More DFU code. This might actually work.

The applications to be uploaded using 'dfu upload' have to have another
FLASH defined in their linker script.

Have to recompile some firmware tomorrow and test if this actually
works.

6 files changed, 193 insertions, 66 deletions

diff --git a/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/TOOLCHAIN_GCC_ARM/STM32F429BI.ld b/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/TOOLCHAIN_GCC_ARM/STM32F429BI.ld
index cb19009..c78e619 100644
--- a/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/TOOLCHAIN_GCC_ARM/STM32F429BI.ld
+++ b/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/TOOLCHAIN_GCC_ARM/STM32F429BI.ld
@@ -1,19 +1,18 @@
 /* Linker script to configure memory regions. */
 MEMORY
-{ 
+{
   /* FLASH (rx) : ORIGIN = 0x08000000, LENGTH = 2048k */
   BOOTLOADER (rx) : ORIGIN = 0x08000000, LENGTH = 128K
-  FIRMWARE (rx)   : ORIGIN = 0x08100000, LENGTH = 2048K - 128K
+  FIRMWARE (rx)   : ORIGIN = 0x08000000 + 128K, LENGTH = 2048K - 128K
   FLASH (rx)   : ORIGIN = 0x08000000, LENGTH = 2048K
-  CCM (rwx) : ORIGIN = 0x10000000, LENGTH = 64K
-  RAM (rwx) : ORIGIN = 0x20000000, LENGTH = 192K
+  RAM (rwx) : ORIGIN = 0x20000000, LENGTH = 192K - 4
 }
 
 /* Linker script to place sections and symbol values. Should be used together
  * with other linker script that defines memory regions FLASH and RAM.
  * It references following symbols, which must be defined in code:
  *   Reset_Handler : Entry of reset handler
- * 
+ *
  * It defines following symbols, which code can use without definition:
  *   __exidx_start
  *   __exidx_end
@@ -152,6 +151,17 @@ SECTIONS
     __StackLimit = __StackTop - SIZEOF(.stack_dummy);
     PROVIDE(__stack = __StackTop);
 
+    /* The DFU code needs to know where the firmware lives */
+    CRYPTECH_BOOTLOADER_START = ORIGIN(BOOTLOADER);
+    CRYPTECH_BOOTLOADER_END = ORIGIN(BOOTLOADER) + LENGTH(BOOTLOADER) - 1;
+    CRYPTECH_FIRMWARE_START = ORIGIN(FIRMWARE);
+    CRYPTECH_FIRMWARE_END = ORIGIN(FIRMWARE) + LENGTH(FIRMWARE) - 1;
+    /* The last 4 bytes of RAM is used to control the DFU reset+jumping.
+     * Have to be reserved in here to not get overwritten by the Reset_Handler
+     * that zeros memory. Maybe there is a better way?
+     */
+    CRYPTECH_DFU_CONTROL = ORIGIN(RAM) + LENGTH(RAM);
+
     /* Check if data + heap + stack exceeds RAM limit */
     ASSERT(__StackLimit >= __HeapLimit, "region RAM overflowed with stack")
 }
diff --git a/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/cmsis_nvic.c b/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/cmsis_nvic.c
index ac2c92a..2da63fc 100644
--- a/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/cmsis_nvic.c
+++ b/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/cmsis_nvic.c
@@ -38,8 +38,7 @@ void NVIC_SetVector(IRQn_Type IRQn, uint32_t vector) {
     uint32_t i;
 
     // Copy and switch to dynamic vectors if the first time called
-    if (SCB->VTOR == NVIC_FLASH_VECTOR_ADDRESS ||
-	SCB->VTOR == 0x08100000) {
+    if (SCB->VTOR == NVIC_FLASH_VECTOR_ADDRESS) {
         uint32_t *old_vectors = vectors;
         vectors = (uint32_t*)NVIC_RAM_VECTOR_ADDRESS;
         for (i=0; i<NVIC_NUM_VECTORS; i++) {
diff --git a/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/system_stm32f4xx.c b/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/system_stm32f4xx.c
index c2816d0..cc527ae 100644
--- a/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/system_stm32f4xx.c
+++ b/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/system_stm32f4xx.c
@@ -75,6 +75,8 @@
   #define HSI_VALUE    ((uint32_t)16000000) /*!< Value of the Internal oscillator in Hz*/
 #endif /* HSI_VALUE */
 
