boardd: misc spi fixes (#26670)

selfdrive/boardd/panda.cc

@@ -230,8 +230,6 @@ bool Panda::can_receive(std::vector<can_frame>& out_vec) {
 }
 
 bool Panda::unpack_can_buffer(uint8_t *data, int size, std::vector<can_frame> &out_vec) {
-  recv_buf.clear();
-
   if (size < sizeof(uint32_t)) {
     return true;
   }
@@ -239,7 +237,7 @@ bool Panda::unpack_can_buffer(uint8_t *data, int size, std::vector<can_frame> &o
   uint32_t magic;
   memcpy(&magic, &data[0], sizeof(uint32_t));
   if (magic != CAN_TRANSACTION_MAGIC) {
-    LOGE("CAN: MALFORMED CAN RECV PACKET");
+    LOGE("CAN recv: buffer didn't start with magic");
     handle->comms_healthy = false;
     return false;
   }

selfdrive/boardd/panda.h

@@ -43,7 +43,6 @@ struct can_frame {
 class Panda {
 private:
   std::unique_ptr<PandaCommsHandle> handle;
-  std::vector<uint8_t> recv_buf;
 
 public:
   Panda(std::string serial="", uint32_t bus_offset=0);

selfdrive/boardd/panda_comms.h

@@ -13,8 +13,6 @@
 #define TIMEOUT 0
 #define SPI_BUF_SIZE 1024
 
-const bool PANDA_NO_RETRY = getenv("PANDA_NO_RETRY");
-
 
 // comms base class
 class PandaCommsHandle {
@@ -52,7 +50,6 @@ public:
 private:
   libusb_context *ctx = NULL;
   libusb_device_handle *dev_handle = NULL;
-  std::vector<uint8_t> recv_buf;
   void handle_usb_issue(int err, const char func[]);
 };
 

selfdrive/boardd/spi.cc

@@ -6,6 +6,7 @@
 #include <cstring>
 
 #include "common/util.h"
+#include "common/timing.h"
 #include "common/swaglog.h"
 #include "panda/board/comms_definitions.h"
 #include "selfdrive/boardd/panda_comms.h"
@@ -24,6 +25,9 @@ struct __attribute__((packed)) spi_header {
   uint16_t max_rx_len;
 };
 
+const int SPI_MAX_RETRIES = 5;
+const int SPI_ACK_TIMEOUT = 50; // milliseconds
+
 
 PandaSpiHandle::PandaSpiHandle(std::string serial) : PandaCommsHandle(serial) {
   LOGD("opening SPI panda: %s", serial.c_str());
@@ -109,7 +113,7 @@ int PandaSpiHandle::bulk_read(unsigned char endpoint, unsigned char* data, int l
 int PandaSpiHandle::bulk_transfer(uint8_t endpoint, uint8_t *tx_data, uint16_t tx_len, uint8_t *rx_data, uint16_t rx_len) {
   std::lock_guard lk(hw_lock);
 
-  const int xfer_size = 0x40;
+  const int xfer_size = 0x40 * 15;
 
   int ret = 0;
   uint16_t length = (tx_data != NULL) ? tx_len : rx_len;
@@ -119,7 +123,8 @@ int PandaSpiHandle::bulk_transfer(uint8_t endpoint, uint8_t *tx_data, uint16_t t
       int len = std::min(xfer_size, tx_len - (xfer_size * i));
       d = spi_transfer_retry(endpoint, tx_data + (xfer_size * i), len, NULL, 0);
     } else {
-      d = spi_transfer_retry(endpoint, NULL, 0, rx_data + (xfer_size * i), xfer_size);
+      uint16_t to_read = std::min(xfer_size, rx_len - ret);
+      d = spi_transfer_retry(endpoint, NULL, 0, rx_data + (xfer_size * i), to_read);
     }
 
     if (d < 0) {
@@ -137,15 +142,11 @@ int PandaSpiHandle::bulk_transfer(uint8_t endpoint, uint8_t *tx_data, uint16_t t
   return ret;
 }
 
-
-
 std::vector<std::string> PandaSpiHandle::list() {
   // TODO: list all pandas available over SPI
   return {};
 }
 
-
-
 void add_checksum(uint8_t *data, int data_len) {
   data[data_len] = SPI_CHECKSUM_START;
   for (int i=0; i < data_len; i++) {
@@ -165,17 +166,19 @@ bool check_checksum(uint8_t *data, int data_len) {
 int PandaSpiHandle::spi_transfer_retry(uint8_t endpoint, uint8_t *tx_data, uint16_t tx_len, uint8_t *rx_data, uint16_t max_rx_len) {
   int ret;
 
+  int count = SPI_MAX_RETRIES;
   std::lock_guard lk(hw_lock);
   do {
     // TODO: handle error
     ret = spi_transfer(endpoint, tx_data, tx_len, rx_data, max_rx_len);
-  } while (ret < 0 && connected && !PANDA_NO_RETRY);
+    count--;
+  } while (ret < 0 && connected && count > 0);
 
   return ret;
 }
 
 int PandaSpiHandle::wait_for_ack(spi_ioc_transfer &transfer, uint8_t ack) {
-  // TODO: add timeout?
+  double start_millis = millis_since_boot();
   while (true) {
     int ret = util::safe_ioctl(spi_fd, SPI_IOC_MESSAGE(1), &transfer);
     if (ret < 0) {
@@ -189,6 +192,12 @@ int PandaSpiHandle::wait_for_ack(spi_ioc_transfer &transfer, uint8_t ack) {
       LOGW("SPI: got NACK");
       return -1;
     }
+
+    // handle timeout
+    if (millis_since_boot() - start_millis > SPI_ACK_TIMEOUT) {
+      LOGE("SPI: timed out waiting for ACK");
+      return -1;
+    }
   }
 
   return 0;