From eb10dc8f591fa2cb88dd30564430680ea420cd58 Mon Sep 17 00:00:00 2001
From: ALTracer <11005378+ALTracer@users.noreply.github.com>
Date: Tue, 1 Oct 2024 19:10:56 +0300
Subject: [PATCH 1/6] spi: blackpill-f4: Implement spi_xfer_block for faster
 reads

* Existing implementation has to walk up and down the function stack per byte,
  which is fine for commands and general poking
* 256-byte long page reads and writes can be accelerated because the length is known ahead of time
* Keep a byte in flight on stm32f1/f4 SPI (this is simpler than IRQ or DMA)
---
 src/include/platform_support.h                |  1 +
 .../common/blackpill-f4/blackpill-f4.c        | 42 +++++++++++++++++++
 src/target/spi.c                              |  8 ++++
 3 files changed, 51 insertions(+)

diff --git a/src/include/platform_support.h b/src/include/platform_support.h
index 5e04ceb332b..91aa6463af0 100644
--- a/src/include/platform_support.h
+++ b/src/include/platform_support.h
@@ -77,6 +77,7 @@ bool platform_spi_deinit(spi_bus_e bus);
 
 bool platform_spi_chip_select(uint8_t device_select);
 uint8_t platform_spi_xfer(spi_bus_e bus, uint8_t value);
+void platform_spi_xfer_block(spi_bus_e bus, uint8_t *const data, size_t count);
 #endif
 
 #ifdef PLATFORM_IDENT_DYNAMIC
diff --git a/src/platforms/common/blackpill-f4/blackpill-f4.c b/src/platforms/common/blackpill-f4/blackpill-f4.c
index 1e26b875770..0f0f49993e7 100644
--- a/src/platforms/common/blackpill-f4/blackpill-f4.c
+++ b/src/platforms/common/blackpill-f4/blackpill-f4.c
@@ -291,6 +291,48 @@ uint8_t platform_spi_xfer(const spi_bus_e bus, const uint8_t value)
 	}
 }
 
+void platform_spi_xfer_block(const spi_bus_e bus, uint8_t *const data, const size_t count)
+{
+	uint32_t spi_base = OB_SPI;
+	switch (bus) {
+	case SPI_BUS_INTERNAL:
+		spi_base = OB_SPI;
+		break;
+	case SPI_BUS_EXTERNAL:
+		spi_base = EXT_SPI;
+		break;
+	default:
+		return;
+	}
+
+#if 0
+	/* Put a byte on MOSI, wait entire transfer, grab the byte from MOSI into buffer, repeat. */
+	for (size_t i = 0; i < count; i++) {
+		uint8_t resp = spi_xfer(spi_base, data[i]);
+		data[i] = resp;
+	}
+#else
+	CM_ATOMIC_CONTEXT();
+	/*
+	 * Start the waveform by putting first MOSI byte into TXDR
+	 * from which it falls immediately into the shadow register
+	 * (pipelined polling version with inter-byte-gaps removed, no IRQ, no DMA)
+	 */
+	spi_write(spi_base, data[0]);
+	/*
+	 * Put the next MOSI byte into TXDR,
+	 * wait for previous byte from MOSI to appear in RXDR and copy it into buffer,
+	 * repeat N-2 times
+	 */
+	for (size_t i = 0; i < count - 1; i++) {
+		spi_send(spi_base, data[i + 1]);
+		data[i] = spi_read(spi_base);
+	}
+	/* Once the last MOSI byte comes out, grab the last MISO byte */
+	data[count - 1] = spi_read(spi_base);
+#endif
+}
+
 int platform_hwversion(void)
 {
 	return 0;
diff --git a/src/target/spi.c b/src/target/spi.c
index 872c68f4451..2c1b3e5eeb6 100644
--- a/src/target/spi.c
+++ b/src/target/spi.c
@@ -67,9 +67,13 @@ void bmp_spi_read(const spi_bus_e bus, const uint8_t device, const uint16_t comm
 	bmp_spi_setup_xfer(bus, device, command, address);
 	/* Now read back the data that elicited */
 	uint8_t *const data = (uint8_t *const)buffer;
+#if 0
 	for (size_t i = 0; i < length; ++i)
 		/* Do a write to read */
 		data[i] = platform_spi_xfer(bus, 0);
+#else
+	platform_spi_xfer_block(bus, data, length);
+#endif
 	/* Deselect the Flash */
 	platform_spi_chip_select(device);
 }
@@ -81,9 +85,13 @@ void bmp_spi_write(const spi_bus_e bus, const uint8_t device, const uint16_t com
 	bmp_spi_setup_xfer(bus, device, command, address);
 	/* Now write out back the data requested */
 	const uint8_t *const data = (const uint8_t *)buffer;
+#if 0
 	for (size_t i = 0; i < length; ++i)
 		/* Do a write to read */
 		platform_spi_xfer(bus, data[i]);
+#else
+	platform_spi_xfer_block(bus, data, length);
+#endif
 	/* Deselect the Flash */
 	platform_spi_chip_select(device);
 }

