aboutsummaryrefslogtreecommitdiff
path: root/src/target/lakemont.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/target/lakemont.c')
-rw-r--r--src/target/lakemont.c1111
1 files changed, 1111 insertions, 0 deletions
diff --git a/src/target/lakemont.c b/src/target/lakemont.c
new file mode 100644
index 0000000..70c785c
--- /dev/null
+++ b/src/target/lakemont.c
@@ -0,0 +1,1111 @@
+/*
+ * 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 version 2 of the GNU General Public License as
+ * published by the Free Software Foundation.
+ *
+ * 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, write to the Free Software
+ * Foundation, Inc., 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Contact Information:
+ * Intel Corporation
+ */
+
+/*
+ * @file
+ * This implements the probemode operations for Lakemont 1 (LMT1).
+ */
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <helper/log.h>
+
+#include "target.h"
+#include "target_type.h"
+#include "lakemont.h"
+#include "register.h"
+#include "breakpoints.h"
+#include "x86_32_common.h"
+
+static int irscan(struct target *t, uint8_t *out,
+ uint8_t *in, uint8_t ir_len);
+static int drscan(struct target *t, uint8_t *out, uint8_t *in, uint8_t len);
+static int save_context(struct target *target);
+static int restore_context(struct target *target);
+static uint32_t get_tapstatus(struct target *t);
+static int enter_probemode(struct target *t);
+static int exit_probemode(struct target *t);
+static int halt_prep(struct target *t);
+static int do_halt(struct target *t);
+static int do_resume(struct target *t);
+static int read_all_core_hw_regs(struct target *t);
+static int write_all_core_hw_regs(struct target *t);
+static int read_hw_reg(struct target *t,
+ int reg, uint32_t *regval, uint8_t cache);
+static int write_hw_reg(struct target *t,
+ int reg, uint32_t regval, uint8_t cache);
+static struct reg_cache *lakemont_build_reg_cache
+ (struct target *target);
+static int submit_reg_pir(struct target *t, int num);
+static int submit_instruction_pir(struct target *t, int num);
+static int submit_pir(struct target *t, uint64_t op);
+static int lakemont_get_core_reg(struct reg *reg);
+static int lakemont_set_core_reg(struct reg *reg, uint8_t *buf);
+
+static struct scan_blk scan;
+
+/* registers and opcodes for register access, pm_idx is used to identify the
+ * registers that are modified for lakemont probemode specific operations
+ */
+static const struct {
+ uint8_t id;
+ const char *name;
+ uint64_t op;
+ uint8_t pm_idx;
+ unsigned bits;
+ enum reg_type type;
+ const char *group;
+ const char *feature;
+} regs[] = {
+ /* general purpose registers */
+ { EAX, "eax", 0x000000D01D660000, 0, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { ECX, "ecx", 0x000000501D660000, 1, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { EDX, "edx", 0x000000901D660000, 2, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { EBX, "ebx", 0x000000101D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { ESP, "esp", 0x000000E01D660000, NOT_PMREG, 32, REG_TYPE_DATA_PTR, "general", "org.gnu.gdb.i386.core" },
+ { EBP, "ebp", 0x000000601D660000, NOT_PMREG, 32, REG_TYPE_DATA_PTR, "general", "org.gnu.gdb.i386.core" },
+ { ESI, "esi", 0x000000A01D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { EDI, "edi", 0x000000201D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+
+ /* instruction pointer & flags */
+ { EIP, "eip", 0x000000C01D660000, 3, 32, REG_TYPE_CODE_PTR, "general", "org.gnu.gdb.i386.core" },
+ { EFLAGS, "eflags", 0x000000401D660000, 4, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+
+ /* segment registers */
+ { CS, "cs", 0x000000281D660000, 5, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { SS, "ss", 0x000000C81D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { DS, "ds", 0x000000481D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { ES, "es", 0x000000A81D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { FS, "fs", 0x000000881D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { GS, "gs", 0x000000081D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+
+ /* floating point unit registers - not accessible via JTAG - here to satisfy GDB */
