);
     //
     lua_getfield( L, stack, "rxlen" );
     rx_len = luaL_optint( L, -1, 0 );
@@ -128,7 +128,7 @@ static int lspi_device_transfer( lua_State *L )
 
   } else {
     // use DMA'able buffer
-    if ((trans.tx_buffer = pvPortMallocCaps( data_len, MALLOC_CAP_DMA ))) {
+    if ((trans.tx_buffer = heap_caps_malloc( data_len, MALLOC_CAP_DMA ))) {
       memcpy( (void *)trans.tx_buffer, data, data_len );
     } else {
       msg = "no memory";
@@ -147,7 +147,7 @@ static int lspi_device_transfer( lua_State *L )
 
   } else {
     // use DMA'able buffer
-    if (!(trans.rx_buffer = pvPortMallocCaps( rx_len, MALLOC_CAP_DMA ))) {
+    if (!(trans.rx_buffer = heap_caps_malloc( rx_len, MALLOC_CAP_DMA ))) {
       msg = "no mem";
       goto free_mem;
     }
@@ -167,9 +167,9 @@ static int lspi_device_transfer( lua_State *L )
 
 free_mem:
   if (!(trans.flags & SPI_TRANS_USE_TXDATA) && trans.tx_buffer)
-    free( (void *)trans.tx_buffer );
+    heap_caps_free( (void *)trans.tx_buffer );
   if (!(trans.flags & SPI_TRANS_USE_RXDATA) && trans.rx_buffer)
-    free( (void *)trans.rx_buffer );
+    heap_caps_free( (void *)trans.rx_buffer );
 
   if (msg)
     return luaL_error( L, msg );
@@ -237,9 +237,12 @@ int lspi_master( lua_State *L )
   CONFIG_BUS_PIN_FROM_FIELD(miso);
   CONFIG_BUS_PIN_FROM_FIELD(quadwp);
   CONFIG_BUS_PIN_FROM_FIELD(quadhd);
-  lua_settop( L, stack );
-  //
-  if (no_err( spi_bus_initialize( host, &config, 1 ) )) {
+  lua_pop( L, 5 );
+
+  int use_dma = luaL_optint( L, ++stack, 1 );
+  luaL_argcheck( L, use_dma >= 0 && use_dma <= 2, stack, "out of range" );
+
+  if (no_err( spi_bus_initialize( host, &config, use_dma ) )) {
     lspi_host_t *ud = (lspi_host_t *)lua_newuserdata( L, sizeof( lspi_host_t ) );
     luaL_getmetatable( L, UD_HOST_STR );
     lua_setmetatable( L, -2 );
diff --git a/components/platform/u8x8_nodemcu_hal.c b/components/platform/u8x8_nodemcu_hal.c
index fdf9b22b..6b16cd24 100644
--- a/components/platform/u8x8_nodemcu_hal.c
+++ b/components/platform/u8x8_nodemcu_hal.c
@@ -10,7 +10,7 @@
 
 #include "u8x8_nodemcu_hal.h"
 
-#include "esp_heap_alloc_caps.h"
+#include "esp_heap_caps.h"
 
 
 uint8_t u8x8_gpio_and_delay_nodemcu(u8x8_t *u8x8, uint8_t msg, uint8_t arg_int, void *arg_ptr)
@@ -289,7 +289,7 @@ uint8_t u8x8_byte_nodemcu_spi(u8x8_t *u8x8, uint8_t msg, uint8_t arg_int, void *
 
   case U8X8_MSG_BYTE_START_TRANSFER:
     hal->buffer.size = 256;
-    if (!(hal->buffer.data = (uint8_t *)pvPortMallocCaps( hal->buffer.size, MALLOC_CAP_DMA )))
+    if (!(hal->buffer.data = (uint8_t *)heap_caps_malloc( hal->buffer.size, MALLOC_CAP_DMA )))
       return 0;
     hal->buffer.used = 0;
 
@@ -300,12 +300,13 @@ uint8_t u8x8_byte_nodemcu_spi(u8x8_t *u8x8, uint8_t msg, uint8_t arg_int, void *
     while (hal->buffer.size - hal->buffer.used < arg_int) {
       hal->buffer.size *= 2;
       uint8_t *tmp;
-      if (!(tmp = (uint8_t *)pvPortMallocCaps( hal->buffer.size, MALLOC_CAP_DMA ))) {
-        free( hal->buffer.data );
+      if (!(tmp = (uint8_t *)heap_caps_malloc( hal->buffer.size, MALLOC_CAP_DMA ))) {
+        heap_caps_free( hal->buffer.data );
+        hal->buffer.data = NULL;
         return 0;
       }
       memcpy( tmp, hal->buffer.data, hal->buffer.used );
-      free( hal->buffer.data );
+      heap_caps_free( hal->buffer.data );
       hal->buffer.data = tmp;
     }
     memcpy( hal->buffer.data + hal->buffer.used, arg_ptr, arg_int );
@@ -318,7 +319,7 @@ uint8_t u8x8_byte_nodemcu_spi(u8x8_t *u8x8, uint8_t msg, uint8_t arg_int, void *
     u8x8_gpio_SetCS( u8x8, u8x8->display_info->chip_disable_level );
 
     if (hal->buffer.data)
-      free( hal->buffer.data );
+      heap_caps_free( hal->buffer.data );
     break;
 
   default:
diff --git a/docs/en/modules/spi.md b/docs/en/modules/spi.md
index 2a586bc3..af351f18 100644
--- a/docs/en/modules/spi.md
+++ b/docs/en/modules/spi.md
@@ -17,7 +17,7 @@ The host signals can be mapped to any suitable GPIO pins.
 Initializes a bus in master mode and returns a bus master object.
 
 #### Syntax
-`spi.master(host, config)`
+`spi.master(host, config[, dma])`
 
 #### Parameters
 - `host` id, one of
@@ -30,6 +30,9 @@ Initializes a bus in master mode and returns a bus master object.
     - `miso`
     - `quadwp`
     - `quadhd`
+- `dma` set DMA channel (1 or 2) or disable DMA (0), defaults to 1 if omitted.
+  Enabling DMA allows sending and receiving an unlimited amount of bytes but has restrictions in halfduplex mode (see [`spi.master:device()`](#spimasterdevice)).
+  Disabling DMA limits a transaction to 32&nbsp;bytes max.
 
 #### Returns
 SPI bus master object
@@ -61,6 +64,10 @@ none
 ## spi.master:device()
 Adds a device on the given master bus. Up to three devices per bus are supported.
 
+!!! note
+
+    Due to restrictions of the ESP IDF, halfduplex mode does not support DMA with both MOSI and MISO phases. Disable DMA during the call to [`spi.master()`](#spimaster) in this case and ensure that transaction data is not larger than 32&nbsp;bytes.
+
 #### Syntax
 `busmaster:device(config)`
 
diff --git a/sdk/esp32-esp-idf b/sdk/esp32-esp-idf
index 9b955f4c..3a1de7db 160000
--- a/sdk/esp32-esp-idf
+++ b/sdk/esp32-esp-idf
@@ -1 +1 @@
-Subproject commit 9b955f4c9f1b32652ea165d3e4cdaad01bba170e
+Subproject commit 3a1de7dba3565424a12f795b71af45dc3c2072fc