From 26dd5d862cc3555b5cb1a76a19045757c84b709a Mon Sep 17 00:00:00 2001
From: ALTracer <11005378+ALTracer@users.noreply.github.com>
Date: Sat, 26 Oct 2024 19:13:01 +0300
Subject: [PATCH 2/6] spi: blackpill-f4: Split out const-qualified tx_buf

---
 src/include/platform_support.h                |  2 +-
 .../common/blackpill-f4/blackpill-f4.c        | 27 ++++++++++++++-----
 src/target/spi.c                              |  4 +--
 3 files changed, 23 insertions(+), 10 deletions(-)

diff --git a/src/include/platform_support.h b/src/include/platform_support.h
index 91aa6463af0..afd74c15f69 100644
--- a/src/include/platform_support.h
+++ b/src/include/platform_support.h
@@ -77,7 +77,7 @@ bool platform_spi_deinit(spi_bus_e bus);
 
 bool platform_spi_chip_select(uint8_t device_select);
 uint8_t platform_spi_xfer(spi_bus_e bus, uint8_t value);
-void platform_spi_xfer_block(spi_bus_e bus, uint8_t *const data, size_t count);
+void platform_spi_xfer_block(spi_bus_e bus, const uint8_t *const tx_buf, uint8_t *const rx_buf, size_t count);
 #endif
 
 #ifdef PLATFORM_IDENT_DYNAMIC
diff --git a/src/platforms/common/blackpill-f4/blackpill-f4.c b/src/platforms/common/blackpill-f4/blackpill-f4.c
index 0f0f49993e7..e488abdd759 100644
--- a/src/platforms/common/blackpill-f4/blackpill-f4.c
+++ b/src/platforms/common/blackpill-f4/blackpill-f4.c
@@ -291,7 +291,15 @@ uint8_t platform_spi_xfer(const spi_bus_e bus, const uint8_t value)
 	}
 }
 
-void platform_spi_xfer_block(const spi_bus_e bus, uint8_t *const data, const size_t count)
+/**
+ * @brief Run a full-duplex SPI transfer
+ * @param bus Which BMP SPI bus to use (internal/external)
+ * @param tx_buf Transmit data buffer of at least `count` size (nonempty)
+ * @param rx_buf Receive data buffer of same size (can alias, can be NULL)
+ * @param count Byte count for the transfer, and each buffer size.
+ */
+void platform_spi_xfer_block(
+	const spi_bus_e bus, const uint8_t *const tx_buf, uint8_t *const rx_buf, const size_t count)
 {
 	uint32_t spi_base = OB_SPI;
 	switch (bus) {
@@ -308,8 +316,9 @@ void platform_spi_xfer_block(const spi_bus_e bus, uint8_t *const data, const siz
 #if 0
 	/* Put a byte on MOSI, wait entire transfer, grab the byte from MOSI into buffer, repeat. */
 	for (size_t i = 0; i < count; i++) {
-		uint8_t resp = spi_xfer(spi_base, data[i]);
-		data[i] = resp;
+		uint8_t resp = spi_xfer(spi_base, tx_buf[i]);
+		if (rx_buf)
+			rx_buf[i] = resp;
 	}
 #else
 	CM_ATOMIC_CONTEXT();
@@ -318,18 +327,22 @@ void platform_spi_xfer_block(const spi_bus_e bus, uint8_t *const data, const siz
 	 * from which it falls immediately into the shadow register
 	 * (pipelined polling version with inter-byte-gaps removed, no IRQ, no DMA)
 	 */
-	spi_write(spi_base, data[0]);
+	spi_write(spi_base, tx_buf[0]);
 	/*
 	 * Put the next MOSI byte into TXDR,
 	 * wait for previous byte from MOSI to appear in RXDR and copy it into buffer,
 	 * repeat N-2 times
 	 */
 	for (size_t i = 0; i < count - 1; i++) {
-		spi_send(spi_base, data[i + 1]);
-		data[i] = spi_read(spi_base);
+		spi_send(spi_base, tx_buf[i + 1]);
+		uint8_t resp = spi_read(spi_base);
+		if (rx_buf)
+			rx_buf[i] = resp;
 	}
 	/* Once the last MOSI byte comes out, grab the last MISO byte */
-	data[count - 1] = spi_read(spi_base);
+	uint8_t resp = spi_read(spi_base);
+	if (rx_buf)
+		rx_buf[count - 1] = resp;
 #endif
 }
 