+extern uint32_t CRYPTECH_FIRMWARE_START;  /* defined in the linker script (STM32F429BI.ld) */
+
 /**
   * @}
   */
@@ -197,11 +199,14 @@ void SystemInit(void)
 #endif /* DATA_IN_ExtSRAM || DATA_IN_ExtSDRAM */
 
   /* Configure the Vector Table location add offset address ------------------*/
+  /* cryptech: Don't change VTOR if it is already set up by the bootloader */
+  if (SCB->VTOR != CRYPTECH_FIRMWARE_START) {
 #ifdef VECT_TAB_SRAM
-  SCB->VTOR = SRAM_BASE | VECT_TAB_OFFSET; /* Vector Table Relocation in Internal SRAM */
+      SCB->VTOR = SRAM_BASE | VECT_TAB_OFFSET; /* Vector Table Relocation in Internal SRAM */
 #else
-  SCB->VTOR = FLASH_BASE | VECT_TAB_OFFSET; /* Vector Table Relocation in Internal FLASH */
+      SCB->VTOR = FLASH_BASE | VECT_TAB_OFFSET; /* Vector Table Relocation in Internal FLASH */
 #endif
+  }
 
   /* Configure the Cube driver */
   SystemCoreClock = 16000000; // At this stage the HSI is used as system clock
diff --git a/projects/cli-test/cli-test.c b/projects/cli-test/cli-test.c
index 30623a4..1a8c6b7 100644
--- a/projects/cli-test/cli-test.c
+++ b/projects/cli-test/cli-test.c
@@ -410,6 +410,9 @@ main()
 {
     static struct cli_def cli;
 
+    /* This is simulating the bootloader from the cli-test. */
+    check_early_dfu_jump();
+
     stm_init();
 
     led_on(LED_RED);
diff --git a/projects/cli-test/mgmt-dfu.c b/projects/cli-test/mgmt-dfu.c
index 1f8aa0a..1c7e052 100644
--- a/projects/cli-test/mgmt-dfu.c
+++ b/projects/cli-test/mgmt-dfu.c
@@ -39,46 +39,136 @@
 
 #include <string.h>
 
+extern uint32_t update_crc(uint32_t crc, uint8_t *buf, int len);
+
+/* symbols defined in the linker script (STM32F429BI.ld) */
+extern uint32_t CRYPTECH_FIRMWARE_START;
+extern uint32_t CRYPTECH_FIRMWARE_END;
+extern uint32_t CRYPTECH_DFU_CONTROL;
+
+#define DFU_FIRMWARE_ADDR         ((uint32_t ) &CRYPTECH_FIRMWARE_START)
+#define DFU_FIRMWARE_PTR	  ((__IO uint32_t *) (CRYPTECH_FIRMWARE_START))
+#define DFU_FIRMWARE_END_ADDR     CRYPTECH_FIRMWARE_END
+#define HARDWARE_EARLY_DFU_JUMP   0xBADABADA
+#define DFU_UPLOAD_CHUNK_SIZE     256
+
+__IO uint32_t *dfu_control = &CRYPTECH_DFU_CONTROL;
+__IO uint32_t *dfu_new_msp = &CRYPTECH_FIRMWARE_START;
+__IO uint32_t *dfu_firmware = &CRYPTECH_FIRMWARE_START + 4;
+
+/* Flash sector offsets from RM0090, Table 6. Flash module - 2 Mbyte dual bank organization */
+#define FLASH_NUM_SECTORS 24 + 1
+uint32_t flash_sector_offsets[FLASH_NUM_SECTORS] = {
+    /* Bank 1 */
+    0x08000000, /* #0,  16 KBytes */
+    0x08004000, /* #1,  16 Kbytes */
+    0x08008000, /* #2,  16 Kbytes */
+    0x0800C000, /* #3,  16 Kbytes */
+    0x08010000, /* #4,  64 Kbytes */
+    0x08020000, /* #5,  128 Kbytes */
+    0x08040000, /* #6,  128 Kbytes */
+    0x08060000, /* #7,  128 Kbytes */
+    0x08080000, /* #8,  128 Kbytes */
+    0x080A0000, /* #9,  128 Kbytes */
+    0x080C0000, /* #10, 128 Kbytes */
+    0x080E0000, /* #11, 128 Kbytes */
+    /* Bank 2 */
+    0x08100000, /* #12,  16 Kbytes */
+    0x08104000, /* #13,  16 Kbytes */
+    0x08108000, /* #14,  16 Kbytes */
+    0x0810C000, /* #15,  16 Kbytes */
+    0x08110000, /* #16,  64 Kbytes */
+    0x08120000, /* #17, 128 Kbytes */
+    0x08140000, /* #18, 128 Kbytes */
+    0x08160000, /* #19, 128 Kbytes */
+    0x08180000, /* #20, 128 Kbytes */
+    0x081A0000, /* #21, 128 Kbytes */
+    0x081C0000, /* #22, 128 Kbytes */
+    0x081E0000, /* #23, 128 Kbytes */
+    0x08200000  /* first address *after* flash */
+};
+
+
+typedef  void (*pFunction)(void);
+
+
+/* This is it's own function to make it more convenient to set a breakpoint at it in gdb */
+void do_early_dfu_jump(void)
+{
+    //pFunction loaded_app = (pFunction) *(DFU_FIRMWARE_PTR + 1);
+    pFunction loaded_app = (pFunction) *dfu_firmware;
+    *dfu_control = 0;
+    __set_MSP(*dfu_new_msp);
+    /* Set the Vector Table Offset Register */
+    SCB->VTOR = DFU_FIRMWARE_ADDR;
+    loaded_app();
+    while (1);
+}
 
