From 020cb12077cb15be1700bf7733178acfb2b26428 Mon Sep 17 00:00:00 2001 From: Ake Rehnman Date: Mon, 6 Nov 2017 19:56:28 +0100 Subject: [PATCH] stm8 : new target New STM8 target based mostly on mips4k. Target communication through STLINK/SWIM. No flash driver yet but it is still possible to program flash through load_image command. The usual target debug methods are implemented. Change-Id: I7216f231d3ac7c70cae20f1cd8463c2ed864a329 Signed-off-by: Ake Rehnman Reviewed-on: http://openocd.zylin.com/3953 Tested-by: jenkins Reviewed-by: Tomas Vanek Reviewed-by: Paul Fertser --- contrib/loaders/Makefile | 5 +- contrib/loaders/erase_check/Makefile | 17 + .../loaders/erase_check/stm8_erase_check.inc | 5 + .../loaders/erase_check/stm8_erase_check.s | 69 + src/target/Makefile.am | 5 + src/target/stm8.c | 2219 +++++++++++++++++ src/target/stm8.h | 75 + src/target/target.c | 2 + tcl/target/stm8l.cfg | 87 + tcl/target/stm8s.cfg | 84 + 10 files changed, 2567 insertions(+), 1 deletion(-) create mode 100644 contrib/loaders/erase_check/stm8_erase_check.inc create mode 100644 contrib/loaders/erase_check/stm8_erase_check.s create mode 100644 src/target/stm8.c create mode 100644 src/target/stm8.h create mode 100644 tcl/target/stm8l.cfg create mode 100644 tcl/target/stm8s.cfg diff --git a/contrib/loaders/Makefile b/contrib/loaders/Makefile index 31cccb5ff..a9a27706d 100644 --- a/contrib/loaders/Makefile +++ b/contrib/loaders/Makefile @@ -1,6 +1,6 @@ .PHONY: arm clean-arm -all: arm +all: arm stm8 common_dirs = \ checksum \ @@ -32,3 +32,6 @@ clean: clean-arm for d in $(common_dirs); do \ $(MAKE) -C $$d clean; \ done + +stm8: + $(MAKE) -C erase_check stm8 diff --git a/contrib/loaders/erase_check/Makefile b/contrib/loaders/erase_check/Makefile index 01e62dead..427fa0c07 100644 --- a/contrib/loaders/erase_check/Makefile +++ b/contrib/loaders/erase_check/Makefile @@ -6,6 +6,12 @@ ARM_OBJCOPY ?= $(ARM_CROSS_COMPILE)objcopy ARM_AFLAGS = -EL +STM8_CROSS_COMPILE ?= stm8- +STM8_AS ?= $(STM8_CROSS_COMPILE)as +STM8_OBJCOPY ?= $(STM8_CROSS_COMPILE)objcopy + +STM8_AFLAGS = + arm: armv4_5_erase_check.inc armv7m_erase_check.inc armv7m_0_erase_check.inc armv4_5_%.elf: armv4_5_%.s @@ -26,5 +32,16 @@ armv7m_%.bin: armv7m_%.elf armv7m_%.inc: armv7m_%.bin $(BIN2C) < $< > $@ +stm8: stm8_erase_check.inc + +stm8_%.elf: stm8_%.s + $(STM8_AS) $(STM8_AFLAGS) $< -o $@ + +stm8_%.bin: stm8_%.elf + $(STM8_OBJCOPY) -Obinary $< $@ + +stm8_%.inc: stm8_%.bin + $(BIN2C) < $< > $@ + clean: -rm -f *.elf *.bin *.inc diff --git a/contrib/loaders/erase_check/stm8_erase_check.inc b/contrib/loaders/erase_check/stm8_erase_check.inc new file mode 100644 index 000000000..66b4ec7f5 --- /dev/null +++ b/contrib/loaders/erase_check/stm8_erase_check.inc @@ -0,0 +1,5 @@ +/* Autogenerated with ../../../src/helper/bin2char.sh */ +0x00,0x80,0x00,0x00,0x80,0x00,0x96,0xcf,0x00,0x22,0x1e,0x01,0x16,0x04,0xa6,0xff, +0x90,0x5d,0x26,0x04,0x0d,0x03,0x27,0x17,0x90,0x5d,0x26,0x02,0x0a,0x03,0x90,0x5a, +0x92,0xbc,0x00,0x00,0xa1,0xff,0x26,0x07,0x5c,0x26,0xe5,0x0c,0x00,0x20,0xe1,0x1f, +0x01,0x17,0x04,0x8b, diff --git a/contrib/loaders/erase_check/stm8_erase_check.s b/contrib/loaders/erase_check/stm8_erase_check.s new file mode 100644 index 000000000..62694006b --- /dev/null +++ b/contrib/loaders/erase_check/stm8_erase_check.s @@ -0,0 +1,69 @@ +/* + Copyright (C) 2017 Ake Rehnman + ake.rehnman(at)gmail.com + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ +;; +;; erase check memory code +;; + .org 0x0 +;; start address + start_addr: .byte 0x00 + .word 0x8000 +;; byte count + byte_cnt: .byte 0x00 + .word 0x8000 +; +; SP must point to start_addr on entry +; first relocate start_addr to the location +; we are running at +start: + ldw X,SP + ldw .cont+2,X + ldw X,(start_addr+1,SP) ;start addr + ldw Y,(byte_cnt+1,SP) ;count + ld A,#0xff +; +; if count == 0 return +.L1: + tnzw Y + jrne .decrcnt ;continue if low word != 0 + tnz (byte_cnt,SP) ;high byte + jreq .exit ;goto exit +; +; decrement count (byte_cnt) +.decrcnt: + tnzw Y ;low word count + jrne .decr1 + dec (byte_cnt,SP) ;high byte +.decr1: + decw Y; decr low word +; +; first check if [start_addr] is 0xff +.cont: + ldf A, [start_addr.e] + cp A,#0xff + jrne .exit ;exit if not 0xff +; +; increment start_addr (addr) + incw X + jrne .L1 + inc (start_addr,SP) ;increment high byte + jra .L1 +; +.exit: + ldw (start_addr+1,SP),X ;start addr + ldw (byte_cnt+1,SP),Y ;count + break diff --git a/src/target/Makefile.am b/src/target/Makefile.am index 597070c4b..d2aab0a5e 100644 --- a/src/target/Makefile.am +++ b/src/target/Makefile.am @@ -19,6 +19,7 @@ noinst_LTLIBRARIES += %D%/libtarget.la $(AVR32_SRC) \ $(MIPS32_SRC) \ $(NDS32_SRC) \ + $(STM8_SRC) \ $(INTEL_IA32_SRC) \ %D%/avrt.c \ %D%/dsp563xx.c \ @@ -124,6 +125,9 @@ NDS32_SRC = \ %D%/nds32_v3m.c \ %D%/nds32_aice.c +STM8_SRC = \ + %D%/stm8.c + INTEL_IA32_SRC = \ %D%/quark_x10xx.c \ %D%/quark_d20xx.c \ @@ -205,6 +209,7 @@ INTEL_IA32_SRC = \ %D%/nds32_v3.h \ %D%/nds32_v3m.h \ %D%/nds32_aice.h \ + %D%/stm8.h \ %D%/lakemont.h \ %D%/x86_32_common.h \ %D%/arm_cti.h diff --git a/src/target/stm8.c b/src/target/stm8.c new file mode 100644 index 000000000..262497b0d --- /dev/null +++ b/src/target/stm8.c @@ -0,0 +1,2219 @@ +/* + OpenOCD STM8 target driver + Copyright (C) 2017 Ake Rehnman + ake.rehnman(at)gmail.com + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include "target.h" +#include "target_type.h" +#include "hello.h" +#include "jtag/jtag.h" +#include "jtag/hla/hla_transport.h" +#include "jtag/hla/hla_interface.h" +#include "jtag/hla/hla_layout.h" +#include "register.h" +#include "breakpoints.h" +#include "algorithm.h" +#include "stm8.h" + +static struct reg_cache *stm8_build_reg_cache(struct target *target); +static int stm8_read_core_reg(struct target *target, unsigned int num); +static int stm8_write_core_reg(struct target *target, unsigned int num); +static int stm8_save_context(struct target *target); +static void stm8_enable_breakpoints(struct target *target); +static int stm8_unset_breakpoint(struct target *target, + struct breakpoint *breakpoint); +static int stm8_set_breakpoint(struct target *target, + struct breakpoint *breakpoint); +static void stm8_enable_watchpoints(struct target *target); +static int stm8_unset_watchpoint(struct target *target, + struct watchpoint *watchpoint); + +static const struct { + unsigned id; + const char *name; + const uint8_t bits; + enum reg_type type; + const char *group; + const char *feature; + int flag; +} stm8_regs[] = { + { 0, "pc", 32, REG_TYPE_UINT32, "general", "org.gnu.gdb.stm8.core", 0 }, + { 1, "a", 8, REG_TYPE_UINT8, "general", "org.gnu.gdb.stm8.core", 0 }, + { 2, "x", 16, REG_TYPE_UINT16, "general", "org.gnu.gdb.stm8.core", 0 }, + { 3, "y", 16, REG_TYPE_UINT16, "general", "org.gnu.gdb.stm8.core", 0 }, + { 4, "sp", 16, REG_TYPE_UINT16, "general", "org.gnu.gdb.stm8.core", 0 }, + { 5, "cc", 8, REG_TYPE_UINT8, "general", "org.gnu.gdb.stm8.core", 0 }, +}; + +#define STM8_NUM_REGS ARRAY_SIZE(stm8_regs) +#define STM8_PC 0 +#define STM8_A 1 +#define STM8_X 2 +#define STM8_Y 3 +#define STM8_SP 4 +#define STM8_CC 5 + +#define CC_I0 0x8 +#define CC_I1 0x20 + +#define DM_REGS 0x7f00 +#define DM_REG_A 0x7f00 +#define DM_REG_PC 0x7f01 +#define DM_REG_X 0x7f04 +#define DM_REG_Y 0x7f06 +#define DM_REG_SP 0x7f08 +#define DM_REG_CC 0x7f0a + +#define DM_BKR1E 0x7f90 +#define DM_BKR2E 0x7f93 +#define DM_CR1 0x7f96 +#define DM_CR2 0x7f97 +#define DM_CSR1 0x7f98 +#define DM_CSR2 0x7f99 + +#define STE 0x40 +#define STF 0x20 +#define RST 0x10 +#define BRW 0x08 +#define BK2F 0x04 +#define BK1F 0x02 + +#define SWBRK 0x20 +#define SWBKF 0x10 +#define STALL 0x08 +#define FLUSH 0x01 + +#define FLASH_CR1_STM8S 0x505A +#define FLASH_CR2_STM8S 0x505B +#define FLASH_NCR2_STM8S 0x505C +#define FLASH_IAPSR_STM8S 0x505F +#define FLASH_PUKR_STM8S 0x5062 +#define FLASH_DUKR_STM8S 0x5064 + +#define FLASH_CR1_STM8L 0x5050 +#define FLASH_CR2_STM8L 0x5051 +#define FLASH_NCR2_STM8L 0 +#define FLASH_PUKR_STM8L 0x5052 +#define FLASH_DUKR_STM8L 0x5053 +#define FLASH_IAPSR_STM8L 0x5054 + +/* FLASH_IAPSR */ +#define HVOFF 0x40 +#define DUL 0x08 +#define EOP 0x04 +#define PUL 0x02 +#define WR_PG_DIS 0x01 + +/* FLASH_CR2 */ +#define OPT 0x80 +#define WPRG 0x40 +#define ERASE 0x20 +#define FPRG 0x10 +#define PRG 0x01 + +/* SWIM_CSR */ +#define SAFE_MASK 0x80 +#define NO_ACCESS 0x40 +#define SWIM_DM 0x20 +#define HS 0x10 +#define OSCOFF 0x08 +#define SWIM_RST 0x04 +#define