diff --git a/src/target/spi.c b/src/target/spi.c
index 2c1b3e5eeb6..15e69c33a25 100644
--- a/src/target/spi.c
+++ b/src/target/spi.c
@@ -72,7 +72,7 @@ void bmp_spi_read(const spi_bus_e bus, const uint8_t device, const uint16_t comm
 		/* Do a write to read */
 		data[i] = platform_spi_xfer(bus, 0);
 #else
-	platform_spi_xfer_block(bus, data, length);
+	platform_spi_xfer_block(bus, data, data, length);
 #endif
 	/* Deselect the Flash */
 	platform_spi_chip_select(device);
@@ -90,7 +90,7 @@ void bmp_spi_write(const spi_bus_e bus, const uint8_t device, const uint16_t com
 		/* Do a write to read */
 		platform_spi_xfer(bus, data[i]);
 #else
-	platform_spi_xfer_block(bus, data, length);
+	platform_spi_xfer_block(bus, data, NULL, length);
 #endif
 	/* Deselect the Flash */
 	platform_spi_chip_select(device);

From 1430564e712360bd5d977c45eaad5d11c90339ef Mon Sep 17 00:00:00 2001
From: ALTracer <11005378+ALTracer@users.noreply.github.com>
Date: Sat, 26 Oct 2024 19:29:07 +0300
Subject: [PATCH 3/6] native: Implement spi_xfer_block for faster reads

---
 src/platforms/native/platform.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/src/platforms/native/platform.c b/src/platforms/native/platform.c
index e889527e872..392a128b117 100644
--- a/src/platforms/native/platform.c
+++ b/src/platforms/native/platform.c
@@ -519,6 +519,19 @@ uint8_t platform_spi_xfer(const spi_bus_e bus, const uint8_t value)
 	return spi_xfer(bus == SPI_BUS_EXTERNAL ? EXT_SPI : AUX_SPI, value);
 }
 
+void platform_spi_xfer_block(
+	const spi_bus_e bus, const uint8_t *const tx_buf, uint8_t *const rx_buf, const size_t count)
+{
+	const uint32_t spi_base = bus == SPI_BUS_EXTERNAL ? EXT_SPI : AUX_SPI;
+
+	/* Put a byte on MOSI, wait entire transfer, grab the byte from MOSI into buffer, repeat. */
+	for (size_t i = 0; i < count; i++) {
+		uint8_t resp = spi_xfer(spi_base, tx_buf[i]);
+		if (rx_buf)
+			rx_buf[i] = resp;
+	}
+}
+
 void exti15_10_isr(void)
 {
 	uint32_t usb_vbus_port;

From 6a88dbf674792ed0cbc2b49c6c6181ecdffad8ce Mon Sep 17 00:00:00 2001
From: ALTracer <11005378+ALTracer@users.noreply.github.com>
Date: Sat, 26 Oct 2024 19:34:02 +0300
Subject: [PATCH 4/6] stlink, swlink: Add no-op stubs for spi_xfer_block

