stm32: Add usbfs double buffer support for bulk tx messages

Implement the usbfs fast buffer switching mechanism on the "bulk in"
endpoint.  This can improve the overall USB throughput and bus
utilization.

Signed-off-by: Kevin O'Connor <kevin@koconnor.net>

1 files changed, 28 insertions, 8 deletions

diff --git a/src/stm32/usbfs.c b/src/stm32/usbfs.c
index 661aa1b1..d166fbb7 100644
--- a/src/stm32/usbfs.c
+++ b/src/stm32/usbfs.c
@@ -118,8 +118,8 @@ btable_configure(void)
     epm_ep_desc_setup(USB_CDC_EP_ACM, BUFRX, 0);
     epm_ep_desc_setup(USB_CDC_EP_BULK_OUT, 0, USB_CDC_EP_BULK_OUT_SIZE);
     epm_ep_desc_setup(USB_CDC_EP_BULK_OUT, 1, USB_CDC_EP_BULK_OUT_SIZE);
-    epm_ep_desc_setup(USB_CDC_EP_BULK_IN, BUFTX, 0);
-    epm_ep_desc_setup(USB_CDC_EP_BULK_IN, BUFRX, 0);
+    epm_ep_desc_setup(USB_CDC_EP_BULK_IN, 0, 0);
+    epm_ep_desc_setup(USB_CDC_EP_BULK_IN, 1, 0);
 }
 
 // Read a packet stored in dedicated usb memory
@@ -236,15 +236,27 @@ usb_read_bulk_out(void *data, uint_fast8_t max_len)
     return count;
 }
 
+static uint32_t bulk_in_push_count, bulk_in_pop_flag;
+
 int_fast8_t
 usb_send_bulk_in(void *data, uint_fast8_t len)
 {
-    uint32_t ep = USB_CDC_EP_BULK_IN, epr = USB_EPR[ep];
-    if ((epr & USB_EPTX_STAT) != USB_EP_TX_NAK)
+    if (readl(&bulk_in_pop_flag))
         // No buffer space available
         return -1;
-    btable_write_packet(ep, BUFTX, data, len);
-    USB_EPR[ep] = calc_epr_bits(epr, USB_EPTX_STAT, USB_EP_TX_VALID);
+    uint32_t ep = USB_CDC_EP_BULK_IN;
+    int bufnum = bulk_in_push_count & 1;
+    bulk_in_push_count++;
+    btable_write_packet(ep, bufnum, data, len);
+    writel(&bulk_in_pop_flag, USB_EP_DTOG_RX);
+
+    // Check if hardware needs to be notified
+    uint32_t epr = USB_EPR[ep];
+    if (epr_is_dbuf_blocking(epr) && readl(&bulk_in_pop_flag)) {
+        writel(&bulk_in_pop_flag, 0);
+        USB_EPR[ep] = calc_epr_bits(epr, 0, 0) | USB_EP_DTOG_RX;
+    }
+
     return len;
 }
 
@@ -305,6 +317,11 @@ usb_set_configure(void)
     uint32_t ep = USB_CDC_EP_BULK_OUT;
     bulk_out_pop_count = 0;
     USB_EPR[ep] = calc_epr_bits(USB_EPR[ep], USB_EPRX_STAT, USB_EP_RX_VALID);
+
+    ep = USB_CDC_EP_BULK_IN;
+    bulk_in_push_count = 0;
+    writel(&bulk_in_pop_flag, 0);
+    USB_EPR[ep] = calc_epr_bits(USB_EPR[ep], USB_EPTX_STAT, USB_EP_TX_VALID);
 }
 
 
@@ -329,8 +346,9 @@ usb_reset(void)
     bulk_out_push_flag = USB_EP_DTOG_TX;
 
     ep = USB_CDC_EP_BULK_IN;
-    USB_EPR[ep] = (USB_CDC_EP_BULK_IN | USB_EP_BULK
+    USB_EPR[ep] = (USB_CDC_EP_BULK_IN | USB_EP_BULK | USB_EP_KIND
                    | USB_EP_RX_NAK | USB_EP_TX_NAK);
+    bulk_in_pop_flag = USB_EP_DTOG_RX;
 
     USB->CNTR = USB_CNTR_CTRM | USB_CNTR_RESETM;
     USB->DADDR = USB_DADDR_EF;
@@ -350,7 +368,9 @@ USB_IRQHandler(void)
             bulk_out_push_flag = 0;
             usb_notify_bulk_out();
         } else if (ep == USB_CDC_EP_BULK_IN) {
-            USB_EPR[ep] = calc_epr_bits(epr, USB_EP_CTR_RX | USB_EP_CTR_TX, 0);
+            USB_EPR[ep] = (calc_epr_bits(epr, USB_EP_CTR_RX | USB_EP_CTR_TX, 0)
+                           | bulk_in_pop_flag);
+            bulk_in_pop_flag = 0;
             usb_notify_bulk_in();
         } else if (ep == 0) {
             USB_EPR[ep] = calc_epr_bits(epr, USB_EP_CTR_RX | USB_EP_CTR_TX, 0);