+ { ST0, "st0", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { ST1, "st1", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { ST2, "st2", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { ST3, "st3", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { ST4, "st4", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { ST5, "st5", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { ST6, "st6", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { ST7, "st7", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { FCTRL, "fctrl", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { FSTAT, "fstat", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { FTAG, "ftag", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { FISEG, "fiseg", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { FIOFF, "fioff", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { FOSEG, "foseg", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { FOOFF, "fooff", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+ { FOP, "fop", 0x0, NOT_AVAIL_REG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.core" },
+
+ /* control registers */
+ { CR0, "cr0", 0x000000001D660000, 6, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { CR2, "cr2", 0x000000BC1D660000, 7, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { CR3, "cr3", 0x000000801D660000, 8, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { CR4, "cr4", 0x0000002C1D660000, 9, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+
+ /* debug registers */
+ { DR0, "dr0", 0x0000007C1D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { DR1, "dr1", 0x000000FC1D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { DR2, "dr2", 0x000000021D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { DR3, "dr3", 0x000000821D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { DR6, "dr6", 0x000000301D660000, 10, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { DR7, "dr7", 0x000000B01D660000, 11, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+
+ /* descriptor tables */
+ { IDTB, "idtbase", 0x000000581D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { IDTL, "idtlimit", 0x000000D81D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { IDTAR, "idtar", 0x000000981D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { GDTB, "gdtbase", 0x000000B81D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { GDTL, "gdtlimit", 0x000000781D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { GDTAR, "gdtar", 0x000000381D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { TR, "tr", 0x000000701D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { LDTR, "ldtr", 0x000000F01D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { LDTB, "ldbase", 0x000000041D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { LDTL, "ldlimit", 0x000000841D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { LDTAR, "ldtar", 0x000000F81D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+
+ /* segment registers */
+ { CSB, "csbase", 0x000000F41D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { CSL, "cslimit", 0x0000000C1D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { CSAR, "csar", 0x000000741D660000, 12, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { DSB, "dsbase", 0x000000941D660000, 13, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { DSL, "dslimit", 0x000000541D660000, 14, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { DSAR, "dsar", 0x000000141D660000, 15, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { ESB, "esbase", 0x0000004C1D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { ESL, "eslimit", 0x000000CC1D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { ESAR, "esar", 0x0000008C1D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { FSB, "fsbase", 0x000000641D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { FSL, "fslimit", 0x000000E41D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { FSAR, "fsar", 0x000000A41D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { GSB, "gsbase", 0x000000C41D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { GSL, "gslimit", 0x000000241D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { GSAR, "gsar", 0x000000441D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { SSB, "ssbase", 0x000000341D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { SSL, "sslimit", 0x000000B41D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { SSAR, "ssar", 0x000000D41D660000, 16, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { TSSB, "tssbase", 0x000000E81D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { TSSL, "tsslimit", 0x000000181D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ { TSSAR, "tssar", 0x000000681D660000, NOT_PMREG, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+ /* probemode control register */
+ { PMCR, "pmcr", 0x000000421D660000, 17, 32, REG_TYPE_INT32, "general", "org.gnu.gdb.i386.sys" },
+};
+
+static const struct {
+ uint8_t id;
+ const char *name;
+ uint64_t op;
+} instructions[] = {
+ /* memory read/write */
+ { MEMRDB32, "MEMRDB32", 0x0909090909090851 },