HSIT 0x02 +#define PRI 0x01 + +#define SWIM_CSR 0x7f80 + +#define STM8_BREAK 0x8B + +enum mem_type { + RAM, + FLASH, + EEPROM, + OPTION +}; + +struct stm8_algorithm { + int common_magic; +}; + +struct stm8_core_reg { + uint32_t num; + struct target *target; + struct stm8_common *stm8_common; +}; + +enum hw_break_type { + /* break on execute */ + HWBRK_EXEC, + /* break on read */ + HWBRK_RD, + /* break on write */ + HWBRK_WR, + /* break on read, write and execute */ + HWBRK_ACC +}; + +struct stm8_comparator { + bool used; + uint32_t bp_value; + uint32_t reg_address; + enum hw_break_type type; +}; + +static inline struct hl_interface_s *target_to_adapter(struct target *target) +{ + return target->tap->priv; +} + +static int stm8_adapter_read_memory(struct target *target, + uint32_t addr, int size, int count, void *buf) +{ + int ret; + struct hl_interface_s *adapter = target_to_adapter(target); + + ret = adapter->layout->api->read_mem(adapter->handle, + addr, size, count, buf); + if (ret != ERROR_OK) + return ret; + return ERROR_OK; +} + +static int stm8_adapter_write_memory(struct target *target, + uint32_t addr, int size, int count, const void *buf) +{ + int ret; + struct hl_interface_s *adapter = target_to_adapter(target); + + ret = adapter->layout->api->write_mem(adapter->handle, + addr, size, count, buf); + if (ret != ERROR_OK) + return ret; + return ERROR_OK; +} + +static int stm8_write_u8(struct target *target, + uint32_t addr, uint8_t val) +{ + int ret; + uint8_t buf[1]; + struct hl_interface_s *adapter = target_to_adapter(target); + + buf[0] = val; + ret = adapter->layout->api->write_mem(adapter->handle, addr, 1, 1, buf); + if (ret != ERROR_OK) + return ret; + return ERROR_OK; +} + +static int stm8_read_u8(struct target *target, + uint32_t addr, uint8_t *val) +{ + int ret; + struct hl_interface_s *adapter = target_to_adapter(target); + + ret = adapter->layout->api->read_mem(adapter->handle, addr, 1, 1, val); + if (ret != ERROR_OK) + return ret; + return ERROR_OK; +} + +static int stm8_set_speed(struct target *target, int speed) +{ + struct hl_interface_s *adapter = target_to_adapter(target); + adapter->layout->api->speed(adapter->handle, speed, 0); + return ERROR_OK; +} + +/* + Disables interrupts. + If interrupts are enabled they are masked and the cc register + is saved. + + Enables interrupts. + Enable interrupts is actually restoring I1 I0 state from previous + call with enable == 0. Note that if stepping and breaking on a sim + instruction will NOT work since the interrupt flags are restored on + debug_entry. We don't have any way for the debugger to exclusively + disable the interrupts +*/ +static int stm8_enable_interrupts(struct target *target, int enable) +{ + struct stm8_common *stm8 = target_to_stm8(target); + uint8_t cc; + + if (enable) { + if (!stm8->cc_valid) + return ERROR_OK; /* cc was not stashed */ + /* fetch current cc */ + stm8_read_u8(target, DM_REG_CC, &cc); + /* clear I1 I0 */ + cc &= ~(CC_I0 + CC_I1); + /* restore I1 & I0 from stash*/ + cc |= (stm8->cc & (CC_I0+CC_I1)); + /* update current cc */ + stm8_write_u8(target, DM_REG_CC, cc); + stm8->cc_valid = false; + } else { + stm8_read_u8(target, DM_REG_CC, &cc); + if ((cc & CC_I0) && (cc & CC_I1)) + return ERROR_OK; /* interrupts already masked */ + /* stash cc */ + stm8->cc = cc; + stm8->cc_valid = true; + /* mask interrupts (disable) */ + cc |= (CC_I0 + CC_I1); + stm8_write_u8(target, DM_REG_CC, cc); + } + + return ERROR_OK; +} + +static int stm8_set_hwbreak(struct target *target, + struct stm8_comparator comparator_list[]) +{ + uint8_t buf[3]; + int i, ret; + + /* Refer to Table 4 in UM0470 */ + uint8_t bc = 0x5; + uint8_t bir = 0; + uint8_t biw = 0; + + uint32_t data; + uint32_t addr; + + if (!comparator_list[0].used) { + comparator_list[0].type = HWBRK_EXEC; + comparator_list[0].bp_value = -1; + } + + if (!comparator_list[1].used) { + comparator_list[1].type = HWBRK_EXEC; + comparator_list[1].bp_value = -1; + } + + if ((comparator_list[0].type == HWBRK_EXEC) + && (comparator_list[1].type == HWBRK_EXEC)) { + comparator_list[0].reg_address = 0; + comparator_list[1].reg_address = 1; + } + + if ((comparator_list[0].type == HWBRK_EXEC) + && (comparator_list[1].type != HWBRK_EXEC)) { + comparator_list[0].reg_address = 0; + comparator_list[1].reg_address = 1; + switch (comparator_list[1].type) { + case HWBRK_RD: + bir = 1; + break; + case HWBRK_WR: + biw = 1; + break; + default: + bir = 1; + biw = 1; + break; + } + } + + if ((comparator_list[1].type == HWBRK_EXEC) + && (comparator_list[0].type != HWBRK_EXEC)) { + comparator_list[0].reg_address = 1; + comparator_list[1].reg_address = 0; + switch (comparator_list[0].type) { + case HWBRK_RD: + bir = 1; + break; + case HWBRK_WR: + biw = 1; + break; + default: + bir = 1; + biw = 1; + break; + } + } + + if ((comparator_list[0].type != HWBRK_EXEC) + && (comparator_list[1].type != HWBRK_EXEC)) { + if ((comparator_list[0].type != comparator_list[1].type)) { + LOG_ERROR("data hw breakpoints must be of same type"); + return ERROR_TARGET_RESOURCE_NOT_AVAILABLE; + } + } + + for (i = 0; i < 2; i++) { + data = comparator_list[i].bp_value; + addr = comparator_list[i].reg_address; + + buf[0] = data >> 16; + buf[1] = data >> 8; + buf[2] = data; + + if (addr == 0) { + ret = stm8_adapter_write_memory(target, DM_BKR1E, 1, 3, buf); + LOG_DEBUG("DM_BKR1E=%" PRIx32, data); + } else if (addr == 1) { + ret = stm8_adapter_write_memory(target, DM_BKR2E, 1, 3, buf); + LOG_DEBUG("DM_BKR2E=%" PRIx32, data); + } else { + LOG_DEBUG("addr=%" PRIu32, addr); + return ERROR_FAIL; + } + + if (ret != ERROR_OK) + return ret; + + ret = stm8_write_u8(target, DM_CR1, + (bc << 3) + (bir << 2) + (biw << 1)); + LOG_DEBUG("DM_CR1=%" PRIx8, buf[0]); + if (ret != ERROR_OK) + return ret; + + } + return ERROR_OK; +} + +/* read DM control and status regs */ +static int stm8_read_dm_csrx(struct target *target, uint8_t *csr1, + uint8_t *csr2) +{ + int ret; + uint8_t buf[2]; + + ret = stm8_adapter_read_memory(target, DM_CSR1, 1, sizeof(buf), buf); + if (ret != ERROR_OK) + return ret; + if (csr1) + *csr1 = buf[0]; + if (csr2) + *csr2 = buf[1]; + return ERROR_OK; +} + +/* set or clear the single step flag in DM */ +static int stm8_config_step(struct target *target, int enable) +{ + int ret; + uint8_t csr1, csr2; + + ret = stm8_read_dm_csrx(target, &csr1, &csr2); + if (ret != ERROR_OK) + return ret; + if (enable) + csr1 |= STE; + else + csr1 &= ~STE; + + ret = stm8_write_u8(target, DM_CSR1, csr1); + if (ret != ERROR_OK) + return ret; + return ERROR_OK; +} + +/* set the stall flag in DM */ +static int stm8_debug_stall(struct target *target) +{ + int ret; + uint8_t csr1, csr2; + + ret = stm8_read_dm_csrx(target, &csr1, &csr2); + if (ret != ERROR_OK) + return ret; + csr2 |= STALL; + ret = stm8_write_u8(target, DM_CSR2, csr2); + if (ret != ERROR_OK) + return ret; + return ERROR_OK; +} + +static int stm8_configure_break_unit(struct target *target) +{ + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + + if (stm8->bp_scanned) + return ERROR_OK; + + stm8->num_hw_bpoints = 2; + stm8->num_hw_bpoints_avail = stm8->num_hw_bpoints; + + stm8->hw_break_list = calloc(stm8->num_hw_bpoints, + sizeof(struct stm8_comparator)); + + stm8->hw_break_list[0].reg_address = 0; + stm8->hw_break_list[1].reg_address = 1; + + LOG_DEBUG("hw breakpoints: numinst %i numdata %i", stm8->num_hw_bpoints, + stm8->num_hw_bpoints); + + stm8->bp_scanned = true; + + return ERROR_OK; +} + +static int stm8_examine_debug_reason(struct target *target) +{ + int retval; + uint8_t csr1, csr2; + + retval = stm8_read_dm_csrx(target, &csr1, &csr2); + LOG_DEBUG("csr1 = 0x%02X csr2 = 0x%02X", csr1, csr2); + + if ((target->debug_reason != DBG_REASON_DBGRQ) + && (target->debug_reason != DBG_REASON_SINGLESTEP)) { + + if (retval != ERROR_OK) + return retval; + + if (csr1 & RST) + /* halted on reset */ + target->debug_reason = DBG_REASON_UNDEFINED; + + if (csr1 & (BK1F+BK2F)) + /* we have halted on a breakpoint (or wp)*/ + target->debug_reason = DBG_REASON_BREAKPOINT; + + if (csr2 & SWBKF) + /* we have halted on a breakpoint */ + target->debug_reason = DBG_REASON_BREAKPOINT; + + } + + return ERROR_OK; +} + +static int stm8_debug_entry(struct target *target) +{ + struct stm8_common *stm8 = target_to_stm8(target); + + /* restore interrupts */ + stm8_enable_interrupts(target, 1); + + stm8_save_context(target); + + /* make sure stepping disabled STE bit in CSR1 cleared */ + stm8_config_step(target, 0); + + /* attempt to find halt reason */ + stm8_examine_debug_reason(target); + + LOG_DEBUG("entered debug state at PC 0x%" PRIx32 ", target->state: %s", + buf_get_u32(stm8->core_cache->reg_list[STM8_PC].value, 0, 32), + target_state_name(target)); + + return ERROR_OK; +} + +/* clear stall flag in DM and flush instruction pipe */ +static int stm8_exit_debug(struct target *target) +{ + int ret; + uint8_t csr1, csr2; + + ret = stm8_read_dm_csrx(target, &csr1, &csr2); + if (ret != ERROR_OK) + return ret; + csr2 |= FLUSH; + ret = stm8_write_u8(target, DM_CSR2, csr2); + if (ret != ERROR_OK) + return ret; + + csr2 &= ~STALL; + csr2 |= SWBRK; + ret = stm8_write_u8(target, DM_CSR2, csr2); + if (ret != ERROR_OK) + return ret; + return ERROR_OK; +} + +static int stm8_read_regs(struct target *target, uint32_t regs[]) +{ + int ret; + uint8_t buf[11]; + + ret = stm8_adapter_read_memory(target, DM_REGS, 1, sizeof(buf), buf); + if (ret != ERROR_OK) + return ret; + + regs[0] = be_to_h_u24(buf+DM_REG_PC-DM_REGS); + regs[1] = buf[DM_REG_A-DM_REGS]; + regs[2] = be_to_h_u16(buf+DM_REG_X-DM_REGS); + regs[3] = be_to_h_u16(buf+DM_REG_Y-DM_REGS); + regs[4] = be_to_h_u16(buf+DM_REG_SP-DM_REGS); + regs[5] = buf[DM_REG_CC-DM_REGS]; + + return ERROR_OK; +} + +static int stm8_write_regs(struct target *target, uint32_t regs[]) +{ + int ret; + uint8_t buf[11]; + + h_u24_to_be(buf+DM_REG_PC-DM_REGS, regs[0]); + buf[DM_REG_A-DM_REGS] = regs[1]; + h_u16_to_be(buf+DM_REG_X-DM_REGS, regs[2]); + h_u16_to_be(buf+DM_REG_Y-DM_REGS, regs[3]); + h_u16_to_be(buf+DM_REG_SP-DM_REGS, regs[4]); + buf[DM_REG_CC-DM_REGS] = regs[5]; + + ret = stm8_adapter_write_memory(target, DM_REGS, 1, sizeof(buf), buf); + if (ret != ERROR_OK) + return ret; + + return ERROR_OK; +} + +static int stm8_get_core_reg(struct reg *reg) +{ + int retval; + struct stm8_core_reg *stm8_reg = reg->arch_info; + struct target *target = stm8_reg->target; + struct stm8_common *stm8_target = target_to_stm8(target); + + if (target->state != TARGET_HALTED) + return ERROR_TARGET_NOT_HALTED; + + retval = stm8_target->read_core_reg(target, stm8_reg->num); + + return retval; +} + +static int stm8_set_core_reg(struct reg *reg, uint8_t *buf) +{ + struct stm8_core_reg *stm8_reg = reg->arch_info; + struct target *target = stm8_reg->target; + uint32_t value = buf_get_u32(buf, 0, reg->size); + + if (target->state != TARGET_HALTED) + return ERROR_TARGET_NOT_HALTED; + + buf_set_u32(reg->value, 0, 32, value); + reg->dirty = true; + reg->valid = true; + + return ERROR_OK; +} + +static int stm8_save_context(struct target *target) +{ + unsigned int i; + + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + + /* read core registers */ + stm8_read_regs(target, stm8->core_regs); + + for (i = 0; i < STM8_NUM_REGS; i++) { + if (!stm8->core_cache->reg_list[i].valid) + stm8->read_core_reg(target, i); + } + + return ERROR_OK; +} + +static int stm8_restore_context(struct target *target) +{ + unsigned int i; + + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + + for (i = 0; i < STM8_NUM_REGS; i++) { + if (stm8->core_cache->reg_list[i].dirty) + stm8->write_core_reg(target, i); + } + + /* write core regs */ + stm8_write_regs(target, stm8->core_regs); + + return ERROR_OK; +} + +static int stm8_unlock_flash(struct target *target) +{ + uint8_t data[1]; + + struct stm8_common *stm8 = target_to_stm8(target); + + /* check if flash is unlocked */ + stm8_read_u8(target, stm8->flash_iapsr, data); + if (~data[0] & PUL) { + /* unlock flash */ + stm8_write_u8(target, stm8->flash_pukr, 0x56); + stm8_write_u8(target, stm8->flash_pukr, 0xae); + } + + stm8_read_u8(target, stm8->flash_iapsr, data); + if (~data[0] & PUL) + return ERROR_FAIL; + return ERROR_OK; +} + +static int stm8_unlock_eeprom(struct target *target) +{ + uint8_t data[1]; + + struct stm8_common *stm8 = target_to_stm8(target); + + /* check if eeprom is unlocked */ + stm8_read_u8(target, stm8->flash_iapsr, data); + if (~data[0] & DUL) { + /* unlock eeprom */ + stm8_write_u8(target, stm8->flash_dukr, 0xae); + stm8_write_u8(target, stm8->flash_dukr, 0x56); + } + + stm8_read_u8(target, stm8->flash_iapsr, data); + if (~data[0] & DUL) + return ERROR_FAIL; + return ERROR_OK; +} + +static int stm8_write_flash(struct target *target, enum mem_type type, + uint32_t address, + uint32_t size, uint32_t count, uint32_t blocksize_param, + const uint8_t *buffer) +{ + struct stm8_common *stm8 = target_to_stm8(target); + + uint8_t iapsr; + uint8_t opt = 0; + unsigned int i; + uint32_t blocksize = 0; + uint32_t bytecnt; + int res; + + switch (type) { + case (FLASH): + stm8_unlock_flash(target); + break; + case (EEPROM): + stm8_unlock_eeprom(target); + break; + case (OPTION): + stm8_unlock_eeprom(target); + opt = OPT; + break; + default: + LOG_ERROR("BUG: wrong mem_type %d", type); + assert(0); + } + + if (size == 2) { + /* we don't support short writes */ + count = count * 2; + size = 1; + } + + bytecnt = count * size; + + while (bytecnt) { + if ((bytecnt >= blocksize_param) && ((address & (blocksize_param-1)) == 0)) { + if (stm8->flash_cr2) + stm8_write_u8(target, stm8->flash_cr2, PRG + opt); + if (stm8->flash_ncr2) + stm8_write_u8(target, stm8->flash_ncr2, ~(PRG + opt)); + blocksize = blocksize_param; + } else + if ((bytecnt >= 4) && ((address & 0x3) == 0)) { + if (stm8->flash_cr2) + stm8_write_u8(target, stm8->flash_cr2, WPRG + opt); + if (stm8->flash_ncr2) + stm8_write_u8(target, stm8->flash_ncr2, ~(WPRG + opt)); + blocksize = 4; + } else + if (blocksize != 1) { + if (stm8->flash_cr2) + stm8_write_u8(target, stm8->flash_cr2, opt); + if (stm8->flash_ncr2) + stm8_write_u8(target, stm8->flash_ncr2, ~opt); + blocksize = 1; + } + + res = stm8_adapter_write_memory(target, address, 1, blocksize, buffer); + if (res != ERROR_OK) + return res; + address += blocksize; + buffer += blocksize; + bytecnt -= blocksize; + + /* lets hang here until end of program (EOP) */ + for (i = 0; i < 16; i++) { + stm8_read_u8(target, stm8->flash_iapsr, &iapsr); + if (iapsr & EOP) + break; + else + usleep(1000); + } + if (i == 16) + return ERROR_FAIL; + } + + /* disable write access */ + res = stm8_write_u8(target, stm8->flash_iapsr, 0x0); + + if (res != ERROR_OK) + return ERROR_FAIL; + + return ERROR_OK; +} + +static int stm8_write_memory(struct target *target, target_addr_t address, + uint32_t size, uint32_t count, + const uint8_t *buffer) +{ + struct stm8_common *stm8 = target_to_stm8(target); + + LOG_DEBUG("address: 0x%8.8" TARGET_PRIxADDR + ", size: 0x%8.8" PRIx32 + ", count: 0x%8.8" PRIx32, + address, size, count); + + if (target->state != TARGET_HALTED) + LOG_WARNING("target not halted"); + + int retval; + + if ((address >= stm8->flashstart) && (address <= stm8->flashend)) + retval = stm8_write_flash(target, FLASH, address, size, count, + stm8->blocksize, buffer); + else if ((address >= stm8->eepromstart) && (address <= stm8->eepromend)) + retval = stm8_write_flash(target, EEPROM, address, size, count, + stm8->blocksize, buffer); + else if ((address >= stm8->optionstart) && (address <= stm8->optionend)) + retval = stm8_write_flash(target, OPTION, address, size, count, 0, buffer); + else + retval = stm8_adapter_write_memory(target, address, size, count, + buffer); + + if (retval != ERROR_OK) + return ERROR_TARGET_FAILURE; + + return retval; +} + +static int stm8_read_memory(struct target *target, target_addr_t address, + uint32_t size, uint32_t count, uint8_t *buffer) +{ + LOG_DEBUG("address: 0x%8.8" TARGET_PRIxADDR + ", size: 0x%8.8" PRIx32 + ", count: 0x%8.8" PRIx32, + address, size, count); + + if (target->state != TARGET_HALTED) + LOG_WARNING("target not halted"); + + int retval; + retval = stm8_adapter_read_memory(target, address, size, count, buffer); + + if (retval != ERROR_OK) + return ERROR_TARGET_FAILURE; + + return retval; +} + +static int stm8_init(struct command_context *cmd_ctx, struct target *target) +{ + stm8_build_reg_cache(target); + + return ERROR_OK; +} + +static int stm8_poll(struct target *target) +{ + int retval = ERROR_OK; + uint8_t csr1, csr2; + +#ifdef LOG_STM8 + LOG_DEBUG("target->state=%d", target->state); +#endif + + /* read dm_csrx control regs */ + retval = stm8_read_dm_csrx(target, &csr1, &csr2); + if (retval != ERROR_OK) { + LOG_DEBUG("stm8_read_dm_csrx failed retval=%d", retval); + /* + We return ERROR_OK here even if we didn't get an answer. + openocd will call target_wait_state until we get target state