diff --git a/src/common/tusb_private.h b/src/common/tusb_private.h
index 8643bb020..6dd0fc842 100644
--- a/src/common/tusb_private.h
+++ b/src/common/tusb_private.h
@@ -53,15 +53,13 @@ typedef struct TU_ATTR_PACKED {
 }tu_edpt_state_t;
 
 typedef struct {
-  struct TU_ATTR_PACKED  {
-    bool is_host   : 1; // 1: host, 0: device
-    bool is_mps512 : 1; // 1: 512, 0: 64 since stream is used for Bulk only
-  };
+  bool is_host; // 1: host, 0: device
   uint8_t ep_addr;
   uint16_t ep_bufsize;
 
   uint8_t  *ep_buf; // set to NULL to use xfer_fifo when CFG_TUD_EDPT_DEDICATED_HWFIFO = 1
   tu_fifo_t ff;
+  uint16_t mps;
 
   // mutex: read if rx, otherwise write
   OSAL_MUTEX_DEF(ff_mutexdef);
@@ -101,7 +99,7 @@ bool tu_edpt_stream_deinit(tu_edpt_stream_t* s);
 // Open an endpoint stream
 TU_ATTR_ALWAYS_INLINE static inline void tu_edpt_stream_open(tu_edpt_stream_t* s, tusb_desc_endpoint_t const *desc_ep) {
   s->ep_addr = desc_ep->bEndpointAddress;
-  s->is_mps512 = tu_edpt_packet_size(desc_ep) == 512;
+  s->mps = tu_edpt_packet_size(desc_ep);
 }
 
 TU_ATTR_ALWAYS_INLINE static inline bool tu_edpt_stream_is_opened(const tu_edpt_stream_t *s) {
diff --git a/src/tusb.c b/src/tusb.c
index b6cfd1260..2f34817b0 100644
--- a/src/tusb.c
+++ b/src/tusb.c
@@ -415,8 +415,7 @@ TU_ATTR_ALWAYS_INLINE static inline bool stream_release(uint8_t hwid, tu_edpt_st
 //--------------------------------------------------------------------+
 bool tu_edpt_stream_write_zlp_if_needed(uint8_t hwid, tu_edpt_stream_t* s, uint32_t last_xferred_bytes) {
   // ZLP condition: no pending data, last transferred bytes is multiple of packet size
-  const uint16_t mps = s->is_mps512 ? TUSB_EPSIZE_BULK_HS : TUSB_EPSIZE_BULK_FS;
-  TU_VERIFY(tu_fifo_empty(&s->ff) && last_xferred_bytes > 0 && (0 == (last_xferred_bytes & (mps - 1))));
+  TU_VERIFY(tu_fifo_empty(&s->ff) && last_xferred_bytes > 0 && (0 == (last_xferred_bytes & (s->mps - 1))));
   TU_VERIFY(stream_claim(hwid, s));
   TU_ASSERT(stream_xfer(hwid, s, 0));
   return true;
@@ -469,8 +468,7 @@ uint32_t tu_edpt_stream_write(uint8_t hwid, tu_edpt_stream_t *s, const void *buf
 
     // flush if fifo has more than packet size or
     // in rare case: fifo depth is configured too small (which never reach packet size)
-    const uint16_t mps = s->is_mps512 ? TUSB_EPSIZE_BULK_HS : TUSB_EPSIZE_BULK_FS;
-    if ((tu_fifo_count(&s->ff) >= mps) || (tu_fifo_depth(&s->ff) < mps)) {
+    if ((tu_fifo_count(&s->ff) >= s->mps) || (tu_fifo_depth(&s->ff) < s->mps)) {
       tu_edpt_stream_write_xfer(hwid, s);
     }
     return ret;
@@ -507,21 +505,22 @@ uint32_t tu_edpt_stream_read_xfer(uint8_t hwid, tu_edpt_stream_t* s) {
     TU_ASSERT(stream_xfer(hwid, s, s->ep_bufsize), 0);
     return s->ep_bufsize;
   } else {
-    const uint16_t mps = s->is_mps512 ? TUSB_EPSIZE_BULK_HS : TUSB_EPSIZE_BULK_FS;
     uint16_t available = tu_fifo_remaining(&s->ff);
 
     // Prepare for incoming data but only allow what we can store in the ring buffer.
     // TODO Actually we can still carry out the transfer, keeping count of received bytes
     // and slowly move it to the FIFO when read().
     // This pre-check reduces endpoint claiming
-    TU_VERIFY(available >= mps);
+    TU_VERIFY(available >= s->mps);
     TU_VERIFY(stream_claim(hwid, s), 0);
     available = tu_fifo_remaining(&s->ff); // re-get available since fifo can be changed
 
-    if (available >= mps) {
+    if (available >= s->mps) {
       // multiple of packet size limit by ep bufsize
-      uint16_t count = (uint16_t) (available & ~(mps - 1));
-      count = tu_min16(count, s->ep_bufsize);
+      uint16_t count = (uint16_t) (available & ~(s->mps - 1));
+      if (s->ep_buf != NULL) {
+        count = tu_min16(count, s->ep_bufsize);
+      }
       TU_ASSERT(stream_xfer(hwid, s, count), 0);
       return count;
     } else {