+ { MEMRDB16, "MEMRDB16", 0x09090909090851E6 },
+ { MEMRDH32, "MEMRDH32", 0x090909090908D166 },
+ { MEMRDH16, "MEMRDH16", 0x090909090908D1E6 },
+ { MEMRDW32, "MEMRDW32", 0x09090909090908D1 },
+ { MEMRDW16, "MEMRDW16", 0x0909090908D1E666 },
+ { MEMWRB32, "MEMWRB32", 0x0909090909090811 },
+ { MEMWRB16, "MEMWRB16", 0x09090909090811E6 },
+ { MEMWRH32, "MEMWRH32", 0x0909090909089166 },
+ { MEMWRH16, "MEMWRH16", 0x09090909090891E6 },
+ { MEMWRW32, "MEMWRW32", 0x0909090909090891 },
+ { MEMWRW16, "MEMWRW16", 0x090909090891E666 },
+ /* IO read/write */
+ { IORDB32, "IORDB32", 0x0909090909090937 },
+ { IORDB16, "IORDB16", 0x09090909090937E6 },
+ { IORDH32, "IORDH32", 0x090909090909B766 },
+ { IORDH16, "IORDH16", 0x090909090909B7E6 },
+ { IORDW32, "IORDW32", 0x09090909090909B7 },
+ { IORDW16, "IORDW16", 0x0909090909B7E666 },
+ { IOWRB32, "IOWRB32", 0x0909090909090977 },
+ { IOWRB16, "IOWRB16", 0x09090909090977E6 },
+ { IOWRH32, "IOWRH32", 0x090909090909F766 },
+ { IOWRH16, "IOWRH16", 0x090909090909F7E6 },
+ { IOWRW32, "IOWRW32", 0x09090909090909F7 },
+ { IOWRW16, "IOWRW16", 0x0909090909F7E666 },
+ /* lakemont1 core shadow ram access opcodes */
+ { SRAMACCESS, "SRAMACCESS", 0x0000000E9D660000 },
+ { SRAM2PDR, "SRAM2PDR", 0x4CF0000000000000 },
+ { PDR2SRAM, "PDR2SRAM", 0x0CF0000000000000 },
+ { WBINVD, "WBINVD", 0x09090909090990F0 },
+};
+
+bool check_not_halted(const struct target *t)
+{
+ bool halted = t->state == TARGET_HALTED;
+ if (!halted)
+ LOG_ERROR("target running, halt it first");
+ return !halted;
+}
+
+static int irscan(struct target *t, uint8_t *out,
+ uint8_t *in, uint8_t ir_len)
+{
+ int retval = ERROR_OK;
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ if (NULL == t->tap) {
+ retval = ERROR_FAIL;
+ LOG_ERROR("%s invalid target tap", __func__);
+ return retval;
+ }
+ if (ir_len != t->tap->ir_length) {
+ retval = ERROR_FAIL;
+ if (t->tap->enabled)
+ LOG_ERROR("%s tap enabled but tap irlen=%d",
+ __func__, t->tap->ir_length);
+ else
+ LOG_ERROR("%s tap not enabled and irlen=%d",
+ __func__, t->tap->ir_length);
+ return retval;
+ }
+ struct scan_field *fields = &scan.field;
+ fields->num_bits = ir_len;
+ fields->out_value = out;
+ fields->in_value = in;
+ jtag_add_ir_scan(x86_32->curr_tap, fields, TAP_IDLE);
+ if (x86_32->flush) {
+ retval = jtag_execute_queue();
+ if (retval != ERROR_OK)
+ LOG_ERROR("%s failed to execute queue", __func__);
+ }
+ return retval;
+}
+
+static int drscan(struct target *t, uint8_t *out, uint8_t *in, uint8_t len)
+{
+ int retval = ERROR_OK;
+ uint64_t data = 0;
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ if (NULL == t->tap) {
+ retval = ERROR_FAIL;
+ LOG_ERROR("%s invalid target tap", __func__);
+ return retval;
+ }
+ if (len > MAX_SCAN_SIZE || 0 == len) {
+ retval = ERROR_FAIL;
+ LOG_ERROR("%s data len is %d bits, max is %d bits",
+ __func__, len, MAX_SCAN_SIZE);
+ return retval;
+ }
+ struct scan_field *fields = &scan.field;
+ fields->out_value = out;
+ fields->in_value = in;
+ fields->num_bits = len;
+ jtag_add_dr_scan(x86_32->curr_tap, 1, fields, TAP_IDLE);
+ if (x86_32->flush) {
+ retval = jtag_execute_queue();
+ if (retval != ERROR_OK) {
+ LOG_ERROR("%s drscan failed to execute queue", __func__);
+ return retval;
+ }
+ }
+ if (in != NULL) {
+ if (len >= 8) {
+ for (int n = (len / 8) - 1 ; n >= 0; n--)
+ data = (data << 8) + *(in+n);
+ } else
+ LOG_DEBUG("dr in 0x%02" PRIx8, *in);
+ } else {
+ LOG_ERROR("%s no drscan data", __func__);
+ retval = ERROR_FAIL;
+ }
+ return retval;
+}
+
+static int save_context(struct target *t)
+{
+ int err;
+ /* read core registers from lakemont sram */
+ err = read_all_core_hw_regs(t);
+ if (err != ERROR_OK) {
+ LOG_ERROR("%s error reading regs", __func__);
+ return err;
+ }
+ return ERROR_OK;
+}
+
+static int restore_context(struct target *t)
+{
+ int err = ERROR_OK;
+ uint32_t i;
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+
+ /* write core regs into the core PM SRAM from the reg_cache */
+ err = write_all_core_hw_regs(t);
+ if (err != ERROR_OK) {
+ LOG_ERROR("%s error writing regs", __func__);
+ return err;
+ }
+
+ for (i = 0; i < (x86_32->cache->num_regs); i++) {
+ x86_32->cache->reg_list[i].dirty = 0;
+ x86_32->cache->reg_list[i].valid = 0;
+ }
+ return err;
+}
+
+/*
+ * we keep reg_cache in sync with hardware at halt/resume time, we avoid
+ * writing to real hardware here bacause pm_regs reflects the hardware
+ * while we are halted then reg_cache syncs with hw on resume
+ * TODO - in order for "reg eip force" to work it assume get/set reads
+ * and writes from hardware, may be other reasons also because generally
+ * other openocd targets read/write from hardware in get/set - watch this!