-#define DFU_BASE_ADDRESS	0x08100000
-#define DFU_BASE_PTR		(__IO uint32_t *) DFU_BASE_ADDRESS
+/* This function is called from main() before any peripherals are initialized */
+void check_early_dfu_jump(void)
+{
+    if (*dfu_control == HARDWARE_EARLY_DFU_JUMP) {
+	do_early_dfu_jump();
+    }
+}
 
+inline int _flash_sector_num(uint32_t offset)
+{
+    int i = FLASH_NUM_SECTORS - 1;
+    while (i-- >= 0) {
+	if (offset >= flash_sector_offsets[i] &&
+	    offset < flash_sector_offsets[i + 1]) {
+	    return i;
+	}
+    }
+    return -1;
+}
 
-extern uint32_t update_crc(uint32_t crc, uint8_t *buf, int len);
+int _write_to_flash(uint32_t offset, const uint32_t *buf, uint32_t elements)
+{
+    uint32_t sector = _flash_sector_num(offset);
+    uint32_t SectorError = 0, i, j;
+
+    if (offset == flash_sector_offsets[sector]) {
+	/* Request to write to beginning of a flash sector, erase it first. */
+	FLASH_EraseInitTypeDef FLASH_EraseInitStruct;
+
+	FLASH_EraseInitStruct.TypeErase = TYPEERASE_SECTORS;
+	FLASH_EraseInitStruct.Sector = sector;
+	FLASH_EraseInitStruct.NbSectors = 1;
+	FLASH_EraseInitStruct.VoltageRange = VOLTAGE_RANGE_3;
+
+	if (HAL_FLASHEx_Erase(&FLASH_EraseInitStruct, &SectorError) != HAL_OK) {
+	    return -1;
+	}
+    }
 
+    for (i = 0; i < elements; i++) {
+	if ((j = HAL_FLASH_Program(FLASH_TYPEPROGRAM_WORD, offset, buf[i])) != HAL_OK) {
+	    return -2;
+	}
+	offset += 4;
+    }
 
-/* The chunk size have to be a multiple of the SPI flash page size (256 bytes),
-   and it has to match the chunk size in the program sending the bitstream over the UART.
-*/
-#define DFU_UPLOAD_CHUNK_SIZE 256
+    return 1;
+}
 
 int cmd_dfu_upload(struct cli_def *cli, const char *command, char *argv[], int argc)
 {
-    uint32_t filesize = 0, crc = 0, my_crc = 0, counter = 0, i, j;
-    uint32_t offset = 0, n = DFU_UPLOAD_CHUNK_SIZE;
+    uint32_t filesize = 0, crc = 0, my_crc = 0, counter = 0;
+    uint32_t offset = DFU_FIRMWARE_ADDR, n = DFU_UPLOAD_CHUNK_SIZE;
     uint32_t buf[DFU_UPLOAD_CHUNK_SIZE / 4];
-    FLASH_EraseInitTypeDef FLASH_EraseInitStruct;
-    uint32_t SectorError = 0;
 
     cli_print(cli, "OK, write DFU application file size (4 bytes), data in %i byte chunks, CRC-32 (4 bytes)",
 	      DFU_UPLOAD_CHUNK_SIZE);
 
