openocd/src/target/x86_32_common.c
Marc Schink fb43f1ff4e target: Rework 'set' variable of break-/watchpoints
The 'set' variable name suggests a boolean data type which determines
whether a breakpoint (or watchpoint) is active. However, it is also
used to store the number of the breakpoint.

This encoding leads to inconsistent value assignments: boolean and
integer values are mixed. Also, associated hardware comparator
numbers, which are usually numbered from 0, cannot be used directly.
An additional offset is required to store the comparator numbers.

In order to make the code more readable and the value assignment more
consistent, change the variable name to 'is_set', its data type to 'bool'
and introduce a dedicated variable for the break-/watchpoint
number.

In order to make the review easier, the data types of various related
variables (e.g. number of breakpoints) are not changed.

While at it, fix a few coding style issues.

Change-Id: I2193f5639247cce6b80580d4c1c6afee916aeb82
Signed-off-by: Marc Schink <dev@zapb.de>
Reviewed-on: https://review.openocd.org/c/openocd/+/6319
Tested-by: jenkins
Reviewed-by: Antonio Borneo <borneo.antonio@gmail.com>
2022-03-19 09:14:39 +00:00

1546 lines
46 KiB
C

/*
* Copyright(c) 2013 Intel Corporation.
*
* Adrian Burns (adrian.burns@intel.com)
* Thomas Faust (thomas.faust@intel.com)
* Ivan De Cesaris (ivan.de.cesaris@intel.com)
* Julien Carreno (julien.carreno@intel.com)
* Jeffrey Maxwell (jeffrey.r.maxwell@intel.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 2 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 <http://www.gnu.org/licenses/>.
*
* Contact Information:
* Intel Corporation
*/
/*
* @file
* This implements generic x86 32 bit memory and breakpoint operations.
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include <helper/log.h>
#include "target.h"
#include "target_type.h"
#include "register.h"
#include "breakpoints.h"
#include "x86_32_common.h"
static int set_debug_regs(struct target *t, uint32_t address,
uint8_t bp_num, uint8_t bp_type, uint8_t bp_length);
static int unset_debug_regs(struct target *t, uint8_t bp_num);
static int read_mem(struct target *t, uint32_t size,
uint32_t addr, uint8_t *buf);
static int write_mem(struct target *t, uint32_t size,
uint32_t addr, const uint8_t *buf);
static int calcaddr_physfromlin(struct target *t, target_addr_t addr,
target_addr_t *physaddr);
static int read_phys_mem(struct target *t, uint32_t phys_address,
uint32_t size, uint32_t count, uint8_t *buffer);
static int write_phys_mem(struct target *t, uint32_t phys_address,
uint32_t size, uint32_t count, const uint8_t *buffer);
static int set_breakpoint(struct target *target,
struct breakpoint *breakpoint);
static int unset_breakpoint(struct target *target,
struct breakpoint *breakpoint);
static int set_watchpoint(struct target *target,
struct watchpoint *watchpoint);
static int unset_watchpoint(struct target *target,
struct watchpoint *watchpoint);
static int read_hw_reg_to_cache(struct target *t, int num);
static int write_hw_reg_from_cache(struct target *t, int num);
int x86_32_get_gdb_reg_list(struct target *t,
struct reg **reg_list[], int *reg_list_size,
enum target_register_class reg_class)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
int i;
*reg_list_size = x86_32->cache->num_regs;
LOG_DEBUG("num_regs=%d, reg_class=%d", (*reg_list_size), reg_class);
*reg_list = malloc(sizeof(struct reg *) * (*reg_list_size));
if (!*reg_list) {
LOG_ERROR("%s out of memory", __func__);
return ERROR_FAIL;
}
/* this will copy the values from our reg list to gdbs */
for (i = 0; i < (*reg_list_size); i++) {
(*reg_list)[i] = &x86_32->cache->reg_list[i];
LOG_DEBUG("value %s = %08" PRIx32, x86_32->cache->reg_list[i].name,
buf_get_u32(x86_32->cache->reg_list[i].value, 0, 32));
}
return ERROR_OK;
}
int x86_32_common_init_arch_info(struct target *t, struct x86_32_common *x86_32)
{
t->arch_info = x86_32;
x86_32->common_magic = X86_32_COMMON_MAGIC;
x86_32->num_hw_bpoints = MAX_DEBUG_REGS;
x86_32->hw_break_list = calloc(x86_32->num_hw_bpoints,
sizeof(struct x86_32_dbg_reg));
if (!x86_32->hw_break_list) {
LOG_ERROR("%s out of memory", __func__);
return ERROR_FAIL;
}
x86_32->curr_tap = t->tap;
x86_32->fast_data_area = NULL;
x86_32->flush = 1;
x86_32->read_hw_reg_to_cache = read_hw_reg_to_cache;
x86_32->write_hw_reg_from_cache = write_hw_reg_from_cache;
return ERROR_OK;
}
int x86_32_common_mmu(struct target *t, int *enabled)
{
*enabled = true;
return ERROR_OK;
}
int x86_32_common_virt2phys(struct target *t, target_addr_t address, target_addr_t *physical)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
/*
* We need to ignore 'segmentation' for now, as OpenOCD can't handle
* segmented addresses.
* In protected mode that is almost OK, as (almost) any known OS is using
* flat segmentation. In real mode we use use the base of the DS segment,
* as we don't know better ...
*/
uint32_t cr0 = buf_get_u32(x86_32->cache->reg_list[CR0].value, 0, 32);
if (!(cr0 & CR0_PG)) {
/* target halted in real mode */
/* TODO: needs validation !!! */
uint32_t dsb = buf_get_u32(x86_32->cache->reg_list[DSB].value, 0, 32);
*physical = dsb + address;
} else {
/* target halted in protected mode */
if (calcaddr_physfromlin(t, address, physical) != ERROR_OK) {
LOG_ERROR("%s failed to calculate physical address from " TARGET_ADDR_FMT,
__func__, address);
return ERROR_FAIL;
}
}
return ERROR_OK;
}
int x86_32_common_read_phys_mem(struct target *t, target_addr_t phys_address,
uint32_t size, uint32_t count, uint8_t *buffer)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
int error;
error = read_phys_mem(t, phys_address, size, count, buffer);
if (error != ERROR_OK)
return error;
/* After reading memory from target, we must replace software breakpoints
* with the original instructions again.