+ */
+static int lakemont_get_core_reg(struct reg *reg)
+{
+ int retval = ERROR_OK;
+ struct lakemont_core_reg *lakemont_reg = reg->arch_info;
+ struct target *t = lakemont_reg->target;
+ if (check_not_halted(t))
+ return ERROR_TARGET_NOT_HALTED;
+ LOG_DEBUG("reg=%s, value=%08" PRIx32, reg->name,
+ buf_get_u32(reg->value, 0, 32));
+ return retval;
+}
+
+static int lakemont_set_core_reg(struct reg *reg, uint8_t *buf)
+{
+ struct lakemont_core_reg *lakemont_reg = reg->arch_info;
+ struct target *t = lakemont_reg->target;
+ uint32_t value = buf_get_u32(buf, 0, 32);
+ LOG_DEBUG("reg=%s, newval=%08" PRIx32, reg->name, value);
+ if (check_not_halted(t))
+ return ERROR_TARGET_NOT_HALTED;
+ buf_set_u32(reg->value, 0, 32, value);
+ reg->dirty = 1;
+ reg->valid = 1;
+ return ERROR_OK;
+}
+
+static const struct reg_arch_type lakemont_reg_type = {
+ /* these get called if reg_cache doesnt have a "valid" value
+ * of an individual reg eg "reg eip" but not for "reg" block
+ */
+ .get = lakemont_get_core_reg,
+ .set = lakemont_set_core_reg,
+};
+
+struct reg_cache *lakemont_build_reg_cache(struct target *t)
+{
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ int num_regs = ARRAY_SIZE(regs);
+ struct reg_cache **cache_p = register_get_last_cache_p(&t->reg_cache);
+ struct reg_cache *cache = malloc(sizeof(struct reg_cache));
+ struct reg *reg_list = malloc(sizeof(struct reg) * num_regs);
+ struct lakemont_core_reg *arch_info = malloc(sizeof(struct lakemont_core_reg) * num_regs);
+ struct reg_feature *feature;
+ int i;
+
+ if (cache == NULL || reg_list == NULL || arch_info == NULL) {
+ free(cache);
+ free(reg_list);
+ free(arch_info);
+ LOG_ERROR("%s out of memory", __func__);
+ return NULL;
+ }
+
+ /* Build the process context cache */
+ cache->name = "lakemont registers";
+ cache->next = NULL;
+ cache->reg_list = reg_list;
+ cache->num_regs = num_regs;
+ (*cache_p) = cache;
+ x86_32->cache = cache;
+
+ for (i = 0; i < num_regs; i++) {
+ arch_info[i].target = t;
+ arch_info[i].x86_32_common = x86_32;
+ arch_info[i].op = regs[i].op;
+ arch_info[i].pm_idx = regs[i].pm_idx;
+ reg_list[i].name = regs[i].name;
+ reg_list[i].size = 32;
+ reg_list[i].value = calloc(1, 4);
+ reg_list[i].dirty = 0;
+ reg_list[i].valid = 0;
+ reg_list[i].type = &lakemont_reg_type;
+ reg_list[i].arch_info = &arch_info[i];
+
+ reg_list[i].group = regs[i].group;
+ reg_list[i].number = i;
+ 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 = regs[i].feature;
+ reg_list[i].feature = feature;
+ } else
+ LOG_ERROR("%s unable to allocate feature list", __func__);
+
+ 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 = regs[i].type;
+ else
+ LOG_ERROR("%s unable to allocate reg type list", __func__);
+ }
+ return cache;
+}
+
+static uint32_t get_tapstatus(struct target *t)
+{
+ scan.out[0] = TAPSTATUS;
+ if (irscan(t, scan.out, NULL, LMT_IRLEN) != ERROR_OK)
