diff --git a/drivers/can/can_mcux_flexcan.c b/drivers/can/can_mcux_flexcan.c
index b31b597324..1ed9911f77 100644
--- a/drivers/can/can_mcux_flexcan.c
+++ b/drivers/can/can_mcux_flexcan.c
@@ -25,6 +25,8 @@ LOG_MODULE_REGISTER(can_mcux_flexcan);
 	(FSL_FEATURE_FLEXCAN_HAS_MESSAGE_BUFFER_MAX_NUMBERn(0) \
 	- MCUX_FLEXCAN_MAX_RX)
 
+#define MCUX_N_TX_ALLOC_ELEM (1 + (MCUX_FLEXCAN_MAX_TX - 1) / ATOMIC_BITS)
+
 /*
  * Convert from RX message buffer index to allocated filter ID and
  * vice versa.
@@ -195,6 +197,45 @@ static void mcux_flexcan_copy_zfilter_to_mbconfig(const struct zcan_filter *src,
 	}
 }
 
+/* mcux_get_tx_alloc is a linear on array, and binary on atomic_val_t search
+ * for the highest bit set in data->tx_allocs. 0 is returned in case of an empty
+ * tx_alloc, the next free bit otherwise.
+ * The reason to always use a higher buffer number than the current in use is
+ * that a FIFO manner is kept. The Controller would otherwise send the frame
+ * that is in the lowest buffer number first.
+ */
+static int mcux_get_tx_alloc(struct mcux_flexcan_data *data)
+{
+	atomic_val_t *allocs = data->tx_allocs;
+	atomic_val_t pivot = ATOMIC_BITS / 2;
+	atomic_val_t alloc, mask;
+	int i;
+
+	for (i = MCUX_N_TX_ALLOC_ELEM - 1; i >= 0; i--) {
+		alloc = allocs[i];
+		if (alloc) {
+			for (atomic_val_t bits = ATOMIC_BITS / 2U;
+			    bits; bits >>= 1) {
+				mask = GENMASK(pivot + bits - 1, pivot);
+				if (alloc & mask) {
+					pivot += bits / 2U;
+				} else {
+					pivot -= bits / 2U;
+				}
+			}
+
+			if (!(alloc & mask)) {
+				pivot--;
+			}
+
+			break;
+		}
+	}
+
+	alloc = alloc ? (pivot + 1 + i * ATOMIC_BITS) : 0;
+	return alloc >= MCUX_FLEXCAN_MAX_TX ? -1 : alloc;
+}
+
 static int mcux_flexcan_send(struct device *dev, const struct zcan_frame *msg,
 			     s32_t timeout, can_tx_callback_t callback_isr,
 			     void *callback_arg)
@@ -205,14 +246,19 @@ static int mcux_flexcan_send(struct device *dev, const struct zcan_frame *msg,
 	status_t status;
 	int alloc;
 
-	if (k_sem_take(&data->tx_allocs_sem, timeout) != 0) {
-		return CAN_TIMEOUT;
-	}
+	while (true) {
+		alloc = mcux_get_tx_alloc(data);
+		if (alloc >= 0) {
+			if (atomic_test_and_set_bit(data->tx_allocs, alloc)) {
+				continue;
+			}
 
-	for (alloc = 0; alloc < MCUX_FLEXCAN_MAX_TX; alloc++) {
-		if (!atomic_test_and_set_bit(data->tx_allocs, alloc)) {
 			break;
 		}
+
+		if (k_sem_take(&data->tx_allocs_sem, timeout) != 0) {
+			return CAN_TIMEOUT;
+		}
 	}
 
 	mcux_flexcan_copy_zframe_to_frame(msg, &data->tx_cbs[alloc].frame);
@@ -488,8 +534,7 @@ static int mcux_flexcan_init(struct device *dev)
 	int i;
 
 	k_mutex_init(&data->rx_mutex);
-	k_sem_init(&data->tx_allocs_sem, MCUX_FLEXCAN_MAX_TX,
-		   MCUX_FLEXCAN_MAX_TX);
+	k_sem_init(&data->tx_allocs_sem, 0, 1);
 
 	for (i = 0; i < ARRAY_SIZE(data->tx_cbs); i++) {
 		k_sem_init(&data->tx_cbs[i].done, 0, 1);