*/
struct swbp_mem_patch *iter = x86_32->swbbp_mem_patch_list;
while (iter) {
if (iter->physaddr >= phys_address && iter->physaddr < phys_address+(size*count)) {
uint32_t offset = iter->physaddr - phys_address;
buffer[offset] = iter->orig_byte;
}
iter = iter->next;
}
return ERROR_OK;
}
static int read_phys_mem(struct target *t, uint32_t phys_address,
uint32_t size, uint32_t count, uint8_t *buffer)
{
int retval = ERROR_OK;
bool pg_disabled = false;
LOG_DEBUG("addr=0x%08" PRIx32 ", size=%" PRIu32 ", count=0x%" PRIx32 ", buf=%p",
phys_address, size, count, buffer);
struct x86_32_common *x86_32 = target_to_x86_32(t);
if (check_not_halted(t))
return ERROR_TARGET_NOT_HALTED;
if (!count || !buffer || !phys_address) {
LOG_ERROR("%s invalid params count=0x%" PRIx32 ", buf=%p, addr=0x%08" PRIx32,
__func__, count, buffer, phys_address);
return ERROR_COMMAND_ARGUMENT_INVALID;
}
/* to access physical memory, switch off the CR0.PG bit */
if (x86_32->is_paging_enabled(t)) {
retval = x86_32->disable_paging(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s could not disable paging", __func__);
return retval;
}
pg_disabled = true;
}
for (uint32_t i = 0; i < count; i++) {
switch (size) {
case BYTE:
retval = read_mem(t, size, phys_address + i, buffer + i);
break;
case WORD:
retval = read_mem(t, size, phys_address + i * 2, buffer + i * 2);
break;
case DWORD:
retval = read_mem(t, size, phys_address + i * 4, buffer + i * 4);
break;
default:
LOG_ERROR("%s invalid read size", __func__);
break;
}
if (retval != ERROR_OK)
break;
}
/* restore CR0.PG bit if needed (regardless of retval) */
if (pg_disabled) {
int retval2 = x86_32->enable_paging(t);
if (retval2 != ERROR_OK) {
LOG_ERROR("%s could not enable paging", __func__);
return retval2;
}
}
/* TODO: After reading memory from target, we must replace
* software breakpoints with the original instructions again.
* Solve this with the breakpoint fix
*/
return retval;
}
int x86_32_common_write_phys_mem(struct target *t, target_addr_t phys_address,
uint32_t size, uint32_t count, const uint8_t *buffer)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
int error = ERROR_OK;
uint8_t *newbuffer = NULL;
check_not_halted(t);
if (!count || !buffer || !phys_address) {
LOG_ERROR("%s invalid params count=0x%" PRIx32 ", buf=%p, addr=" TARGET_ADDR_FMT,
__func__, count, buffer, phys_address);
return ERROR_COMMAND_ARGUMENT_INVALID;
}
/* Before writing memory to target, we must update software breakpoints
* with the new instructions and patch the memory buffer with the
* breakpoint instruction.
*/
newbuffer = malloc(size*count);
if (!newbuffer) {
LOG_ERROR("%s out of memory", __func__);
return ERROR_FAIL;
}
memcpy(newbuffer, buffer, size*count);
struct swbp_mem_patch *iter = x86_32->swbbp_mem_patch_list;
while (iter) {
if (iter->physaddr >= phys_address && iter->physaddr < phys_address+(size*count)) {
uint32_t offset = iter->physaddr - phys_address;
newbuffer[offset] = SW_BP_OPCODE;
/* update the breakpoint */
struct breakpoint *pbiter = t->breakpoints;
while (pbiter && pbiter->unique_id != iter->swbp_unique_id)
pbiter = pbiter->next;
if (pbiter)
pbiter->orig_instr[0] = buffer[offset];
}
iter = iter->next;
}
error = write_phys_mem(t, phys_address, size, count, newbuffer);
free(newbuffer);
return error;
}
static int write_phys_mem(struct target *t, uint32_t phys_address,
uint32_t size, uint32_t count, const uint8_t *buffer)
{
int retval = ERROR_OK;
bool pg_disabled = false;
struct x86_32_common *x86_32 = target_to_x86_32(t);
LOG_DEBUG("addr=0x%08" PRIx32 ", size=%" PRIu32 ", count=0x%" PRIx32 ", buf=%p",
phys_address, size, count, buffer);
check_not_halted(t);
if (!count || !buffer || !phys_address) {
LOG_ERROR("%s invalid params count=0x%" PRIx32 ", buf=%p, addr=0x%08" PRIx32,
__func__, count, buffer, phys_address);
return ERROR_COMMAND_ARGUMENT_INVALID;
}
/* TODO: Before writing memory to target, we must update
* software breakpoints with the new instructions and
* patch the memory buffer with the breakpoint instruction.
* Solve this with the breakpoint fix
*/
/* to access physical memory, switch off the CR0.PG bit */
if (x86_32->is_paging_enabled(t)) {
retval = x86_32->disable_paging(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s could not disable paging", __func__);
return retval;
}
pg_disabled = true;
}
for (uint32_t i = 0; i < count; i++) {
switch (size) {
case BYTE:
retval = write_mem(t, size, phys_address + i, buffer + i);
break;
case WORD:
retval = write_mem(t, size, phys_address + i * 2, buffer + i * 2);
break;
case DWORD:
retval = write_mem(t, size, phys_address + i * 4, buffer + i * 4);
break;
default:
LOG_DEBUG("invalid read size");
break;
}
}
/* restore CR0.PG bit if needed (regardless of retval) */
if (pg_disabled) {
retval = x86_32->enable_paging(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s could not enable paging", __func__);
return retval;
}
}
return retval;
}
static int read_mem(struct target *t, uint32_t size,
uint32_t addr, uint8_t *buf)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
/* if CS.D bit=1 then its a 32 bit code segment, else 16 */
bool use32 = (buf_get_u32(x86_32->cache->reg_list[CSAR].value, 0, 32)) & CSAR_D;
int retval = x86_32->write_hw_reg(t, EAX, addr, 0);
if (retval != ERROR_OK) {
LOG_ERROR("%s error write EAX", __func__);
return retval;
}
switch (size) {
case BYTE:
if (use32)
retval = x86_32->submit_instruction(t, MEMRDB32);
else
retval = x86_32->submit_instruction(t, MEMRDB16);
break;
case WORD:
if (use32)
retval = x86_32->submit_instruction(t, MEMRDH32);
else
retval = x86_32->submit_instruction(t, MEMRDH16);
break;
case DWORD:
if (use32)
retval = x86_32->submit_instruction(t, MEMRDW32);
else
retval = x86_32->submit_instruction(t, MEMRDW16);
break;
default:
LOG_ERROR("%s invalid read mem size", __func__);
break;
}
if (retval != ERROR_OK)
return retval;
/* read_hw_reg() will write to 4 bytes (uint32_t)
* Watch out, the buffer passed into read_mem() might be 1 or 2 bytes.