+ return 0;
+ if (drscan(t, NULL, scan.out, TS_SIZE) != ERROR_OK)
+ return 0;
+ return buf_get_u32(scan.out, 0, 32);
+}
+
+static int enter_probemode(struct target *t)
+{
+ uint32_t tapstatus = 0;
+ tapstatus = get_tapstatus(t);
+ LOG_DEBUG("TS before PM enter = %08" PRIx32, tapstatus);
+ if (tapstatus & TS_PM_BIT) {
+ LOG_DEBUG("core already in probemode");
+ return ERROR_OK;
+ }
+ scan.out[0] = PROBEMODE;
+ if (irscan(t, scan.out, NULL, LMT_IRLEN) != ERROR_OK)
+ return ERROR_FAIL;
+ scan.out[0] = 1;
+ if (drscan(t, scan.out, scan.in, 1) != ERROR_OK)
+ return ERROR_FAIL;
+ tapstatus = get_tapstatus(t);
+ LOG_DEBUG("TS after PM enter = %08" PRIx32, tapstatus);
+ if ((tapstatus & TS_PM_BIT) && (!(tapstatus & TS_EN_PM_BIT)))
+ return ERROR_OK;
+ else {
+ LOG_ERROR("%s PM enter error, tapstatus = %08" PRIx32
+ , __func__, tapstatus);
+ return ERROR_FAIL;
+ }
+}
+
+static int exit_probemode(struct target *t)
+{
+ uint32_t tapstatus = get_tapstatus(t);
+ LOG_DEBUG("TS before PM exit = %08" PRIx32, tapstatus);
+
+ if (!(tapstatus & TS_PM_BIT)) {
+ LOG_USER("core not in PM");
+ return ERROR_OK;
+ }
+ scan.out[0] = PROBEMODE;
+ if (irscan(t, scan.out, NULL, LMT_IRLEN) != ERROR_OK)
+ return ERROR_FAIL;
+ scan.out[0] = 0;
+ if (drscan(t, scan.out, scan.in, 1) != ERROR_OK)
+ return ERROR_FAIL;
+ return ERROR_OK;
+}
+
+/* do whats needed to properly enter probemode for debug on lakemont */
+static int halt_prep(struct target *t)
+{
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ if (write_hw_reg(t, DSB, PM_DSB, 0) != ERROR_OK)
+ return ERROR_FAIL;
+ LOG_DEBUG("write %s %08" PRIx32, regs[DSB].name, PM_DSB);
+ if (write_hw_reg(t, DSL, PM_DSL, 0) != ERROR_OK)
+ return ERROR_FAIL;
+ LOG_DEBUG("write %s %08" PRIx32, regs[DSL].name, PM_DSL);
+ if (write_hw_reg(t, DSAR, PM_DSAR, 0) != ERROR_OK)
+ return ERROR_FAIL;
+ LOG_DEBUG("write DSAR %08" PRIx32, PM_DSAR);
+ if (write_hw_reg(t, DR7, PM_DR7, 0) != ERROR_OK)
+ return ERROR_FAIL;
+ LOG_DEBUG("write DR7 %08" PRIx32, PM_DR7);
+
+ uint32_t eflags = buf_get_u32(x86_32->cache->reg_list[EFLAGS].value, 0, 32);
+ uint32_t csar = buf_get_u32(x86_32->cache->reg_list[CSAR].value, 0, 32);
+ uint32_t ssar = buf_get_u32(x86_32->cache->reg_list[SSAR].value, 0, 32);
+ uint32_t cr0 = buf_get_u32(x86_32->cache->reg_list[CR0].value, 0, 32);
+
+ /* clear VM86 and IF bits if they are set */
+ LOG_DEBUG("EFLAGS = %08" PRIx32 ", VM86 = %d, IF = %d", eflags,
+ eflags & EFLAGS_VM86 ? 1 : 0,
+ eflags & EFLAGS_IF ? 1 : 0);
+ if (eflags & EFLAGS_VM86
+ || eflags & EFLAGS_IF) {
+ x86_32->pm_regs[I(EFLAGS)] = eflags & ~(EFLAGS_VM86 | EFLAGS_IF);
+ if (write_hw_reg(t, EFLAGS, x86_32->pm_regs[I(EFLAGS)], 0) != ERROR_OK)
+ return ERROR_FAIL;
+ LOG_DEBUG("EFLAGS now = %08" PRIx32 ", VM86 = %d, IF = %d",
+ x86_32->pm_regs[I(EFLAGS)],
+ x86_32->pm_regs[I(EFLAGS)] & EFLAGS_VM86 ? 1 : 0,