     /* Read file size (4 bytes) */
     uart_receive_bytes(STM_UART_MGMT, (void *) &filesize, 4, 1000);
-    cli_print(cli, "File size %li", filesize);
+    cli_print(cli, "File size %li, will write it to 0x%lx", filesize, offset);
 
     HAL_FLASH_Unlock();
 
-    FLASH_EraseInitStruct.TypeErase = TYPEERASE_SECTORS;
-    FLASH_EraseInitStruct.Sector = 12; /* the sector for DFU_BASE_ADDRESS (0x08100000) */
-    FLASH_EraseInitStruct.NbSectors = 1;
-    FLASH_EraseInitStruct.VoltageRange = VOLTAGE_RANGE_3;
-
-    if (HAL_FLASHEx_Erase(&FLASH_EraseInitStruct, &SectorError) != HAL_OK) {
-	cli_print(cli, "Failed erasing flash sector");
-	return CLI_ERROR;
-    }
-
     while (filesize) {
 	/* By initializing buf to the same value that erased flash has (0xff), we don't
 	 * have to try and be smart when writing the last page of data to the memory.
@@ -99,14 +189,8 @@ int cmd_dfu_upload(struct cli_def *cli, const char *command, char *argv[], int a
 	 * calculate CRC and write it to flash.
 	 */
 	my_crc = update_crc(my_crc, (uint8_t *) buf, n);
-
-	for (i = 0; i < DFU_UPLOAD_CHUNK_SIZE / 4; i++) {
-	    if ((j = HAL_FLASH_Program(FLASH_TYPEPROGRAM_WORD, DFU_BASE_ADDRESS + offset, buf[i])) != HAL_OK) {
-		cli_print(cli, "Failed writing data at offset %li: %li", offset, j);
-		return CLI_ERROR;
-	    }
-	    offset += 4;
-	}
+	_write_to_flash(offset, buf, sizeof(buf) / 4);
+	offset += DFU_UPLOAD_CHUNK_SIZE;
 
 	/* ACK this chunk by sending the current chunk counter (4 bytes) */
 	counter++;
@@ -130,48 +214,71 @@ int cmd_dfu_upload(struct cli_def *cli, const char *command, char *argv[], int a
 
 int cmd_dfu_dump(struct cli_def *cli, const char *command, char *argv[], int argc)
 {
-    cli_print(cli, "First 256 bytes from DFU application address %p:\r\n", DFU_BASE_PTR);
+    cli_print(cli, "First 256 bytes from DFU application address %p:\r\n", DFU_FIRMWARE_PTR);
 
-    uart_send_hexdump(STM_UART_MGMT, (uint8_t *) DFU_BASE_PTR, 0, 0xff);
+    uart_send_hexdump(STM_UART_MGMT, (uint8_t *) DFU_FIRMWARE_ADDR, 0, 0xff);
     uart_send_string2(STM_UART_MGMT, (char *) "\r\n\r\n");
 