*/
uint32_t regval;
retval = x86_32->read_hw_reg(t, EDX, &regval, 0);
if (retval != ERROR_OK) {
LOG_ERROR("%s error read EDX", __func__);
return retval;
}
for (uint8_t i = 0; i < size; i++)
buf[i] = (regval >> (i*8)) & 0x000000FF;
retval = x86_32->transaction_status(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s error on mem read", __func__);
return retval;
}
return retval;
}
static int write_mem(struct target *t, uint32_t size,
uint32_t addr, const uint8_t *buf)
{
uint32_t i = 0;
uint32_t buf4bytes = 0;
int retval = ERROR_OK;
struct x86_32_common *x86_32 = target_to_x86_32(t);
for (i = 0; i < size; ++i) {
buf4bytes = buf4bytes << 8; /* first time we only shift 0s */
buf4bytes += buf[(size-1)-i]; /* it was hard to write, should be hard to read! */
}
/* if CS.D bit=1 then its a 32 bit code segment, else 16 */
bool use32 = (buf_get_u32(x86_32->cache->reg_list[CSAR].value, 0, 32)) & CSAR_D;
retval = x86_32->write_hw_reg(t, EAX, addr, 0);
if (retval != ERROR_OK) {
LOG_ERROR("%s error write EAX", __func__);
return retval;
}
/* write_hw_reg() will write to 4 bytes (uint32_t)
* Watch out, the buffer passed into write_mem() might be 1 or 2 bytes.
*/
retval = x86_32->write_hw_reg(t, EDX, buf4bytes, 0);
if (retval != ERROR_OK) {
LOG_ERROR("%s error write EDX", __func__);
return retval;
}
switch (size) {
case BYTE:
if (use32)
retval = x86_32->submit_instruction(t, MEMWRB32);
else
retval = x86_32->submit_instruction(t, MEMWRB16);
break;
case WORD:
if (use32)
retval = x86_32->submit_instruction(t, MEMWRH32);
else
retval = x86_32->submit_instruction(t, MEMWRH16);
break;
case DWORD:
if (use32)
retval = x86_32->submit_instruction(t, MEMWRW32);
else
retval = x86_32->submit_instruction(t, MEMWRW16);
break;
default:
LOG_ERROR("%s invalid write mem size", __func__);
return ERROR_FAIL;
}
if (retval != ERROR_OK)
return retval;
retval = x86_32->transaction_status(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s error on mem write", __func__);
return retval;
}
return retval;
}
int calcaddr_physfromlin(struct target *t, target_addr_t addr, target_addr_t *physaddr)
{
uint8_t entry_buffer[8];
if (!physaddr || !t)
return ERROR_FAIL;
struct x86_32_common *x86_32 = target_to_x86_32(t);
/* The 'user-visible' CR0.PG should be set - otherwise the function shouldn't be called
* (Don't check the CR0.PG on the target, this might be temporally disabled at this point)
*/
uint32_t cr0 = buf_get_u32(x86_32->cache->reg_list[CR0].value, 0, 32);
if (!(cr0 & CR0_PG)) {
/* you are wrong in this function, never mind */
*physaddr = addr;
return ERROR_OK;
}
uint32_t cr4 = buf_get_u32(x86_32->cache->reg_list[CR4].value, 0, 32);
bool is_pae = cr4 & 0x00000020; /* PAE - Physical Address Extension */
uint32_t cr3 = buf_get_u32(x86_32->cache->reg_list[CR3].value, 0, 32);
if (is_pae) {
uint32_t pdpt_base = cr3 & 0xFFFFF000; /* lower 12 bits of CR3 must always be 0 */
uint32_t pdpt_index = (addr & 0xC0000000) >> 30; /* A[31:30] index to PDPT */
uint32_t pdpt_addr = pdpt_base + (8 * pdpt_index);
if (x86_32_common_read_phys_mem(t, pdpt_addr, 4, 2, entry_buffer) != ERROR_OK) {
LOG_ERROR("%s couldn't read page directory pointer table entry at 0x%08" PRIx32,
__func__, pdpt_addr);
return ERROR_FAIL;
}
uint64_t pdpt_entry = target_buffer_get_u64(t, entry_buffer);
if (!(pdpt_entry & 0x0000000000000001)) {
LOG_ERROR("%s page directory pointer table entry at 0x%08" PRIx32 " is not present",
__func__, pdpt_addr);
return ERROR_FAIL;
}
uint32_t pd_base = pdpt_entry & 0xFFFFF000; /* A[31:12] is PageTable/Page Base Address */
uint32_t pd_index = (addr & 0x3FE00000) >> 21; /* A[29:21] index to PD entry with PAE */
uint32_t pd_addr = pd_base + (8 * pd_index);
if (x86_32_common_read_phys_mem(t, pd_addr, 4, 2, entry_buffer) != ERROR_OK) {
LOG_ERROR("%s couldn't read page directory entry at 0x%08" PRIx32,
__func__, pd_addr);
return ERROR_FAIL;
}
uint64_t pd_entry = target_buffer_get_u64(t, entry_buffer);
if (!(pd_entry & 0x0000000000000001)) {
LOG_ERROR("%s page directory entry at 0x%08" PRIx32 " is not present",
__func__, pd_addr);
return ERROR_FAIL;
}
/* PS bit in PD entry is indicating 4KB or 2MB page size */
if (pd_entry & 0x0000000000000080) {
uint32_t page_base = (uint32_t)(pd_entry & 0x00000000FFE00000); /* [31:21] */
uint32_t offset = addr & 0x001FFFFF; /* [20:0] */
*physaddr = page_base + offset;
return ERROR_OK;
} else {
uint32_t pt_base = (uint32_t)(pd_entry & 0x00000000FFFFF000); /*[31:12]*/
uint32_t pt_index = (addr & 0x001FF000) >> 12; /*[20:12]*/
uint32_t pt_addr = pt_base + (8 * pt_index);
if (x86_32_common_read_phys_mem(t, pt_addr, 4, 2, entry_buffer) != ERROR_OK) {
LOG_ERROR("%s couldn't read page table entry at 0x%08" PRIx32, __func__, pt_addr);
return ERROR_FAIL;
}
uint64_t pt_entry = target_buffer_get_u64(t, entry_buffer);
if (!(pt_entry & 0x0000000000000001)) {
LOG_ERROR("%s page table entry at 0x%08" PRIx32 " is not present", __func__, pt_addr);
return ERROR_FAIL;
}
uint32_t page_base = (uint32_t)(pt_entry & 0x00000000FFFFF000); /*[31:12]*/
uint32_t offset = addr & 0x00000FFF; /*[11:0]*/
*physaddr = page_base + offset;
return ERROR_OK;
}
} else {
uint32_t pd_base = cr3 & 0xFFFFF000; /* lower 12 bits of CR3 must always be 0 */
uint32_t pd_index = (addr & 0xFFC00000) >> 22; /* A[31:22] index to PD entry */
uint32_t pd_addr = pd_base + (4 * pd_index);
if (x86_32_common_read_phys_mem(t, pd_addr, 4, 1, entry_buffer) != ERROR_OK) {
LOG_ERROR("%s couldn't read page directory entry at 0x%08" PRIx32, __func__, pd_addr);
return ERROR_FAIL;
}
uint32_t pd_entry = target_buffer_get_u32(t, entry_buffer);
if (!(pd_entry & 0x00000001)) {
LOG_ERROR("%s page directory entry at 0x%08" PRIx32 " is not present", __func__, pd_addr);
return ERROR_FAIL;
}
/* Bit 7 in page directory entry is page size.