TARGET_HALTED + */ + return ERROR_OK; + } + + /* check for processor halted */ + if (csr2 & STALL) { + if (target->state != TARGET_HALTED) { + if (target->state == TARGET_UNKNOWN) + LOG_DEBUG("DM_CSR2_STALL already set during server startup."); + + retval = stm8_debug_entry(target); + if (retval != ERROR_OK) { + LOG_DEBUG("stm8_debug_entry failed retval=%d", retval); + return ERROR_TARGET_FAILURE; + } + + if (target->state == TARGET_DEBUG_RUNNING) { + target->state = TARGET_HALTED; + target_call_event_callbacks(target, TARGET_EVENT_DEBUG_HALTED); + } else { + target->state = TARGET_HALTED; + target_call_event_callbacks(target, TARGET_EVENT_HALTED); + } + } + } else + target->state = TARGET_RUNNING; +#ifdef LOG_STM8 + LOG_DEBUG("csr1 = 0x%02X csr2 = 0x%02X", csr1, csr2); +#endif + return ERROR_OK; +} + +static int stm8_halt(struct target *target) +{ + LOG_DEBUG("target->state: %s", target_state_name(target)); + + if (target->state == TARGET_HALTED) { + LOG_DEBUG("target was already halted"); + return ERROR_OK; + } + + if (target->state == TARGET_UNKNOWN) + LOG_WARNING("target was in unknown state when halt was requested"); + + if (target->state == TARGET_RESET) { + /* we came here in a reset_halt or reset_init sequence + * debug entry was already prepared in stm8_assert_reset() + */ + target->debug_reason = DBG_REASON_DBGRQ; + + return ERROR_OK; + } + + + /* break processor */ + stm8_debug_stall(target); + + target->debug_reason = DBG_REASON_DBGRQ; + + return ERROR_OK; +} + +static int stm8_reset_assert(struct target *target) +{ + int res = ERROR_OK; + struct hl_interface_s *adapter = target_to_adapter(target); + struct stm8_common *stm8 = target_to_stm8(target); + bool use_srst_fallback = true; + + enum reset_types jtag_reset_config = jtag_get_reset_config(); + + if (jtag_reset_config & RESET_HAS_SRST) { + jtag_add_reset(0, 1); + res = adapter->layout->api->assert_srst(adapter->handle, 0); + + if (res == ERROR_OK) + /* hardware srst supported */ + use_srst_fallback = false; + else if (res != ERROR_COMMAND_NOTFOUND) + /* some other failure */ + return res; + } + + if (use_srst_fallback) { + LOG_DEBUG("Hardware srst not supported, falling back to swim reset"); + res = adapter->layout->api->reset(adapter->handle); + if (res != ERROR_OK) + return res; + } + + /* registers are now invalid */ + register_cache_invalidate(stm8->core_cache); + + target->state = TARGET_RESET; + target->debug_reason = DBG_REASON_NOTHALTED; + + if (target->reset_halt) { + res = target_halt(target); + if (res != ERROR_OK) + return res; + } + + return ERROR_OK; +} + +static int stm8_reset_deassert(struct target *target) +{ + int res; + struct hl_interface_s *adapter = target_to_adapter(target); + + enum reset_types jtag_reset_config = jtag_get_reset_config(); + + if (jtag_reset_config & RESET_HAS_SRST) { + res = adapter->layout->api->assert_srst(adapter->handle, 1); + if ((res != ERROR_OK) && (res != ERROR_COMMAND_NOTFOUND)) + return res; + } + + /* virtual deassert reset, we need it for the internal + * jtag state machine + */ + jtag_add_reset(0, 0); + + /* The cpu should now be stalled. If halt was requested + let poll detect the stall */ + if (target->reset_halt) + return ERROR_OK; + + /* Instead of going thrugh saving context, polling and + then resuming target again just clear stall and proceed. */ + target->state = TARGET_RUNNING; + return stm8_exit_debug(target); +} + +/* stm8_single_step_core() is only used for stepping over breakpoints + from stm8_resume() */ +static int stm8_single_step_core(struct target *target) +{ + struct stm8_common *stm8 = target_to_stm8(target); + + /* configure single step mode */ + stm8_config_step(target, 1); + + /* disable interrupts while stepping */ + if (!stm8->enable_step_irq) + stm8_enable_interrupts(target, 0); + + /* exit debug mode */ + stm8_exit_debug(target); + + stm8_debug_entry(target); + + return ERROR_OK; +} + +static int stm8_resume(struct target *target, int current, + target_addr_t address, int handle_breakpoints, + int debug_execution) +{ + struct stm8_common *stm8 = target_to_stm8(target); + struct breakpoint *breakpoint = NULL; + uint32_t resume_pc; + + LOG_DEBUG("%d " TARGET_ADDR_FMT " %d %d", current, address, + handle_breakpoints, debug_execution); + + if (target->state != TARGET_HALTED) { + LOG_WARNING("target not halted"); + return ERROR_TARGET_NOT_HALTED; + } + + if (!debug_execution) { + target_free_all_working_areas(target); + stm8_enable_breakpoints(target); + stm8_enable_watchpoints(target); + struct stm8_comparator *comparator_list = stm8->hw_break_list; + stm8_set_hwbreak(target, comparator_list); + } + + /* current = 1: continue on current pc, + otherwise continue at
*/ + if (!current) { + buf_set_u32(stm8->core_cache->reg_list[STM8_PC].value, + 0, 32, address); + stm8->core_cache->reg_list[STM8_PC].dirty = true; + stm8->core_cache->reg_list[STM8_PC].valid = true; + } + + if (!current) + resume_pc = address; + else + resume_pc = buf_get_u32( + stm8->core_cache->reg_list[STM8_PC].value, + 0, 32); + + stm8_restore_context(target); + + /* the front-end may request us not to handle breakpoints */ + if (handle_breakpoints) { + /* Single step past breakpoint at current address */ + breakpoint = breakpoint_find(target, resume_pc); + if (breakpoint) { + LOG_DEBUG("unset breakpoint at " TARGET_ADDR_FMT, + breakpoint->address); + stm8_unset_breakpoint(target, breakpoint); + stm8_single_step_core(target); + stm8_set_breakpoint(target, breakpoint); + } + } + + /* disable interrupts if we are debugging */ + if (debug_execution) + stm8_enable_interrupts(target, 0); + + /* exit debug mode */ + stm8_exit_debug(target); + target->debug_reason = DBG_REASON_NOTHALTED; + + /* registers are now invalid */ + register_cache_invalidate(stm8->core_cache); + + if (!debug_execution) { + target->state = TARGET_RUNNING; + target_call_event_callbacks(target, TARGET_EVENT_RESUMED); + LOG_DEBUG("target resumed at 0x%" PRIx32 "", resume_pc); + } else { + target->state = TARGET_DEBUG_RUNNING; + target_call_event_callbacks(target, TARGET_EVENT_DEBUG_RESUMED); + LOG_DEBUG("target debug resumed at 0x%" PRIx32 "", resume_pc); + } + + return ERROR_OK; +} + +static int stm8_init_flash_regs(bool enable_stm8l, struct stm8_common *stm8) +{ + stm8->enable_stm8l = enable_stm8l; + + if (stm8->enable_stm8l) { + stm8->flash_cr2 = FLASH_CR2_STM8L; + stm8->flash_ncr2 = FLASH_NCR2_STM8L; + stm8->flash_iapsr = FLASH_IAPSR_STM8L; + stm8->flash_dukr = FLASH_DUKR_STM8L; + stm8->flash_pukr = FLASH_PUKR_STM8L; + } else { + stm8->flash_cr2 = FLASH_CR2_STM8S; + stm8->flash_ncr2 = FLASH_NCR2_STM8S; + stm8->flash_iapsr = FLASH_IAPSR_STM8S; + stm8->flash_dukr = FLASH_DUKR_STM8S; + stm8->flash_pukr = FLASH_PUKR_STM8S; + } + return ERROR_OK; +} + +static int stm8_init_arch_info(struct target *target, + struct stm8_common *stm8, struct jtag_tap *tap) +{ + target->endianness = TARGET_BIG_ENDIAN; + target->arch_info = stm8; + stm8->common_magic = STM8_COMMON_MAGIC; + stm8->fast_data_area = NULL; + stm8->blocksize = 0x80; + stm8->flashstart = 0x8000; + stm8->flashend = 0xffff; + stm8->eepromstart = 0x4000; + stm8->eepromend = 0x43ff; + stm8->optionstart = 0x4800; + stm8->optionend = 0x487F; + + /* has breakpoint/watchpoint unit been scanned */ + stm8->bp_scanned = false; + stm8->hw_break_list = NULL; + + stm8->read_core_reg = stm8_read_core_reg; + stm8->write_core_reg = stm8_write_core_reg; + + stm8_init_flash_regs(0, stm8); + + return ERROR_OK; +} + +static int stm8_target_create(struct target *target, + Jim_Interp *interp) +{ + + struct stm8_common *stm8 = calloc(1, sizeof(struct stm8_common)); + + stm8_init_arch_info(target, stm8, target->tap); + stm8_configure_break_unit(target); + + return ERROR_OK; +} + +static int stm8_read_core_reg(struct target *target, unsigned int num) +{ + uint32_t reg_value; + + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + + if (num >= STM8_NUM_REGS) + return ERROR_COMMAND_SYNTAX_ERROR; + + reg_value = stm8->core_regs[num]; + LOG_DEBUG("read core reg %i value 0x%" PRIx32 "", num , reg_value); + buf_set_u32(stm8->core_cache->reg_list[num].value, 0, 32, reg_value); + stm8->core_cache->reg_list[num].valid = true; + stm8->core_cache->reg_list[num].dirty = false; + + return ERROR_OK; +} + +static int stm8_write_core_reg(struct target *target, unsigned int num) +{ + uint32_t reg_value; + + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + + if (num >= STM8_NUM_REGS) + return