     return CLI_OK;
 }
 
-typedef  int (*pFunction)(void);
+int cmd_dfu_erase(struct cli_def *cli, const char *command, char *argv[], int argc)
+{
+    uint32_t start_sector = _flash_sector_num(DFU_FIRMWARE_ADDR);
+    uint32_t end_sector = _flash_sector_num(DFU_FIRMWARE_END_ADDR);
+    uint32_t sector;
+
+    cli_print(cli, "Erasing flash sectors %li to %li (address %p to %p)",
+	      start_sector, end_sector,
+	      (uint32_t *) DFU_FIRMWARE_ADDR,
+	      (uint32_t *) DFU_FIRMWARE_END_ADDR);
+
+    if (start_sector > end_sector) {
+	cli_print(cli, "ERROR: Bad sectors");
+	return CLI_ERROR;
+    }
+
+    HAL_FLASH_Unlock();
+
+    for (sector = start_sector; sector <= end_sector; sector++) {
+	uint32_t SectorError = 0;
+	FLASH_EraseInitTypeDef FLASH_EraseInitStruct;
+
+	FLASH_EraseInitStruct.TypeErase = TYPEERASE_SECTORS;
+	FLASH_EraseInitStruct.Sector = sector;
+	FLASH_EraseInitStruct.NbSectors = 1;
+	FLASH_EraseInitStruct.VoltageRange = VOLTAGE_RANGE_3;
+
+	if (HAL_FLASHEx_Erase(&FLASH_EraseInitStruct, &SectorError) != HAL_OK) {
+	    cli_print(cli, "ERROR: Failed erasing sector %li", sector);
+	}
+    }
+    HAL_FLASH_Lock();
+
+    return CLI_OK;
+}
 
 int cmd_dfu_jump(struct cli_def *cli, const char *command, char *argv[], int argc)
 {
-    uint32_t new_msp, i;
-    /* Load first byte from the DFU_BASE_PTR to verify it contains an IVT before
+    uint32_t i;
+    /* Load first byte from the DFU_FIRMWARE_PTR to verify it contains an IVT before
      * jumping there.
      */
-    new_msp = *DFU_BASE_PTR;
-    i = new_msp & 0xFF000000;
-    /* 'i' is supposed to be a pointer to the new applications stack, it should
+    cli_print(cli, "Checking for application at %p", DFU_FIRMWARE_PTR);
+
+    //new_msp = (uint32_t) DFU_FIRMWARE_PTR;
+    i = *dfu_new_msp & 0xFF000000;
+    /* 'new_msp' is supposed to be a pointer to the new applications stack, it should
      * point either at RAM (0x20000000) or at the CCM memory (0x10000000).
      */
     if (i == 0x20000000 || i == 0x10000000) {
-	uint32_t jmp_to = *(DFU_BASE_PTR + 1);
-	pFunction loaded_app = (pFunction) jmp_to;
-
-	__disable_irq();
-	HAL_NVIC_DisableIRQ(SysTick_IRQn);
-
-	HAL_DeInit();
-
-	/* Relocate interrupt vector table */
-	//NVIC_SetVectorTable(DFU_BASE_ADDRESS);
-	SCB->VTOR == DFU_BASE_ADDRESS;
-	NVIC_SetVector(WWDG_IRQn, *DFU_BASE_PTR + 1);
-
-	/* Re-initialize stack pointer */
-	__set_MSP(new_msp);
-	/* Jump to the DFU loaded application */
-	loaded_app();
-	Error_Handler();
+	*dfu_control = HARDWARE_EARLY_DFU_JUMP;
+	cli_print(cli, "Making the leap");
+	HAL_NVIC_SystemReset();
+	while (1) { ; }
     } else {
-	cli_print(cli, "No loaded application found at %p", DFU_BASE_PTR);
+	cli_print(cli, "No loaded application found at %p (read 0x%x)",
+		  DFU_FIRMWARE_PTR, (unsigned int) *dfu_new_msp);
     }
 
     return CLI_OK;
@@ -184,4 +291,5 @@ void configure_cli_dfu(struct cli_def *cli)
     cli_command_node(dfu, dump, "Show the first 256 bytes of the loaded application");
     cli_command_node(dfu, jump, "Jump to the loaded application");
     cli_command_node(dfu, upload, "Load a new application");
+    cli_command_node(dfu, erase, "Erase the application memory");
 }
diff --git a/projects/cli-test/mgmt-dfu.h b/projects/cli-test/mgmt-dfu.h
index c38a63e..e1e3932 100644
--- a/projects/cli-test/mgmt-dfu.h
+++ b/projects/cli-test/mgmt-dfu.h
@@ -38,7 +38,9 @@
 #include "stm-init.h"
 #include <libcli.h>
 
+#define DFU_UPLOAD_CHUNK_SIZE 256
 
 extern void configure_cli_dfu(struct cli_def *cli);
+extern void check_early_dfu_jump(void);
 
 #endif /* __STM32_CLI_MGMT_DFU_H */