*/
if (pd_entry & 0x00000080) {
/* 4MB pages */
uint32_t page_base = pd_entry & 0xFFC00000;
*physaddr = page_base + (addr & 0x003FFFFF);
} else {
/* 4KB pages */
uint32_t pt_base = pd_entry & 0xFFFFF000; /* A[31:12] is PageTable/Page Base Address */
uint32_t pt_index = (addr & 0x003FF000) >> 12; /* A[21:12] index to page table entry */
uint32_t pt_addr = pt_base + (4 * pt_index);
if (x86_32_common_read_phys_mem(t, pt_addr, 4, 1, entry_buffer) != ERROR_OK) {
LOG_ERROR("%s couldn't read page table entry at 0x%08" PRIx32, __func__, pt_addr);
return ERROR_FAIL;
}
uint32_t pt_entry = target_buffer_get_u32(t, entry_buffer);
if (!(pt_entry & 0x00000001)) {
LOG_ERROR("%s page table entry at 0x%08" PRIx32 " is not present", __func__, pt_addr);
return ERROR_FAIL;
}
uint32_t page_base = pt_entry & 0xFFFFF000; /* A[31:12] is PageTable/Page Base Address */
*physaddr = page_base + (addr & 0x00000FFF); /* A[11:0] offset to 4KB page in linear address */
}
}
return ERROR_OK;
}
int x86_32_common_read_memory(struct target *t, target_addr_t addr,
uint32_t size, uint32_t count, uint8_t *buf)
{
int retval = ERROR_OK;
struct x86_32_common *x86_32 = target_to_x86_32(t);
LOG_DEBUG("addr=" TARGET_ADDR_FMT ", size=%" PRIu32 ", count=0x%" PRIx32 ", buf=%p",
addr, size, count, buf);
check_not_halted(t);
if (!count || !buf || !addr) {
LOG_ERROR("%s invalid params count=0x%" PRIx32 ", buf=%p, addr=" TARGET_ADDR_FMT,
__func__, count, buf, addr);
return ERROR_COMMAND_ARGUMENT_INVALID;
}
if (x86_32->is_paging_enabled(t)) {
/* all memory accesses from debugger must be physical (CR0.PG == 0)
* conversion to physical address space needed
*/
retval = x86_32->disable_paging(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s could not disable paging", __func__);
return retval;
}
target_addr_t physaddr = 0;
if (calcaddr_physfromlin(t, addr, &physaddr) != ERROR_OK) {
LOG_ERROR("%s failed to calculate physical address from " TARGET_ADDR_FMT,
__func__, addr);
retval = ERROR_FAIL;
}
/* TODO: !!! Watch out for page boundaries
* for every 4kB, the physical address has to be re-calculated
* This should be fixed together with bulk memory reads
*/
if (retval == ERROR_OK
&& x86_32_common_read_phys_mem(t, physaddr, size, count, buf) != ERROR_OK) {
LOG_ERROR("%s failed to read memory from physical address " TARGET_ADDR_FMT,
__func__, physaddr);
}
/* restore PG bit if it was cleared prior (regardless of retval) */
retval = x86_32->enable_paging(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s could not enable paging", __func__);
return retval;
}
} else {
/* paging is off - linear address is physical address */
if (x86_32_common_read_phys_mem(t, addr, size, count, buf) != ERROR_OK) {
LOG_ERROR("%s failed to read memory from address " TARGET_ADDR_FMT,
__func__, addr);
retval = ERROR_FAIL;
}
}
return retval;
}
int x86_32_common_write_memory(struct target *t, target_addr_t addr,
uint32_t size, uint32_t count, const uint8_t *buf)
{
int retval = ERROR_OK;
struct x86_32_common *x86_32 = target_to_x86_32(t);
LOG_DEBUG("addr=" TARGET_ADDR_FMT ", size=%" PRIu32 ", count=0x%" PRIx32 ", buf=%p",
addr, size, count, buf);
check_not_halted(t);
if (!count || !buf || !addr) {
LOG_ERROR("%s invalid params count=0x%" PRIx32 ", buf=%p, addr=" TARGET_ADDR_FMT,
__func__, count, buf, addr);
return ERROR_COMMAND_ARGUMENT_INVALID;
}
if (x86_32->is_paging_enabled(t)) {
/* all memory accesses from debugger must be physical (CR0.PG == 0)
* conversion to physical address space needed
*/
retval = x86_32->disable_paging(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s could not disable paging", __func__);
return retval;
}
target_addr_t physaddr = 0;
if (calcaddr_physfromlin(t, addr, &physaddr) != ERROR_OK) {
LOG_ERROR("%s failed to calculate physical address from " TARGET_ADDR_FMT,
__func__, addr);
retval = ERROR_FAIL;
}
/* TODO: !!! Watch out for page boundaries
* for every 4kB, the physical address has to be re-calculated
* This should be fixed together with bulk memory reads
*/
if (retval == ERROR_OK
&& x86_32_common_write_phys_mem(t, physaddr, size, count, buf) != ERROR_OK) {
LOG_ERROR("%s failed to write memory to physical address " TARGET_ADDR_FMT,
__func__, physaddr);
}
/* restore PG bit if it was cleared prior (regardless of retval) */
retval = x86_32->enable_paging(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s could not enable paging", __func__);
return retval;
}
} else {
/* paging is off - linear address is physical address */
if (x86_32_common_write_phys_mem(t, addr, size, count, buf) != ERROR_OK) {
LOG_ERROR("%s failed to write memory to address " TARGET_ADDR_FMT,
__func__, addr);
retval = ERROR_FAIL;
}
}
return retval;
}
int x86_32_common_read_io(struct target *t, uint32_t addr,
uint32_t size, uint8_t *buf)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
/* if CS.D bit=1 then its a 32 bit code segment, else 16 */
bool use32 = (buf_get_u32(x86_32->cache->reg_list[CSAR].value, 0, 32)) & CSAR_D;
int retval = ERROR_FAIL;
bool pg_disabled = false;
LOG_DEBUG("addr=0x%08" PRIx32 ", size=%" PRIu32 ", buf=%p", addr, size, buf);
check_not_halted(t);
if (!buf || !addr) {
LOG_ERROR("%s invalid params buf=%p, addr=%08" PRIx32, __func__, buf, addr);
return retval;
}
retval = x86_32->write_hw_reg(t, EDX, addr, 0);
if (retval != ERROR_OK) {
LOG_ERROR("%s error EDX write", __func__);
return retval;
}
/* to access physical memory, switch off the CR0.PG bit */