+ x86_32->pm_regs[I(EFLAGS)] & EFLAGS_IF ? 1 : 0);
+ }
+
+ /* set CPL to 0 for memory access */
+ if (csar & CSAR_DPL) {
+ x86_32->pm_regs[I(CSAR)] = csar & ~CSAR_DPL;
+ if (write_hw_reg(t, CSAR, x86_32->pm_regs[I(CSAR)], 0) != ERROR_OK)
+ return ERROR_FAIL;
+ LOG_DEBUG("write CSAR_CPL to 0 %08" PRIx32, x86_32->pm_regs[I(CSAR)]);
+ }
+ if (ssar & SSAR_DPL) {
+ x86_32->pm_regs[I(SSAR)] = ssar & ~CSAR_DPL;
+ if (write_hw_reg(t, SSAR, x86_32->pm_regs[I(SSAR)], 0) != ERROR_OK)
+ return ERROR_FAIL;
+ LOG_DEBUG("write SSAR_CPL to 0 %08" PRIx32, x86_32->pm_regs[I(SSAR)]);
+ }
+
+ /* if cache's are enabled, disable and flush */
+ if (!(cr0 & CR0_CD)) {
+ LOG_DEBUG("caching enabled CR0 = %08" PRIx32, cr0);
+ if (cr0 & CR0_PG) {
+ x86_32->pm_regs[I(CR0)] = cr0 & ~CR0_PG;
+ if (write_hw_reg(t, CR0, x86_32->pm_regs[I(CR0)], 0) != ERROR_OK)
+ return ERROR_FAIL;
+ LOG_DEBUG("cleared paging CR0_PG = %08" PRIx32, x86_32->pm_regs[I(CR0)]);
+ /* submit wbinvd to flush cache */
+ if (submit_reg_pir(t, WBINVD) != ERROR_OK)
+ return ERROR_FAIL;
+ x86_32->pm_regs[I(CR0)] =
+ x86_32->pm_regs[I(CR0)] | (CR0_CD | CR0_NW | CR0_PG);
+ if (write_hw_reg(t, CR0, x86_32->pm_regs[I(CR0)], 0) != ERROR_OK)
+ return ERROR_FAIL;
+ LOG_DEBUG("set CD, NW and PG, CR0 = %08" PRIx32, x86_32->pm_regs[I(CR0)]);
+ }
+ }
+ return ERROR_OK;
+}
+
+static int do_halt(struct target *t)
+{
+ /* needs proper handling later if doing a halt errors out */
+ t->state = TARGET_DEBUG_RUNNING;
+ if (enter_probemode(t) != ERROR_OK)
+ return ERROR_FAIL;
+ if (save_context(t) != ERROR_OK)
+ return ERROR_FAIL;
+ if (halt_prep(t) != ERROR_OK)
+ return ERROR_FAIL;
+ t->state = TARGET_HALTED;
+
+ return target_call_event_callbacks(t, TARGET_EVENT_HALTED);
+}
+
+static int do_resume(struct target *t)
+{
+ /* needs proper handling later */
+ t->state = TARGET_DEBUG_RUNNING;
+ if (restore_context(t) != ERROR_OK)
+ return ERROR_FAIL;
+ if (exit_probemode(t) != ERROR_OK)
+ return ERROR_FAIL;
+ t->state = TARGET_RUNNING;
+
+ t->debug_reason = DBG_REASON_NOTHALTED;
+ LOG_USER("target running");
+
+ return target_call_event_callbacks(t, TARGET_EVENT_RESUMED);
+}
+
+static int read_all_core_hw_regs(struct target *t)
+{
+ int err;
+ uint32_t regval, i;
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ for (i = 0; i < (x86_32->cache->num_regs); i++) {
+ if (NOT_AVAIL_REG == regs[i].pm_idx)
+ continue;
+ err = read_hw_reg(t, regs[i].id, &regval, 1);
+ if (err != ERROR_OK) {
+ LOG_ERROR("%s error saving reg %s",
+ __func__, x86_32->cache->reg_list[i].name);
+ return err;
+ }
+ }
+ LOG_DEBUG("read_all_core_hw_regs read %d registers ok", i);
+ return ERROR_OK;
+}
+
+static int write_all_core_hw_regs(struct target *t)
+{
+ int err;
+ uint32_t i;
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ for (i = 0; i < (x86_32->cache->num_regs); i++) {