ERROR_COMMAND_SYNTAX_ERROR; + + reg_value = buf_get_u32(stm8->core_cache->reg_list[num].value, 0, 32); + stm8->core_regs[num] = reg_value; + LOG_DEBUG("write core reg %i value 0x%" PRIx32 "", num , reg_value); + stm8->core_cache->reg_list[num].valid = true; + stm8->core_cache->reg_list[num].dirty = false; + + return ERROR_OK; +} + +static int stm8_get_gdb_reg_list(struct target *target, struct reg **reg_list[], + int *reg_list_size, enum target_register_class reg_class) +{ + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + unsigned int i; + + *reg_list_size = STM8_NUM_REGS; + *reg_list = malloc(sizeof(struct reg *) * (*reg_list_size)); + + for (i = 0; i < STM8_NUM_REGS; i++) + (*reg_list)[i] = &stm8->core_cache->reg_list[i]; + + return ERROR_OK; +} + +static const struct reg_arch_type stm8_reg_type = { + .get = stm8_get_core_reg, + .set = stm8_set_core_reg, +}; + +static struct reg_cache *stm8_build_reg_cache(struct target *target) +{ + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + + int num_regs = STM8_NUM_REGS; + struct reg_cache **cache_p = register_get_last_cache_p(&target->reg_cache); + struct reg_cache *cache = malloc(sizeof(struct reg_cache)); + struct reg *reg_list = calloc(num_regs, sizeof(struct reg)); + struct stm8_core_reg *arch_info = malloc( + sizeof(struct stm8_core_reg) * num_regs); + struct reg_feature *feature; + int i; + + /* Build the process context cache */ + cache->name = "stm8 registers"; + cache->next = NULL; + cache->reg_list = reg_list; + cache->num_regs = num_regs; + (*cache_p) = cache; + stm8->core_cache = cache; + + for (i = 0; i < num_regs; i++) { + arch_info[i].num = stm8_regs[i].id; + arch_info[i].target = target; + arch_info[i].stm8_common = stm8; + + reg_list[i].name = stm8_regs[i].name; + reg_list[i].size = stm8_regs[i].bits; + + reg_list[i].value = calloc(1, 4); + reg_list[i].valid = false; + reg_list[i].type = &stm8_reg_type; + reg_list[i].arch_info = &arch_info[i]; + + reg_list[i].reg_data_type = calloc(1, sizeof(struct reg_data_type)); + if (reg_list[i].reg_data_type) + reg_list[i].reg_data_type->type = stm8_regs[i].type; + else { + LOG_ERROR("unable to allocate reg type list"); + return NULL; + } + + reg_list[i].dirty = false; + reg_list[i].group = stm8_regs[i].group; + reg_list[i].number = stm8_regs[i].id; + reg_list[i].exist = true; + reg_list[i].caller_save = true; /* gdb defaults to true */ + + feature = calloc(1, sizeof(struct reg_feature)); + if (feature) { + feature->name = stm8_regs[i].feature; + reg_list[i].feature = feature; + } else + LOG_ERROR("unable to allocate feature list"); + } + + return cache; +} + +static void stm8_free_reg_cache(struct target *target) +{ + struct stm8_common *stm8 = target_to_stm8(target); + struct reg_cache *cache; + struct reg *reg; + unsigned int i; + + cache = stm8->core_cache; + + if (!cache) + return; + + for (i = 0; i < cache->num_regs; i++) { + reg = &cache->reg_list[i]; + + free(reg->feature); + free(reg->reg_data_type); + free(reg->value); + } + + free(cache->reg_list[0].arch_info); + free(cache->reg_list); + free(cache); + + stm8->core_cache = NULL; +} + +static void stm8_deinit(struct target *target) +{ + struct stm8_common *stm8 = target_to_stm8(target); + + free(stm8->hw_break_list); + + stm8_free_reg_cache(target); + + free(stm8); +} + +static int stm8_arch_state(struct target *target) +{ + struct stm8_common *stm8 = target_to_stm8(target); + + LOG_USER("target halted due to %s, pc: 0x%8.8" PRIx32 "", + debug_reason_name(target), + buf_get_u32(stm8->core_cache->reg_list[STM8_PC].value, 0, 32)); + + return ERROR_OK; +} + +static int stm8_step(struct target *target, int current, + target_addr_t address, int handle_breakpoints) +{ + LOG_DEBUG("%" PRIx32 " " TARGET_ADDR_FMT " %" PRIx32, + current, address, handle_breakpoints); + + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + struct breakpoint *breakpoint = NULL; + + if (target->state != TARGET_HALTED) { + LOG_WARNING("target not halted"); + return ERROR_TARGET_NOT_HALTED; + } + + /* current = 1: continue on current pc, otherwise continue at
*/ + if (!current) { + buf_set_u32(stm8->core_cache->reg_list[STM8_PC].value, 0, 32, address); + stm8->core_cache->reg_list[STM8_PC].dirty = true; + stm8->core_cache->reg_list[STM8_PC].valid = true; + } + + /* the front-end may request us not to handle breakpoints */ + if (handle_breakpoints) { + breakpoint = breakpoint_find(target, + buf_get_u32(stm8->core_cache->reg_list[STM8_PC].value, 0, 32)); + if (breakpoint) + stm8_unset_breakpoint(target, breakpoint); + } + + /* restore context */ + stm8_restore_context(target); + + /* configure single step mode */ + stm8_config_step(target, 1); + + target->debug_reason = DBG_REASON_SINGLESTEP; + + target_call_event_callbacks(target, TARGET_EVENT_RESUMED); + + /* disable interrupts while stepping */ + if (!stm8->enable_step_irq) + stm8_enable_interrupts(target, 0); + + /* exit debug mode */ + stm8_exit_debug(target); + + /* registers are now invalid */ + register_cache_invalidate(stm8->core_cache); + + LOG_DEBUG("target stepped "); + stm8_debug_entry(target); + + if (breakpoint) + stm8_set_breakpoint(target, breakpoint); + + target_call_event_callbacks(target, TARGET_EVENT_HALTED); + + return ERROR_OK; +} + +static void stm8_enable_breakpoints(struct target *target) +{ + struct breakpoint *breakpoint = target->breakpoints; + + /* set any pending breakpoints */ + while (breakpoint) { + if (breakpoint->set == 0) + stm8_set_breakpoint(target, breakpoint); + breakpoint = breakpoint->next; + } +} + +static int stm8_set_breakpoint(struct target *target, + struct breakpoint *breakpoint) +{ + struct stm8_common *stm8 = target_to_stm8(target); + struct stm8_comparator *comparator_list = stm8->hw_break_list; + int retval; + + if (breakpoint->set) { + LOG_WARNING("breakpoint already set"); + return ERROR_OK; + } + + if (breakpoint->type == BKPT_HARD) { + int bp_num = 0; + + while (comparator_list[bp_num].used && (bp_num < stm8->num_hw_bpoints)) + bp_num++; + if (bp_num >= stm8->num_hw_bpoints) { + LOG_ERROR("Can not find free breakpoint register (bpid: %" PRIu32 ")", + breakpoint->unique_id); + return ERROR_TARGET_RESOURCE_NOT_AVAILABLE; + } + breakpoint->set = bp_num + 1; + comparator_list[bp_num].used = true; + comparator_list[bp_num].bp_value = breakpoint->address; + comparator_list[bp_num].type = HWBRK_EXEC; + + retval = stm8_set_hwbreak(target, comparator_list); + if (retval != ERROR_OK) + return retval; + + LOG_DEBUG("bpid: %" PRIu32 ", bp_num %i bp_value 0x%" PRIx32 "", + breakpoint->unique_id, + bp_num, comparator_list[bp_num].bp_value); + } else if (breakpoint->type == BKPT_SOFT) { + LOG_DEBUG("bpid: %" PRIu32, breakpoint->unique_id); + if (breakpoint->length == 1) { + uint8_t verify = 0x55; + + retval = target_read_u8(target, breakpoint->address, + breakpoint->orig_instr); + if (retval != ERROR_OK) + return retval; + retval = target_write_u8(target, breakpoint->address, STM8_BREAK); + if (retval != ERROR_OK) + return retval; + + retval = target_read_u8(target, breakpoint->address, &verify); + if (retval != ERROR_OK) + return retval; + if (verify != STM8_BREAK) { + LOG_ERROR("Unable to set breakpoint at address " TARGET_ADDR_FMT + " - check that memory is read/writable", + breakpoint->address); + return ERROR_TARGET_RESOURCE_NOT_AVAILABLE; + } + } else { + return ERROR_TARGET_RESOURCE_NOT_AVAILABLE; + } + breakpoint->set = 1; /* Any nice value but 0 */ + } + + return ERROR_OK; +} + +static int stm8_add_breakpoint(struct target *target, + struct breakpoint *breakpoint) +{ + struct stm8_common *stm8 = target_to_stm8(target); + int ret; + + if (breakpoint->type == BKPT_HARD) { + if (stm8->num_hw_bpoints_avail < 1) { + LOG_INFO("no hardware breakpoint available"); + return ERROR_TARGET_RESOURCE_NOT_AVAILABLE; + } + + ret = stm8_set_breakpoint(target, breakpoint); + if (ret != ERROR_OK) + return ret; + + stm8->num_hw_bpoints_avail--; + return ERROR_OK; + } + + ret = stm8_set_breakpoint(target, breakpoint); + if (ret != ERROR_OK) + return ret; + + return ERROR_OK; +} + +static int stm8_unset_breakpoint(struct target *target, + struct breakpoint *breakpoint) +{ + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + struct stm8_comparator *comparator_list = stm8->hw_break_list; + int retval; + + if (!breakpoint->set) { + LOG_WARNING("breakpoint not set"); + return ERROR_OK; + } + + if (breakpoint->type == BKPT_HARD) { + int bp_num = breakpoint->set - 1; + if ((bp_num < 0) || (bp_num >= stm8->num_hw_bpoints)) { + LOG_DEBUG("Invalid comparator number in breakpoint (bpid: %" PRIu32 ")", + breakpoint->unique_id); + return ERROR_OK; + } + LOG_DEBUG("bpid: %" PRIu32 " - releasing hw: %d", + breakpoint->unique_id, + bp_num); + comparator_list[bp_num].used = false; + retval = stm8_set_hwbreak(target, comparator_list); + if (retval != ERROR_OK) + return retval; + } else { + /* restore original instruction (kept in target endianness) */ + LOG_DEBUG("bpid: %" PRIu32, breakpoint->unique_id); + if (breakpoint->length == 1) { + uint8_t current_instr; + + /* check that user program has not + modified breakpoint instruction */ + retval = target_read_memory(target, breakpoint->address, 1, 1, + (uint8_t *)¤t_instr); + if (retval != ERROR_OK) + return retval; + + if (current_instr == STM8_BREAK) { + retval = target_write_memory(target, breakpoint->address, 1, 1, + breakpoint->orig_instr); + if (retval != ERROR_OK) + return retval; + } + } else + return ERROR_FAIL; + } + breakpoint->set = 0; + + return ERROR_OK; +} + +static int stm8_remove_breakpoint(struct target *target, + struct breakpoint *breakpoint) +{ + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + + if (target->state != TARGET_HALTED) { + LOG_WARNING("target not halted"); + return ERROR_TARGET_NOT_HALTED; + } + + if (breakpoint->set) + stm8_unset_breakpoint(target, breakpoint); + + if (breakpoint->type == BKPT_HARD) + stm8->num_hw_bpoints_avail++; + + return ERROR_OK; +} + +static int stm8_set_watchpoint(struct target *target, + struct watchpoint *watchpoint) +{ + struct stm8_common *stm8 = target_to_stm8(target); + struct stm8_comparator *comparator_list = stm8->hw_break_list; + int wp_num = 0; + int ret; + + if (watchpoint->set) { + LOG_WARNING("watchpoint already set"); + return ERROR_OK; + } + + while (comparator_list[wp_num].used && (wp_num < stm8->num_hw_bpoints)) + wp_num++; + if (wp_num >= stm8->num_hw_bpoints) { + LOG_ERROR("Can not find free hw breakpoint"); + return ERROR_TARGET_RESOURCE_NOT_AVAILABLE; + } + + if (watchpoint->length != 1) { + LOG_ERROR("Only watchpoints of length 1 are supported"); + return ERROR_TARGET_UNALIGNED_ACCESS; + } + + enum hw_break_type enable = 0; + + switch (watchpoint->rw) { + case WPT_READ: + enable = HWBRK_RD; + break; + case WPT_WRITE: + enable = HWBRK_WR; + break; + case WPT_ACCESS: + enable = HWBRK_ACC; + break; + default: + LOG_ERROR("BUG: watchpoint->rw neither read, write nor access"); + } + + comparator_list[wp_num].used = true; + comparator_list[wp_num].bp_value = watchpoint->address; + comparator_list[wp_num].type = enable; + + ret = stm8_set_hwbreak(target, comparator_list); + if (ret != ERROR_OK) { + comparator_list[wp_num].used = false; + return ret; + } + + watchpoint->set = wp_num + 1; + + LOG_DEBUG("wp_num %i bp_value 0x%" PRIx32 "", + wp_num, + comparator_list[wp_num].bp_value); + + return ERROR_OK; +} + +static int stm8_add_watchpoint(struct target *target, + struct watchpoint *watchpoint) +{ + int ret; + struct stm8_common *stm8 = target_to_stm8(target); + + if (stm8->num_hw_bpoints_avail < 1) { + LOG_INFO("no hardware watchpoints available"); + return ERROR_TARGET_RESOURCE_NOT_AVAILABLE; + } + + ret = stm8_set_watchpoint(target, watchpoint); + if (ret != ERROR_OK) + return ret; + + stm8->num_hw_bpoints_avail--; + return ERROR_OK; +} + +static void stm8_enable_watchpoints(struct target *target) +{ + struct watchpoint *watchpoint = target->watchpoints; + + /* set any pending watchpoints */ + while (watchpoint) { + if (watchpoint->set == 0) + stm8_set_watchpoint(target, watchpoint); + watchpoint = watchpoint->next; + } +} + +static int stm8_unset_watchpoint(struct target *target, + struct watchpoint *watchpoint) +{ + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + struct stm8_comparator *comparator_list = stm8->hw_break_list; + + if (!watchpoint->set) { + LOG_WARNING("watchpoint not set"); + return ERROR_OK; + } + + int wp_num = watchpoint->set - 1; + if ((wp_num < 0) || (wp_num >= stm8->num_hw_bpoints)) { + LOG_DEBUG("Invalid hw comparator number in watchpoint"); + return ERROR_OK; + } + comparator_list[wp_num].used = false; + watchpoint->set = 0; + + stm8_set_hwbreak(target, comparator_list); + + return ERROR_OK; +} + +static int stm8_remove_watchpoint(struct target *target, + struct watchpoint *watchpoint) +{ + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + + if (target->state != TARGET_HALTED) { + LOG_WARNING("target not halted"); + return ERROR_TARGET_NOT_HALTED; + } + + if (watchpoint->set) + stm8_unset_watchpoint(target, watchpoint); + + stm8->num_hw_bpoints_avail++; + + return ERROR_OK; +} + +static int stm8_examine(struct target *target) +{ + int retval; + uint8_t csr1, csr2; + /* get pointers to arch-specific information */ + struct stm8_common *stm8 = target_to_stm8(target); + struct hl_interface_s *adapter = target_to_adapter(target); + + if (!target_was_examined(target)) { + if (!stm8->swim_configured) { + /* set SWIM_CSR = 0xa0 (enable mem access & mask reset) */ + LOG_DEBUG("writing A0 to SWIM_CSR (SAFE_MASK + SWIM_DM)"); + retval = stm8_write_u8(target, SWIM_CSR, SAFE_MASK + SWIM_DM); + if (retval != ERROR_OK) + return retval; + /* set high speed */ + LOG_DEBUG("writing B0 to SWIM_CSR (SAFE_MASK + SWIM_DM + HS)"); + retval = stm8_write_u8(target, SWIM_CSR, SAFE_MASK + SWIM_DM + HS); + if (retval != ERROR_OK) + return retval; + retval = stm8_set_speed(target, 1); + if (retval == ERROR_OK) + stm8->swim_configured = true; + /* + Now is the time to deassert reset if connect_under_reset. + Releasing reset line will cause the option bytes to load. + The core will still be stalled. + */ + if (adapter->param.connect_under_reset) + stm8_reset_deassert(target); + } else { + LOG_INFO("trying to reconnect"); + + retval = adapter->layout->api->state(adapter->handle); + if (retval != ERROR_OK) { + LOG_ERROR("reconnect failed"); + return ERROR_FAIL; + } + + /* read dm_csrx control regs */ + retval = stm8_read_dm_csrx(target, &csr1, &csr2); + if (retval != ERROR_OK) { + LOG_ERROR("state query failed"); + return ERROR_FAIL; + } + } + + target_set_examined(target); + + return ERROR_OK; + } + + return ERROR_OK; +} + +/** Checks whether a memory region is erased. */ +static int stm8_blank_check_memory(struct target *target, + target_addr_t address, uint32_t count, uint32_t *blank, uint8_t erased_value) +{ + struct working_area *erase_check_algorithm; + struct reg_param reg_params[2]; + struct mem_param mem_params[2]; + struct stm8_algorithm stm8_info; + + static const uint8_t stm8_erase_check_code[] = { +#include "../../contrib/loaders/erase_check/stm8_erase_check.inc" + }; + + if (erased_value != 0xff) { + LOG_ERROR("Erase value 0x%02" PRIx8 " not yet supported for STM8", + erased_value); + return ERROR_FAIL; + } + + /* make sure we have a working area */ + if (target_alloc_working_area(target, sizeof(stm8_erase_check_code), + &erase_check_algorithm) != ERROR_OK) + return ERROR_TARGET_RESOURCE_NOT_AVAILABLE; + + target_write_buffer(target, erase_check_algorithm->address, + sizeof(stm8_erase_check_code), stm8_erase_check_code); + + stm8_info.common_magic = STM8_COMMON_MAGIC; + + init_mem_param(&mem_params[0], 0x0, 3, PARAM_OUT); + buf_set_u32(mem_params[0].value, 0, 24, address); + + init_mem_param(&mem_params[1], 0x3, 3, PARAM_OUT); + buf_set_u32(mem_params[1].value, 0, 24, count); + + init_reg_param(®_params[0], "a", 32, PARAM_IN_OUT); + buf_set_u32(reg_params[0].value, 0, 32, erased_value); + + init_reg_param(®_params[1], "sp", 32, PARAM_OUT); + buf_set_u32(reg_params[1].value, 0, 32, erase_check_algorithm->address); + + int retval = target_run_algorithm(target, 2, mem_params, 2, reg_params, + erase_check_algorithm->address + 6, + erase_check_algorithm->address + (sizeof(stm8_erase_check_code) - 1), + 10000, &stm8_info); + + if (retval == ERROR_OK) + *blank = (*(reg_params[0].value) == 0xff); + + destroy_mem_param(&mem_params[0]); + destroy_mem_param(&mem_params[1]); + destroy_reg_param(®_params[0]); + + target_free_working_area(target, erase_check_algorithm); + + return retval; +} + +static int stm8_checksum_memory(struct target *target, target_addr_t address, + uint32_t count, uint32_t *checksum) +{ + /* let image_calculate_checksum() take care of business */ + return ERROR_TARGET_RESOURCE_NOT_AVAILABLE; +} + +/* run to exit point. return error if exit point was not reached. */ +static int