if (x86_32->is_paging_enabled(t)) {
retval = x86_32->disable_paging(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s could not disable paging", __func__);
return retval;
}
pg_disabled = true;
}
switch (size) {
case BYTE:
if (use32)
retval = x86_32->submit_instruction(t, IORDB32);
else
retval = x86_32->submit_instruction(t, IORDB16);
break;
case WORD:
if (use32)
retval = x86_32->submit_instruction(t, IORDH32);
else
retval = x86_32->submit_instruction(t, IORDH16);
break;
case DWORD:
if (use32)
retval = x86_32->submit_instruction(t, IORDW32);
else
retval = x86_32->submit_instruction(t, IORDW16);
break;
default:
LOG_ERROR("%s invalid read io size", __func__);
return ERROR_FAIL;
}
/* restore CR0.PG bit if needed */
if (pg_disabled) {
int retval2 = x86_32->enable_paging(t);
if (retval2 != ERROR_OK) {
LOG_ERROR("%s could not enable paging", __func__);
return retval2;
}
}
if (retval != ERROR_OK)
return retval;
uint32_t regval = 0;
retval = x86_32->read_hw_reg(t, EAX, &regval, 0);
if (retval != ERROR_OK) {
LOG_ERROR("%s error on read EAX", __func__);
return retval;
}
for (uint8_t i = 0; i < size; i++)
buf[i] = (regval >> (i*8)) & 0x000000FF;
retval = x86_32->transaction_status(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s error on io read", __func__);
return retval;
}
return retval;
}
int x86_32_common_write_io(struct target *t, uint32_t addr,
uint32_t size, const uint8_t *buf)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
/* if CS.D bit=1 then its a 32 bit code segment, else 16 */
bool use32 = (buf_get_u32(x86_32->cache->reg_list[CSAR].value, 0, 32)) & CSAR_D;
LOG_DEBUG("addr=0x%08" PRIx32 ", size=%" PRIu32 ", buf=%p", addr, size, buf);
check_not_halted(t);
int retval = ERROR_FAIL;
bool pg_disabled = false;
if (!buf || !addr) {
LOG_ERROR("%s invalid params buf=%p, addr=0x%08" PRIx32, __func__, buf, addr);
return retval;
}
/* no do the write */
retval = x86_32->write_hw_reg(t, EDX, addr, 0);
if (retval != ERROR_OK) {
LOG_ERROR("%s error on EDX write", __func__);
return retval;
}
uint32_t regval = 0;
for (uint8_t i = 0; i < size; i++)
regval += (buf[i] << (i*8));
retval = x86_32->write_hw_reg(t, EAX, regval, 0);
if (retval != ERROR_OK) {
LOG_ERROR("%s error on EAX write", __func__);
return retval;
}
/* to access physical memory, switch off the CR0.PG bit */
if (x86_32->is_paging_enabled(t)) {
retval = x86_32->disable_paging(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s could not disable paging", __func__);
return retval;
}
pg_disabled = true;
}
switch (size) {
case BYTE:
if (use32)
retval = x86_32->submit_instruction(t, IOWRB32);
else
retval = x86_32->submit_instruction(t, IOWRB16);
break;
case WORD:
if (use32)
retval = x86_32->submit_instruction(t, IOWRH32);
else
retval = x86_32->submit_instruction(t, IOWRH16);
break;
case DWORD:
if (use32)
retval = x86_32->submit_instruction(t, IOWRW32);
else
retval = x86_32->submit_instruction(t, IOWRW16);
break;
default:
LOG_ERROR("%s invalid write io size", __func__);
return ERROR_FAIL;
}
/* restore CR0.PG bit if needed */
if (pg_disabled) {
int retval2 = x86_32->enable_paging(t);
if (retval2 != ERROR_OK) {
LOG_ERROR("%s could not enable paging", __func__);
return retval2;
}
}
if (retval != ERROR_OK)
return retval;
retval = x86_32->transaction_status(t);
if (retval != ERROR_OK) {
LOG_ERROR("%s error on io write", __func__);
return retval;
}
return retval;
}
int x86_32_common_add_watchpoint(struct target *t, struct watchpoint *wp)
{
check_not_halted(t);
/* set_watchpoint() will return ERROR_TARGET_RESOURCE_NOT_AVAILABLE if all
* hardware registers are gone
*/
return set_watchpoint(t, wp);
}
int x86_32_common_remove_watchpoint(struct target *t, struct watchpoint *wp)
{
if (check_not_halted(t))
return ERROR_TARGET_NOT_HALTED;
if (wp->is_set)
unset_watchpoint(t, wp);
return ERROR_OK;
}
int x86_32_common_add_breakpoint(struct target *t, struct breakpoint *bp)
{
LOG_DEBUG("type=%d, addr=" TARGET_ADDR_FMT, bp->type, bp->address);
if (check_not_halted(t))
return ERROR_TARGET_NOT_HALTED;
/* set_breakpoint() will return ERROR_TARGET_RESOURCE_NOT_AVAILABLE if all
* hardware registers are gone (for hardware breakpoints)
*/
return set_breakpoint(t, bp);
}
int x86_32_common_remove_breakpoint(struct target *t, struct breakpoint *bp)
{
LOG_DEBUG("type=%d, addr=" TARGET_ADDR_FMT, bp->type, bp->address);
if (check_not_halted(t))
return ERROR_TARGET_NOT_HALTED;
if (bp->is_set)
unset_breakpoint(t, bp);
return ERROR_OK;
}
static int set_debug_regs(struct target *t, uint32_t address,
uint8_t bp_num, uint8_t bp_type, uint8_t bp_length)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
LOG_DEBUG("addr=0x%08" PRIx32 ", bp_num=%" PRIu8 ", bp_type=%" PRIu8 ", pb_length=%" PRIu8,
address, bp_num, bp_type, bp_length);
/* DR7 - set global enable */
uint32_t dr7 = buf_get_u32(x86_32->cache->reg_list[DR7].value, 0, 32);
if (bp_length != 1 && bp_length != 2 && bp_length != 4)
return ERROR_FAIL;
if (DR7_BP_FREE(dr7, bp_num))
DR7_GLOBAL_ENABLE(dr7, bp_num);
else {
LOG_ERROR("%s dr7 error, already enabled, val=%08" PRIx32, __func__, dr7);
return ERROR_TARGET_RESOURCE_NOT_AVAILABLE;
}
switch (bp_type) {
case 0:
/* 00 - only on instruction execution */
DR7_SET_EXE(dr7, bp_num);
DR7_SET_LENGTH(dr7, bp_num, bp_length);
break;
case 1:
/* 01 - only on data writes */
DR7_SET_WRITE(dr7, bp_num);
DR7_SET_LENGTH(dr7, bp_num, bp_length);
