arch/arm: fix DMA transfers with sizes smaller than 4 bytes.

When the buffer to send has a size not multiple of 4 bytes, 4-byte words
are sent correctly, but when the code reaches the section to send the
remaining bytes (1, 2 or 3) it was taking the remaining byte count as origin
of the copy, instead of the buffer (as it should). This commit fixes it
to correctly copy from the buffer pointer.

Signed-off-by: Carlos Sanchez <carlossanchez@geotab.com>

arch/arm/src/common/stm32/stm32_sdio_m3m4_v1.c

@@ -1087,7 +1087,7 @@ static void stm32_sendfifo(struct stm32_dev_s *priv)
            * padding with zero as necessary to extend to a full word.
            */
 
-          uint8_t *ptr = (uint8_t *)priv->remaining;
+          uint8_t *ptr = (uint8_t *)priv->buffer;
           int i;
 
           data.w = 0;

arch/arm/src/stm32f7/stm32_sdmmc.c

@@ -1313,7 +1313,7 @@ static void stm32_sendfifo(struct stm32_dev_s *priv)
            * padding with zero as necessary to extend to a full word.
            */
 
-          uint8_t *ptr = (uint8_t *)priv->remaining;
+          uint8_t *ptr = (uint8_t *)priv->buffer;
           int i;
 
           data.w = 0;

arch/arm/src/stm32h7/stm32_sdmmc.c

@@ -1232,7 +1232,7 @@ static void stm32_sendfifo(struct stm32_dev_s *priv)
            * padding with zero as necessary to extend to a full word.
            */
 
-          uint8_t *ptr = (uint8_t *)priv->remaining;
+          uint8_t *ptr = (uint8_t *)priv->buffer;
           int i;
 
           data.w = 0;

arch/arm/src/stm32l4/stm32l4_sdmmc.c

@@ -1177,7 +1177,7 @@ static void stm32_sendfifo(struct stm32_dev_s *priv)
            * padding with zero as necessary to extend to a full word.
            */
 
-          uint8_t *ptr = (uint8_t *)priv->remaining;
+          uint8_t *ptr = (uint8_t *)priv->buffer;
           int i;
 
           data.w = 0;