stm8_run_and_wait(struct target *target, uint32_t entry_point, + int timeout_ms, uint32_t exit_point, struct stm8_common *stm8) +{ + uint32_t pc; + int retval; + /* This code relies on the target specific resume() and + poll()->debug_entry() sequence to write register values to the + processor and the read them back */ + retval = target_resume(target, 0, entry_point, 0, 1); + if (retval != ERROR_OK) + return retval; + + retval = target_wait_state(target, TARGET_HALTED, timeout_ms); + /* If the target fails to halt due to the breakpoint, force a halt */ + if (retval != ERROR_OK || target->state != TARGET_HALTED) { + retval = target_halt(target); + if (retval != ERROR_OK) + return retval; + retval = target_wait_state(target, TARGET_HALTED, 500); + if (retval != ERROR_OK) + return retval; + return ERROR_TARGET_TIMEOUT; + } + + pc = buf_get_u32(stm8->core_cache->reg_list[STM8_PC].value, 0, 32); + if (exit_point && (pc != exit_point)) { + LOG_DEBUG("failed algorithm halted at 0x%" PRIx32 " ", pc); + return ERROR_TARGET_TIMEOUT; + } + + return ERROR_OK; +} + +static int stm8_run_algorithm(struct target *target, int num_mem_params, + struct mem_param *mem_params, int num_reg_params, + struct reg_param *reg_params, target_addr_t entry_point, + target_addr_t exit_point, int timeout_ms, void *arch_info) +{ + struct stm8_common *stm8 = target_to_stm8(target); + + uint32_t context[STM8_NUM_REGS]; + int retval = ERROR_OK; + + LOG_DEBUG("Running algorithm"); + + /* NOTE: stm8_run_algorithm requires that each + algorithm uses a software breakpoint + at the exit point */ + + if (stm8->common_magic != STM8_COMMON_MAGIC) { + LOG_ERROR("current target isn't a STM8 target"); + return ERROR_TARGET_INVALID; + } + + if (target->state != TARGET_HALTED) { + LOG_WARNING("target not halted"); + return ERROR_TARGET_NOT_HALTED; + } + + /* refresh core register cache */ + for (unsigned int i = 0; i < STM8_NUM_REGS; i++) { + if (!stm8->core_cache->reg_list[i].valid) + stm8->read_core_reg(target, i); + context[i] = buf_get_u32(stm8->core_cache->reg_list[i].value, 0, 32); + } + + for (int i = 0; i < num_mem_params; i++) { + retval = target_write_buffer(target, mem_params[i].address, + mem_params[i].size, mem_params[i].value); + if (retval != ERROR_OK) + return retval; + } + + for (int i = 0; i < num_reg_params; i++) { + struct reg *reg = register_get_by_name(stm8->core_cache, + reg_params[i].reg_name, 0); + + if (!reg) { + LOG_ERROR("BUG: register '%s' not found", reg_params[i].reg_name); + return ERROR_COMMAND_SYNTAX_ERROR; + } + + if (reg_params[i].size != 32) { + LOG_ERROR("BUG: register '%s' size doesn't match reg_params[i].size", + reg_params[i].reg_name); + return ERROR_COMMAND_SYNTAX_ERROR; + } + + stm8_set_core_reg(reg, reg_params[i].value); + } + + retval = stm8_run_and_wait(target, entry_point, + timeout_ms, exit_point, stm8); + + if (retval != ERROR_OK) + return retval; + + for (int i = 0; i < num_mem_params; i++) { + if (mem_params[i].direction != PARAM_OUT) { + retval = target_read_buffer(target, mem_params[i].address, + mem_params[i].size, mem_params[i].value); + if (retval != ERROR_OK) + return retval; + } + } + + for (int i = 0; i < num_reg_params; i++) { + if (reg_params[i].direction != PARAM_OUT) { + struct reg *reg = register_get_by_name(stm8->core_cache, + reg_params[i].reg_name, 0); + if (!reg) { + LOG_ERROR("BUG: register '%s' not found", + reg_params[i].reg_name); + return ERROR_COMMAND_SYNTAX_ERROR; + } + + if (reg_params[i].size != 32) { + LOG_ERROR("BUG: register '%s' size doesn't match reg_params[i].size", + reg_params[i].reg_name); + return ERROR_COMMAND_SYNTAX_ERROR; + } + + buf_set_u32(reg_params[i].value, + 0, 32, buf_get_u32(reg->value, 0, 32)); + } + } + + /* restore everything we saved before */ + for (unsigned int i = 0; i < STM8_NUM_REGS; i++) { + uint32_t regvalue; + regvalue = buf_get_u32(stm8->core_cache->reg_list[i].value, 0, 32); + if (regvalue != context[i]) { + LOG_DEBUG("restoring register %s with value 0x%8.8" PRIx32, + stm8->core_cache->reg_list[i].name, context[i]); + buf_set_u32(stm8->core_cache->reg_list[i].value, + 0, 32, context[i]); + stm8->core_cache->reg_list[i].valid = true; + stm8->core_cache->reg_list[i].dirty = true; + } + } + + return ERROR_OK; +} + +int stm8_jim_configure(struct target *target, Jim_GetOptInfo *goi) +{ + struct stm8_common *stm8 = target_to_stm8(target); + jim_wide w; + int e; + const char *arg; + + arg = Jim_GetString(goi->argv[0], NULL); + if (!strcmp(arg, "-blocksize")) { + e = Jim_GetOpt_String(goi, &arg, NULL); + if (e != JIM_OK) + return e; + + if (goi->argc == 0) { + Jim_WrongNumArgs(goi->interp, goi->argc, goi->argv, + "-blocksize ?bytes? ..."); + return JIM_ERR; + } + + e = Jim_GetOpt_Wide(goi, &w); + if (e != JIM_OK) + return e; + + stm8->blocksize = w; + LOG_DEBUG("blocksize=%8.8x", stm8->blocksize); + return JIM_OK; + } + if (!strcmp(arg, "-flashstart")) { + e = Jim_GetOpt_String(goi, &arg, NULL); + if (e != JIM_OK) + return e; + + if (goi->argc == 0) { + Jim_WrongNumArgs(goi->interp, goi->argc, goi->argv, + "-flashstart ?address? ..."); + return JIM_ERR; + } + + e = Jim_GetOpt_Wide(goi, &w); + if (e != JIM_OK) + return e; + + stm8->flashstart = w; + LOG_DEBUG("flashstart=%8.8x", stm8->flashstart); + return JIM_OK; + } + if (!strcmp(arg, "-flashend")) { + e = Jim_GetOpt_String(goi, &arg, NULL); + if (e != JIM_OK) + return e; + + if (goi->argc == 0) { + Jim_WrongNumArgs(goi->interp, goi->argc, goi->argv, + "-flashend ?address? ..."); + return JIM_ERR; + } + + e = Jim_GetOpt_Wide(goi, &w); + if (e != JIM_OK) + return e; + + stm8->flashend = w; + LOG_DEBUG("flashend=%8.8x", stm8->flashend); + return JIM_OK; + } + if (!strcmp(arg, "-eepromstart")) { + e = Jim_GetOpt_String(goi, &arg, NULL); + if (e != JIM_OK) + return e; + + if (goi->argc == 0) { + Jim_WrongNumArgs(goi->interp, goi->argc, goi->argv, + "-eepromstart ?address? ..."); + return JIM_ERR; + } + + e = Jim_GetOpt_Wide(goi, &w); + if (e != JIM_OK) + return e; + + stm8->eepromstart = w; + LOG_DEBUG("eepromstart=%8.8x", stm8->eepromstart); + return JIM_OK; + } + if (!strcmp(arg, "-eepromend")) { + e = Jim_GetOpt_String(goi, &arg, NULL); + if (e != JIM_OK) + return e; + + if (goi->argc == 0) { + Jim_WrongNumArgs(goi->interp, goi->argc, goi->argv, + "-eepromend ?address? ..."); + return JIM_ERR; + } + + e = Jim_GetOpt_Wide(goi, &w); + if (e != JIM_OK) + return e; + + stm8->eepromend = w; + LOG_DEBUG("eepromend=%8.8x", stm8->eepromend); + return JIM_OK; + } + if (!strcmp(arg, "-optionstart")) { + e = Jim_GetOpt_String(goi, &arg, NULL); + if (e != JIM_OK) + return e; + + if (goi->argc == 0) { + Jim_WrongNumArgs(goi->interp, goi->argc, goi->argv, + "-optionstart ?address? ..."); + return JIM_ERR; + } + + e = Jim_GetOpt_Wide(goi, &w); + if (e != JIM_OK) + return e; + + stm8->optionstart = w; + LOG_DEBUG("optionstart=%8.8x", stm8->optionstart); + return JIM_OK; + } + if (!strcmp(arg, "-optionend")) { + e = Jim_GetOpt_String(goi, &arg, NULL); + if (e != JIM_OK) + return e; + + if (goi->argc == 0) { + Jim_WrongNumArgs(goi->interp, goi->argc, goi->argv, + "-optionend ?address? ..."); + return JIM_ERR; + } + + e = Jim_GetOpt_Wide(goi, &w); + if (e != JIM_OK) + return e; + + stm8->optionend = w; + LOG_DEBUG("optionend=%8.8x", stm8->optionend); + return JIM_OK; + } + if (!strcmp(arg, "-enable_step_irq")) { + e = Jim_GetOpt_String(goi, &arg, NULL); + if (e != JIM_OK) + return e; + + stm8->enable_step_irq = true; + LOG_DEBUG("enable_step_irq=%8.8x", stm8->enable_step_irq); + return JIM_OK; + } + if (!strcmp(arg, "-enable_stm8l")) { + e = Jim_GetOpt_String(goi, &arg, NULL); + if (e != JIM_OK) + return e; + + stm8->enable_stm8l = true; + LOG_DEBUG("enable_stm8l=%8.8x", stm8->enable_stm8l); + stm8_init_flash_regs(stm8->enable_stm8l, stm8); + return JIM_OK; + } + return JIM_CONTINUE; +} + +COMMAND_HANDLER(stm8_handle_enable_step_irq_command) +{ + const char *msg; + struct target *target = get_current_target(CMD_CTX); + struct stm8_common *stm8 = target_to_stm8(target); + bool enable = stm8->enable_step_irq; + + if (CMD_ARGC > 0) { + COMMAND_PARSE_ENABLE(CMD_ARGV[0], enable); + stm8->enable_step_irq = enable; + } + msg = stm8->enable_step_irq ? "enabled" : "disabled"; + command_print(CMD_CTX, "enable_step_irq = %s", msg); + return ERROR_OK; +} + +COMMAND_HANDLER(stm8_handle_enable_stm8l_command) +{ + const char *msg; + struct target *target = get_current_target(CMD_CTX); + struct stm8_common *stm8 = target_to_stm8(target); + bool enable = stm8->enable_stm8l; + + if (CMD_ARGC > 0) { + COMMAND_PARSE_ENABLE(CMD_ARGV[0], enable); + stm8->enable_stm8l = enable; + } + msg = stm8->enable_stm8l ? "enabled" : "disabled"; + command_print(CMD_CTX, "enable_stm8l = %s", msg); + stm8_init_flash_regs(stm8->enable_stm8l, stm8); + return ERROR_OK; +} + +static const struct command_registration stm8_exec_command_handlers[] = { + { + .name = "enable_step_irq", + .handler = stm8_handle_enable_step_irq_command, + .mode = COMMAND_ANY, + .help = "Enable/disable irq handling during step", + .usage = "[1/0]", + }, + { + .name = "enable_stm8l", + .handler = stm8_handle_enable_stm8l_command, + .mode = COMMAND_ANY, + .help = "Enable/disable STM8L flash programming", + .usage = "[1/0]", + }, + COMMAND_REGISTRATION_DONE +}; + +const struct command_registration stm8_command_handlers[] = { + { + .name = "stm8", + .mode = COMMAND_ANY, + .help = "stm8 command group", + .usage = "", + .chain = stm8_exec_command_handlers, + }, + COMMAND_REGISTRATION_DONE +}; + +struct target_type stm8_target = { + .name = "stm8", + + .poll = stm8_poll, + .arch_state = stm8_arch_state, + + .halt = stm8_halt, + .resume = stm8_resume, + .step = stm8_step, + + .assert_reset = stm8_reset_assert, + .deassert_reset = stm8_reset_deassert, + + .get_gdb_reg_list = stm8_get_gdb_reg_list, + + .read_memory = stm8_read_memory, + .write_memory = stm8_write_memory, + .checksum_memory = stm8_checksum_memory, + .blank_check_memory = stm8_blank_check_memory, + + .run_algorithm = stm8_run_algorithm, + + .add_breakpoint = stm8_add_breakpoint, + .remove_breakpoint = stm8_remove_breakpoint, + .add_watchpoint = stm8_add_watchpoint, + .remove_watchpoint = stm8_remove_watchpoint, + + .commands = stm8_command_handlers, + .target_create = stm8_target_create, + .init_target = stm8_init, + .examine = stm8_examine, + + .deinit_target = stm8_deinit, + .target_jim_configure = stm8_jim_configure, +}; diff --git a/src/target/stm8.h b/src/target/stm8.h new file mode 100644 index 000000000..39fac3e32 --- /dev/null +++ b/src/target/stm8.h @@ -0,0 +1,75 @@ +/* + OpenOCD STM8 target driver + Copyright (C) 2017 Ake Rehnman + ake.rehnman(at)gmail.com + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ + +#ifndef OPENOCD_TARGET_STM8_H +#define OPENOCD_TARGET_STM8_H + +struct target; + +#define STM8_COMMON_MAGIC 0x53544D38 +#define STM8_NUM_CORE_REGS 6 + +struct stm8_common { + uint32_t common_magic; + void *arch_info; + struct reg_cache *core_cache; + uint32_t core_regs[STM8_NUM_CORE_REGS]; + + /* working area for fastdata access */ + struct working_area *fast_data_area; + + bool swim_configured; + bool bp_scanned; + uint8_t num_hw_bpoints; + uint8_t num_hw_bpoints_avail; + struct stm8_comparator *hw_break_list; + uint32_t blocksize; + uint32_t flashstart; + uint32_t flashend; + uint32_t eepromstart; + uint32_t eepromend; + uint32_t optionstart; + uint32_t optionend; + bool enable_step_irq; + + bool enable_stm8l; + uint32_t flash_cr2; + uint32_t flash_ncr2; + uint32_t flash_iapsr; + uint32_t flash_dukr; + uint32_t flash_pukr; + + /* cc value used for interrupt flags restore */ + uint32_t cc; + bool cc_valid; + + /* register cache to processor synchronization */ + int (*read_core_reg)(struct target *target, unsigned int num); + int (*write_core_reg)(struct target *target, unsigned int num); +}; + +static inline struct stm8_common * +target_to_stm8(struct target *target) +{ + return target->arch_info; +} + +const struct command_registration stm8_command_handlers[]; + +#endif /* OPENOCD_TARGET_STM8_H */ diff --git a/src/target/target.c b/src/target/target.c index 36318d88e..dcb672506 100644 --- a/src/target/target.c +++ b/src/target/target.c @@ -105,6 +105,7 @@ extern struct target_type nds32_v3m_target; extern struct target_type or1k_target; extern struct target_type quark_x10xx_target; extern struct target_type quark_d20xx_target; +extern struct target_type stm8_target; static struct target_type *target_types[] = { &arm7tdmi_target, @@ -136,6 +137,7 @@ static struct target_type *target_types[] = { &or1k_target, &quark_x10xx_target, &quark_d20xx_target, + &stm8_target, #if BUILD_TARGET64 &aarch64_target, #endif diff --git a/tcl/target/stm8l.cfg b/tcl/target/stm8l.cfg new file mode 100644 index 000000000..5cc99e191 --- /dev/null +++ b/tcl/target/stm8l.cfg @@ -0,0 +1,87 @@ +# script for stm8l family + +# +# stm8 devices support SWIM transports only. +# + +transport select stlink_swim + +if { [info exists CHIPNAME] } { + set _CHIPNAME $CHIPNAME +} else { + set _CHIPNAME stm8l +} + +# Work-area is a space in RAM used for flash programming +# By default use 1kB +if { [info exists WORKAREASIZE] } { + set _WORKAREASIZE $WORKAREASIZE +} else { + set _WORKAREASIZE 0x400 +} + +if { [info exists FLASHSTART] } { + set _FLASHSTART $FLASHSTART +} else { + set _FLASHSTART 0x8000 +} + +if { [info exists FLASHEND] } { + set _FLASHEND $FLASHEND +} else { + set _FLASHEND 0xffff +} + +if { [info exists EEPROMSTART] } { + set _EEPROMSTART $EEPROMSTART +} else { + set _EEPROMSTART 0x4000 +} + +if { [info exists EEPROMEND] } { + set _EEPROMEND $EEPROMEND +} else { + set _EEPROMEND 0x43ff +} + +if { [info exists OPTIONSTART] } { + set _OPTIONSTART $OPTIONSTART +} else { + set _OPTIONSTART 0x4800 +} + +if { [info exists OPTIONEND] } { + set _OPTIONEND $OPTIONEND +} else { + set _OPTIONEND 0x487f +} + +if { [info exists BLOCKSIZE] } { + set _BLOCKSIZE $BLOCKSIZE +} else { + set _BLOCKSIZE 0x80 +} + +hla newtap $_CHIPNAME cpu -irlen 4 -ircapture 0x1 -irmask 0xf -expected-id 0 + +set _TARGETNAME $_CHIPNAME.cpu + +target create $_TARGETNAME stm8 -chain-position $_CHIPNAME.cpu + +$_TARGETNAME configure -work-area-phys 0x0 -work-area-size $_WORKAREASIZE -work-area-backup 1 +$_TARGETNAME configure -flashstart $_FLASHSTART -flashend $_FLASHEND -eepromstart $_EEPROMSTART -eepromend $_EEPROMEND +$_TARGETNAME configure -optionstart $_OPTIONSTART -optionend $_OPTIONEND -blocksize $_BLOCKSIZE + +# Uncomment this line to enable interrupts while instruction step +#$_TARGETNAME configure -enable_step_irq + +# Set stm8l type +$_TARGETNAME configure -enable_stm8l + +# The khz rate does not apply here, only slow <0> and fast <1> +adapter_khz 1 + +reset_config srst_only + +#uncomment this line to connect under reset +#reset_config srst_nogate connect_assert_srst diff --git a/tcl/target/stm8s.cfg b/tcl/target/stm8s.cfg new file mode 100644 index 000000000..d55e61b08 --- /dev/null +++ b/tcl/target/stm8s.cfg @@ -0,0 +1,84 @@ +# script for stm8s family + +# +# stm8 devices support SWIM transports only. +# + +transport select stlink_swim + +if { [info exists CHIPNAME] } { + set _CHIPNAME $CHIPNAME +} else { + set _CHIPNAME stm8s +} + +# Work-area is a space in RAM used for flash programming +# By default use 1kB +if { [info exists WORKAREASIZE] } { + set _WORKAREASIZE $WORKAREASIZE +} else { + set _WORKAREASIZE 0x400 +} + +if { [info exists FLASHSTART] } { + set _FLASHSTART $FLASHSTART +} else { + set _FLASHSTART 0x8000 +} + +if { [info exists FLASHEND] } { + set _FLASHEND $FLASHEND +} else { + set _FLASHEND 0xffff +} + +if { [info exists EEPROMSTART] } { + set _EEPROMSTART $EEPROMSTART +} else { + set _EEPROMSTART 0x4000 +} + +if { [info exists EEPROMEND] } { + set _EEPROMEND $EEPROMEND +} else { + set _EEPROMEND 0x43ff +} + +if { [info exists OPTIONSTART] } { + set _OPTIONSTART $OPTIONSTART +} else { + set _OPTIONSTART 0x4800 +} + +if { [info exists OPTIONEND] } { + set _OPTIONEND $OPTIONEND +} else { + set _OPTIONEND 0x487f +} + +if { [info exists BLOCKSIZE] } { + set _BLOCKSIZE $BLOCKSIZE +} else { + set _BLOCKSIZE 0x80 +} + +hla newtap $_CHIPNAME cpu -irlen 4 -ircapture 0x1 -irmask 0xf -expected-id 0 + +set _TARGETNAME $_CHIPNAME.cpu + +target create $_TARGETNAME stm8 -chain-position $_CHIPNAME.cpu + +$_TARGETNAME configure -work-area-phys 0x0 -work-area-size $_WORKAREASIZE -work-area-backup 1 +$_TARGETNAME configure -flashstart $_FLASHSTART -flashend $_FLASHEND -eepromstart $_EEPROMSTART -eepromend $_EEPROMEND +$_TARGETNAME configure -optionstart $_OPTIONSTART -optionend $_OPTIONEND -blocksize $_BLOCKSIZE + +# Uncomment this line to enable interrupts while instruction step +#$_TARGETNAME configure -enable_step_irq + +# The khz rate does not apply here, only slow <0> and fast <1> +adapter_khz 1 + +reset_config srst_only + +# uncomment this line to connect under reset +#reset_config srst_nogate connect_assert_srst