break;
case 2:
/* 10 UNSUPPORTED - an I/O read and I/O write */
LOG_ERROR("%s unsupported feature bp_type=%d", __func__, bp_type);
return ERROR_FAIL;
break;
case 3:
/* on data read or data write */
DR7_SET_ACCESS(dr7, bp_num);
DR7_SET_LENGTH(dr7, bp_num, bp_length);
break;
default:
LOG_ERROR("%s invalid request [only 0-3] bp_type=%d", __func__, bp_type);
return ERROR_FAIL;
}
/* update regs in the reg cache ready to be written to hardware
* when we exit PM
*/
buf_set_u32(x86_32->cache->reg_list[bp_num+DR0].value, 0, 32, address);
x86_32->cache->reg_list[bp_num+DR0].dirty = true;
x86_32->cache->reg_list[bp_num+DR0].valid = true;
buf_set_u32(x86_32->cache->reg_list[DR6].value, 0, 32, PM_DR6);
x86_32->cache->reg_list[DR6].dirty = true;
x86_32->cache->reg_list[DR6].valid = true;
buf_set_u32(x86_32->cache->reg_list[DR7].value, 0, 32, dr7);
x86_32->cache->reg_list[DR7].dirty = true;
x86_32->cache->reg_list[DR7].valid = true;
return ERROR_OK;
}
static int unset_debug_regs(struct target *t, uint8_t bp_num)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
LOG_DEBUG("bp_num=%" PRIu8, bp_num);
uint32_t dr7 = buf_get_u32(x86_32->cache->reg_list[DR7].value, 0, 32);
if (!(DR7_BP_FREE(dr7, bp_num))) {
DR7_GLOBAL_DISABLE(dr7, bp_num);
} else {
LOG_ERROR("%s dr7 error, not enabled, val=0x%08" PRIx32, __func__, dr7);
return ERROR_TARGET_RESOURCE_NOT_AVAILABLE;
}
/* this will clear rw and len bits */
DR7_RESET_RWLEN_BITS(dr7, bp_num);
/* update regs in the reg cache ready to be written to hardware
* when we exit PM
*/
buf_set_u32(x86_32->cache->reg_list[bp_num+DR0].value, 0, 32, 0);
x86_32->cache->reg_list[bp_num+DR0].dirty = true;
x86_32->cache->reg_list[bp_num+DR0].valid = true;
buf_set_u32(x86_32->cache->reg_list[DR6].value, 0, 32, PM_DR6);
x86_32->cache->reg_list[DR6].dirty = true;
x86_32->cache->reg_list[DR6].valid = true;
buf_set_u32(x86_32->cache->reg_list[DR7].value, 0, 32, dr7);
x86_32->cache->reg_list[DR7].dirty = true;
x86_32->cache->reg_list[DR7].valid = true;
return ERROR_OK;
}
static int set_hwbp(struct target *t, struct breakpoint *bp)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
struct x86_32_dbg_reg *debug_reg_list = x86_32->hw_break_list;
uint8_t hwbp_num = 0;
while (debug_reg_list[hwbp_num].used && (hwbp_num < x86_32->num_hw_bpoints))
hwbp_num++;
if (hwbp_num >= x86_32->num_hw_bpoints) {
LOG_ERROR("%s no free hw breakpoint bpid=0x%" PRIx32, __func__, bp->unique_id);
return ERROR_TARGET_RESOURCE_NOT_AVAILABLE;
}
if (set_debug_regs(t, bp->address, hwbp_num, DR7_BP_EXECUTE, 1) != ERROR_OK)
return ERROR_FAIL;
breakpoint_hw_set(bp, hwbp_num);
debug_reg_list[hwbp_num].used = 1;
debug_reg_list[hwbp_num].bp_value = bp->address;
LOG_USER("%s hardware breakpoint %" PRIu32 " set at 0x%08" PRIx32 " (hwreg=%" PRIu8 ")", __func__,
bp->unique_id, debug_reg_list[hwbp_num].bp_value, hwbp_num);
return ERROR_OK;
}
static int unset_hwbp(struct target *t, struct breakpoint *bp)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
struct x86_32_dbg_reg *debug_reg_list = x86_32->hw_break_list;
int hwbp_num = bp->number;
if (hwbp_num >= x86_32->num_hw_bpoints) {
LOG_ERROR("%s invalid breakpoint number=%d, bpid=%" PRIu32,
__func__, hwbp_num, bp->unique_id);
return ERROR_OK;
}
if (unset_debug_regs(t, hwbp_num) != ERROR_OK)
return ERROR_FAIL;
debug_reg_list[hwbp_num].used = 0;
debug_reg_list[hwbp_num].bp_value = 0;
LOG_USER("%s hardware breakpoint %" PRIu32 " removed from " TARGET_ADDR_FMT " (hwreg=%d)",
__func__, bp->unique_id, bp->address, hwbp_num);
return ERROR_OK;
}
static int set_swbp(struct target *t, struct breakpoint *bp)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
LOG_DEBUG("id %" PRIx32, bp->unique_id);
target_addr_t physaddr;
uint8_t opcode = SW_BP_OPCODE;
uint8_t readback;
if (calcaddr_physfromlin(t, bp->address, &physaddr) != ERROR_OK)
return ERROR_FAIL;
if (read_phys_mem(t, physaddr, 1, 1, bp->orig_instr))
return ERROR_FAIL;
LOG_DEBUG("set software breakpoint - orig byte=0x%02" PRIx8 "", *bp->orig_instr);
/* just write the instruction trap byte */
if (write_phys_mem(t, physaddr, 1, 1, &opcode))
return ERROR_FAIL;
/* verify that this is not invalid/read-only memory */
if (read_phys_mem(t, physaddr, 1, 1, &readback))
return ERROR_FAIL;
if (readback != SW_BP_OPCODE) {
LOG_ERROR("%s software breakpoint error at " TARGET_ADDR_FMT ", check memory",
__func__, bp->address);
LOG_ERROR("%s readback=0x%02" PRIx8 " orig=0x%02" PRIx8 "",
__func__, readback, *bp->orig_instr);
return ERROR_FAIL;
}
bp->is_set = true;
/* add the memory patch */
struct swbp_mem_patch *new_patch = malloc(sizeof(struct swbp_mem_patch));
if (!new_patch) {
LOG_ERROR("%s out of memory", __func__);
return ERROR_FAIL;
}
new_patch->next = NULL;
new_patch->orig_byte = *bp->orig_instr;
new_patch->physaddr = physaddr;
new_patch->swbp_unique_id = bp->unique_id;
struct swbp_mem_patch *addto = x86_32->swbbp_mem_patch_list;
if (!addto)
x86_32->swbbp_mem_patch_list = new_patch;
else {
while (addto->next)
addto = addto->next;
addto->next = new_patch;
}
LOG_USER("%s software breakpoint %" PRIu32 " set at " TARGET_ADDR_FMT,
__func__, bp->unique_id, bp->address);
return ERROR_OK;
}
static int unset_swbp(struct target *t, struct breakpoint *bp)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
LOG_DEBUG("id %" PRIx32, bp->unique_id);
target_addr_t physaddr;
uint8_t current_instr;
/* check that user program has not modified breakpoint instruction */