---
 src/platforms/stlink/platform.c | 10 ++++++++++
 src/platforms/swlink/platform.c | 10 ++++++++++
 2 files changed, 20 insertions(+)

diff --git a/src/platforms/stlink/platform.c b/src/platforms/stlink/platform.c
index 4a423ff6039..71627b917c7 100644
--- a/src/platforms/stlink/platform.c
+++ b/src/platforms/stlink/platform.c
@@ -209,3 +209,13 @@ uint8_t platform_spi_xfer(const spi_bus_e bus, const uint8_t value)
 	(void)bus;
 	return value;
 }
+
+void platform_spi_xfer_block(
+	const spi_bus_e bus, const uint8_t *const tx_buf, uint8_t *const rx_buf, const size_t count)
+{
+	(void)bus;
+	(void)tx_buf;
+	(void)rx_buf;
+	(void)count;
+	return;
+}
diff --git a/src/platforms/swlink/platform.c b/src/platforms/swlink/platform.c
index bff34040943..cff93831fd0 100644
--- a/src/platforms/swlink/platform.c
+++ b/src/platforms/swlink/platform.c
@@ -215,3 +215,13 @@ uint8_t platform_spi_xfer(const spi_bus_e bus, const uint8_t value)
 	(void)bus;
 	return value;
 }
+
+void platform_spi_xfer_block(
+	const spi_bus_e bus, const uint8_t *const tx_buf, uint8_t *const rx_buf, const size_t count)
+{
+	(void)bus;
+	(void)tx_buf;
+	(void)rx_buf;
+	(void)count;
+	return;
+}

From 845ecb8152169d1ad6afcf0de2d8b23100515f2a Mon Sep 17 00:00:00 2001
From: ALTracer <11005378+ALTracer@users.noreply.github.com>
Date: Fri, 22 Nov 2024 20:42:31 +0300
Subject: [PATCH 5/6] fixup! spi: blackpill-f4: Implement spi_xfer_block for
 faster reads

---
 src/platforms/common/blackpill-f4/blackpill-f4.c | 9 ---------
 1 file changed, 9 deletions(-)

diff --git a/src/platforms/common/blackpill-f4/blackpill-f4.c b/src/platforms/common/blackpill-f4/blackpill-f4.c
index e488abdd759..a4dc8585a2f 100644
--- a/src/platforms/common/blackpill-f4/blackpill-f4.c
+++ b/src/platforms/common/blackpill-f4/blackpill-f4.c
@@ -313,14 +313,6 @@ void platform_spi_xfer_block(
 		return;
 	}
 
-#if 0
-	/* Put a byte on MOSI, wait entire transfer, grab the byte from MOSI into buffer, repeat. */
-	for (size_t i = 0; i < count; i++) {
-		uint8_t resp = spi_xfer(spi_base, tx_buf[i]);
-		if (rx_buf)
-			rx_buf[i] = resp;
-	}
-#else
 	CM_ATOMIC_CONTEXT();
 	/*
 	 * Start the waveform by putting first MOSI byte into TXDR
@@ -343,7 +335,6 @@ void platform_spi_xfer_block(
 	uint8_t resp = spi_read(spi_base);
 	if (rx_buf)
 		rx_buf[count - 1] = resp;
-#endif
 }
 
 int platform_hwversion(void)

From 8fb69b8c5b448b3ca8ee13953a6ba96e773fe670 Mon Sep 17 00:00:00 2001
From: ALTracer <11005378+ALTracer@users.noreply.github.com>
Date: Fri, 22 Nov 2024 20:48:36 +0300
Subject: [PATCH 6/6] fixup! spi: blackpill-f4: Split out const-qualified
 tx_buf

---
 src/include/platform_support.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/include/platform_support.h b/src/include/platform_support.h
index afd74c15f69..fdb519bc33c 100644
--- a/src/include/platform_support.h
+++ b/src/include/platform_support.h
@@ -77,7 +77,7 @@ bool platform_spi_deinit(spi_bus_e bus);
 
 bool platform_spi_chip_select(uint8_t device_select);
 uint8_t platform_spi_xfer(spi_bus_e bus, uint8_t value);
-void platform_spi_xfer_block(spi_bus_e bus, const uint8_t *const tx_buf, uint8_t *const rx_buf, size_t count);
+void platform_spi_xfer_block(spi_bus_e bus, const uint8_t *tx_buf, uint8_t *rx_buf, size_t count);
 #endif
 
 #ifdef PLATFORM_IDENT_DYNAMIC