if (calcaddr_physfromlin(t, bp->address, &physaddr) != ERROR_OK)
return ERROR_FAIL;
if (read_phys_mem(t, physaddr, 1, 1, &current_instr))
return ERROR_FAIL;
if (current_instr == SW_BP_OPCODE) {
if (write_phys_mem(t, physaddr, 1, 1, bp->orig_instr))
return ERROR_FAIL;
} else {
LOG_ERROR("%s software breakpoint remove error at " TARGET_ADDR_FMT ", check memory",
__func__, bp->address);
LOG_ERROR("%s current=0x%02" PRIx8 " orig=0x%02" PRIx8 "",
__func__, current_instr, *bp->orig_instr);
return ERROR_FAIL;
}
/* remove from patch */
struct swbp_mem_patch *iter = x86_32->swbbp_mem_patch_list;
if (iter) {
if (iter->swbp_unique_id == bp->unique_id) {
/* it's the first item */
x86_32->swbbp_mem_patch_list = iter->next;
free(iter);
} else {
while (iter->next && iter->next->swbp_unique_id != bp->unique_id)
iter = iter->next;
if (iter->next) {
/* it's the next one */
struct swbp_mem_patch *freeme = iter->next;
iter->next = iter->next->next;
free(freeme);
}
}
}
LOG_USER("%s software breakpoint %" PRIu32 " removed from " TARGET_ADDR_FMT,
__func__, bp->unique_id, bp->address);
return ERROR_OK;
}
static int set_breakpoint(struct target *t, struct breakpoint *bp)
{
int error = ERROR_OK;
struct x86_32_common *x86_32 = target_to_x86_32(t);
LOG_DEBUG("type=%d, addr=" TARGET_ADDR_FMT, bp->type, bp->address);
if (bp->is_set) {
LOG_ERROR("breakpoint already set");
return error;
}
if (bp->type == BKPT_HARD) {
error = set_hwbp(t, bp);
if (error != ERROR_OK) {
LOG_ERROR("%s error setting hardware breakpoint at " TARGET_ADDR_FMT,
__func__, bp->address);
return error;
}
} else {
if (x86_32->sw_bpts_supported(t)) {
error = set_swbp(t, bp);
if (error != ERROR_OK) {
LOG_ERROR("%s error setting software breakpoint at " TARGET_ADDR_FMT,
__func__, bp->address);
return error;
}
} else {
LOG_ERROR("%s core doesn't support SW breakpoints", __func__);
return ERROR_FAIL;
}
}
return error;
}
static int unset_breakpoint(struct target *t, struct breakpoint *bp)
{
LOG_DEBUG("type=%d, addr=" TARGET_ADDR_FMT, bp->type, bp->address);
if (!bp->is_set) {
LOG_WARNING("breakpoint not set");
return ERROR_OK;
}
if (bp->type == BKPT_HARD) {
if (unset_hwbp(t, bp) != ERROR_OK) {
LOG_ERROR("%s error removing hardware breakpoint at " TARGET_ADDR_FMT,
__func__, bp->address);
return ERROR_FAIL;
}
} else {
if (unset_swbp(t, bp) != ERROR_OK) {
LOG_ERROR("%s error removing software breakpoint at " TARGET_ADDR_FMT,
__func__, bp->address);
return ERROR_FAIL;
}
}
bp->is_set = false;
return ERROR_OK;
}
static int set_watchpoint(struct target *t, struct watchpoint *wp)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
struct x86_32_dbg_reg *debug_reg_list = x86_32->hw_break_list;
int wp_num = 0;
LOG_DEBUG("type=%d, addr=" TARGET_ADDR_FMT, wp->rw, wp->address);
if (wp->is_set) {
LOG_ERROR("%s watchpoint already set", __func__);
return ERROR_OK;
}
if (wp->rw == WPT_READ) {
LOG_ERROR("%s no support for 'read' watchpoints, use 'access' or 'write'"
, __func__);
return ERROR_TARGET_RESOURCE_NOT_AVAILABLE;
}
while (debug_reg_list[wp_num].used && (wp_num < x86_32->num_hw_bpoints))
wp_num++;
if (wp_num >= x86_32->num_hw_bpoints) {
LOG_ERROR("%s no debug registers left", __func__);
return ERROR_TARGET_RESOURCE_NOT_AVAILABLE;
}
if (wp->length != 4 && wp->length != 2 && wp->length != 1) {
LOG_ERROR("%s only watchpoints of length 1, 2 or 4 are supported", __func__);
return ERROR_TARGET_RESOURCE_NOT_AVAILABLE;
}
switch (wp->rw) {
case WPT_WRITE:
if (set_debug_regs(t, wp->address, wp_num,
DR7_BP_WRITE, wp->length) != ERROR_OK) {
return ERROR_FAIL;
}
break;
case WPT_ACCESS:
if (set_debug_regs(t, wp->address, wp_num, DR7_BP_READWRITE,
wp->length) != ERROR_OK) {
return ERROR_FAIL;
}
break;
default:
LOG_ERROR("%s only 'access' or 'write' watchpoints are supported", __func__);
break;
}
watchpoint_set(wp, wp_num);
debug_reg_list[wp_num].used = 1;
debug_reg_list[wp_num].bp_value = wp->address;
LOG_USER("'%s' watchpoint %d set at " TARGET_ADDR_FMT " with length %" PRIu32 " (hwreg=%d)",
wp->rw == WPT_READ ? "read" : wp->rw == WPT_WRITE ?
"write" : wp->rw == WPT_ACCESS ? "access" : "?",
wp->unique_id, wp->address, wp->length, wp_num);
return ERROR_OK;
}
static int unset_watchpoint(struct target *t, struct watchpoint *wp)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
struct x86_32_dbg_reg *debug_reg_list = x86_32->hw_break_list;
LOG_DEBUG("type=%d, addr=" TARGET_ADDR_FMT, wp->rw, wp->address);
if (!wp->is_set) {
LOG_WARNING("watchpoint not set");
return ERROR_OK;
}
int wp_num = wp->number;
if (wp_num >= x86_32->num_hw_bpoints) {
LOG_DEBUG("Invalid FP Comparator number in watchpoint");
return ERROR_OK;
}
if (unset_debug_regs(t, wp_num) != ERROR_OK)
return ERROR_FAIL;
debug_reg_list[wp_num].used = 0;
debug_reg_list[wp_num].bp_value = 0;
wp->is_set = false;
LOG_USER("'%s' watchpoint %d removed from " TARGET_ADDR_FMT " with length %" PRIu32 " (hwreg=%d)",
wp->rw == WPT_READ ? "read" : wp->rw == WPT_WRITE ?
"write" : wp->rw == WPT_ACCESS ? "access" : "?",
wp->unique_id, wp->address, wp->length, wp_num);
return ERROR_OK;
}
/* after reset breakpoints and watchpoints in memory are not valid anymore and
* debug registers are cleared.
* we can't afford to remove sw breakpoints using the default methods as the
* memory doesn't have the same layout yet and an access might crash the target,
* so we just clear the openocd breakpoints structures.
*/
void x86_32_common_reset_breakpoints_watchpoints(struct target *t)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
struct x86_32_dbg_reg *debug_reg_list = x86_32->hw_break_list;
struct breakpoint *next_b;
struct watchpoint *next_w;
while (t->breakpoints) {
next_b = t->breakpoints->next;
free(t->breakpoints->orig_instr);
free(t->breakpoints);
t->breakpoints = next_b;
}
while (t->watchpoints) {
next_w = t->watchpoints->next;
free(t->watchpoints);
t->watchpoints = next_w;
}
for (int i = 0; i < x86_32->num_hw_bpoints; i++) {
debug_reg_list[i].used = 0;
debug_reg_list[i].bp_value = 0;
}
}
static int read_hw_reg_to_cache(struct target *t, int num)
{
uint32_t reg_value;
struct x86_32_common *x86_32 = target_to_x86_32(t);
if (check_not_halted(t))
return ERROR_TARGET_NOT_HALTED;
if ((num < 0) || (num >= x86_32->get_num_user_regs(t)))
return ERROR_COMMAND_SYNTAX_ERROR;
if (x86_32->read_hw_reg(t, num, &reg_value, 1) != ERROR_OK) {
LOG_ERROR("%s fail for %s", x86_32->cache->reg_list[num].name, __func__);
return ERROR_FAIL;
}
LOG_DEBUG("reg %s value 0x%08" PRIx32,
x86_32->cache->reg_list[num].name, reg_value);
return ERROR_OK;
}
static int write_hw_reg_from_cache(struct target *t, int num)
{
struct x86_32_common *x86_32 = target_to_x86_32(t);
if (check_not_halted(t))
return ERROR_TARGET_NOT_HALTED;
if ((num < 0) || (num >= x86_32->get_num_user_regs(t)))
return ERROR_COMMAND_SYNTAX_ERROR;
if (x86_32->write_hw_reg(t, num, 0, 1) != ERROR_OK) {
LOG_ERROR("%s fail for %s", x86_32->cache->reg_list[num].name, __func__);
return ERROR_FAIL;
}
LOG_DEBUG("reg %s value 0x%08" PRIx32, x86_32->cache->reg_list[num].name,
buf_get_u32(x86_32->cache->reg_list[num].value, 0, 32));
return ERROR_OK;
}
/* x86 32 commands */
static void handle_iod_output(struct command_invocation *cmd,
struct target *target, uint32_t address, unsigned size,
unsigned count, const uint8_t *buffer)
{
const unsigned line_bytecnt = 32;
unsigned line_modulo = line_bytecnt / size;
char output[line_bytecnt * 4 + 1];
unsigned output_len = 0;
const char *value_fmt;
switch (size) {
case 4:
value_fmt = "%8.8x ";
break;
case 2:
value_fmt = "%4.4x ";
break;
case 1:
value_fmt = "%2.2x ";
break;
default:
/* "can't happen", caller checked */
LOG_ERROR("%s invalid memory read size: %u", __func__, size);
return;
}
for (unsigned i = 0; i < count; i++) {
if (i % line_modulo == 0) {
output_len += snprintf(output + output_len,
sizeof(output) - output_len,
"0x%8.8x: ",
(unsigned)(address + (i*size)));
}
uint32_t value = 0;
const uint8_t *value_ptr = buffer + i * size;
switch (size) {
case 4:
value = target_buffer_get_u32(target, value_ptr);
break;
case 2:
value = target_buffer_get_u16(target, value_ptr);
break;
case 1:
value = *value_ptr;
}
output_len += snprintf(output + output_len,
sizeof(output) - output_len,
value_fmt, value);
if ((i % line_modulo == line_modulo - 1) || (i == count - 1)) {
command_print(cmd, "%s", output);
output_len = 0;
}
}
}
COMMAND_HANDLER(handle_iod_command)
{
if (CMD_ARGC != 1)
return ERROR_COMMAND_SYNTAX_ERROR;
uint32_t address;
COMMAND_PARSE_NUMBER(u32, CMD_ARGV[0], address);
if (address > 0xffff) {
LOG_ERROR("%s IA-32 I/O space is 2^16, 0x%08" PRIx32 " exceeds max", __func__, address);
return ERROR_COMMAND_SYNTAX_ERROR;
}
unsigned size = 0;
switch (CMD_NAME[2]) {
case 'w':
size = 4;
break;
case 'h':
size = 2;
break;
case 'b':
size = 1;
break;
default:
return ERROR_COMMAND_SYNTAX_ERROR;
}
unsigned count = 1;
uint8_t *buffer = calloc(count, size);
struct target *target = get_current_target(CMD_CTX);
int retval = x86_32_common_read_io(target, address, size, buffer);
if (retval == ERROR_OK)
handle_iod_output(CMD, target, address, size, count, buffer);
free(buffer);
return retval;
}
static int target_fill_io(struct target *target,
uint32_t address,
unsigned data_size,
/* value */
uint32_t b)
{
LOG_DEBUG("address=0x%08" PRIx32 ", data_size=%u, b=0x%08" PRIx32,
address, data_size, b);
uint8_t target_buf[data_size];
switch (data_size) {
case 4:
target_buffer_set_u32(target, target_buf, b);
break;
case 2:
target_buffer_set_u16(target, target_buf, b);
break;
case 1:
target_buf[0] = (b & 0x0ff);
break;
default:
exit(-1);
}
return x86_32_common_write_io(target, address, data_size, target_buf);
}
COMMAND_HANDLER(handle_iow_command)
{
if (CMD_ARGC != 2)
return ERROR_COMMAND_SYNTAX_ERROR;
uint32_t address;
COMMAND_PARSE_NUMBER(u32, CMD_ARGV[0], address);
uint32_t value;
COMMAND_PARSE_NUMBER(u32, CMD_ARGV[1], value);
struct target *target = get_current_target(CMD_CTX);
unsigned wordsize;
switch (CMD_NAME[2]) {
case 'w':
wordsize = 4;
break;
case 'h':
wordsize = 2;
break;
case 'b':
wordsize = 1;
break;
default:
return ERROR_COMMAND_SYNTAX_ERROR;
}
return target_fill_io(target, address, wordsize, value);
}
static const struct command_registration x86_32_exec_command_handlers[] = {
{
.name = "iww",
.mode = COMMAND_EXEC,
.handler = handle_iow_command,
.help = "write I/O port word",
.usage = "port data[word]",
},
{
.name = "iwh",
.mode = COMMAND_EXEC,
.handler = handle_iow_command,
.help = "write I/O port halfword",
.usage = "port data[halfword]",
},
{
.name = "iwb",
.mode = COMMAND_EXEC,
.handler = handle_iow_command,
.help = "write I/O port byte",
.usage = "port data[byte]",
},
{
.name = "idw",
.mode = COMMAND_EXEC,
.handler = handle_iod_command,
.help = "display I/O port word",
.usage = "port",
},
{
.name = "idh",
.mode = COMMAND_EXEC,
.handler = handle_iod_command,
.help = "display I/O port halfword",
.usage = "port",
},
{
.name = "idb",
.mode = COMMAND_EXEC,
.handler = handle_iod_command,
.help = "display I/O port byte",
.usage = "port",
},
COMMAND_REGISTRATION_DONE
};
const struct command_registration x86_32_command_handlers[] = {
{
.name = "x86_32",
.mode = COMMAND_ANY,
.help = "x86_32 target commands",
.usage = "",
.chain = x86_32_exec_command_handlers,
},
COMMAND_REGISTRATION_DONE
};