+ if (NOT_AVAIL_REG == regs[i].pm_idx)
+ continue;
+ err = write_hw_reg(t, i, 0, 1);
+ if (err != ERROR_OK) {
+ LOG_ERROR("%s error restoring reg %s",
+ __func__, x86_32->cache->reg_list[i].name);
+ return err;
+ }
+ }
+ LOG_DEBUG("write_all_core_hw_regs wrote %d registers ok", i);
+ return ERROR_OK;
+}
+
+/* read reg from lakemont core shadow ram, update reg cache if needed */
+static int read_hw_reg(struct target *t, int reg, uint32_t *regval, uint8_t cache)
+{
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ struct lakemont_core_reg *arch_info;
+ arch_info = x86_32->cache->reg_list[reg].arch_info;
+ x86_32->flush = 0; /* dont flush scans till we have a batch */
+ if (submit_reg_pir(t, reg) != ERROR_OK)
+ return ERROR_FAIL;
+ if (submit_instruction_pir(t, SRAMACCESS) != ERROR_OK)
+ return ERROR_FAIL;
+ if (submit_instruction_pir(t, SRAM2PDR) != ERROR_OK)
+ return ERROR_FAIL;
+ x86_32->flush = 1;
+ scan.out[0] = RDWRPDR;
+ if (irscan(t, scan.out, NULL, LMT_IRLEN) != ERROR_OK)
+ return ERROR_FAIL;
+ if (drscan(t, NULL, scan.out, PDR_SIZE) != ERROR_OK)
+ return ERROR_FAIL;
+
+ jtag_add_sleep(DELAY_SUBMITPIR);
+ *regval = buf_get_u32(scan.out, 0, 32);
+ if (cache) {
+ buf_set_u32(x86_32->cache->reg_list[reg].value, 0, 32, *regval);
+ x86_32->cache->reg_list[reg].valid = 1;
+ x86_32->cache->reg_list[reg].dirty = 0;
+ }
+ LOG_DEBUG("reg=%s, op=0x%016" PRIx64 ", val=%08" PRIx32,
+ x86_32->cache->reg_list[reg].name,
+ arch_info->op,
+ *regval);
+ return ERROR_OK;
+}
+
+/* write lakemont core shadow ram reg, update reg cache if needed */
+static int write_hw_reg(struct target *t, int reg, uint32_t regval, uint8_t cache)
+{
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ struct lakemont_core_reg *arch_info;
+ arch_info = x86_32->cache->reg_list[reg].arch_info;
+
+ uint8_t reg_buf[4];
+ if (cache)
+ regval = buf_get_u32(x86_32->cache->reg_list[reg].value, 0, 32);
+ buf_set_u32(reg_buf, 0, 32, regval);
+ LOG_DEBUG("reg=%s, op=0x%016" PRIx64 ", val=%08" PRIx32,
+ x86_32->cache->reg_list[reg].name,
+ arch_info->op,
+ regval);
+
+ scan.out[0] = RDWRPDR;
+ x86_32->flush = 0; /* dont flush scans till we have a batch */
+ if (irscan(t, scan.out, NULL, LMT_IRLEN) != ERROR_OK)
+ return ERROR_FAIL;
+ if (drscan(t, reg_buf, scan.out, PDR_SIZE) != ERROR_OK)
+ return ERROR_FAIL;
+ if (submit_reg_pir(t, reg) != ERROR_OK)
+ return ERROR_FAIL;
+ if (submit_instruction_pir(t, SRAMACCESS) != ERROR_OK)
+ return ERROR_FAIL;
+ x86_32->flush = 1;
+ if (submit_instruction_pir(t, PDR2SRAM) != ERROR_OK)
+ return ERROR_FAIL;
+
+ /* we are writing from the cache so ensure we reset flags */
+ if (cache) {
+ x86_32->cache->reg_list[reg].dirty = 0;
+ x86_32->cache->reg_list[reg].valid = 0;
+ }
+ return ERROR_OK;
+}
+
+static bool is_paging_enabled(struct target *t)
+{
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ if (x86_32->pm_regs[I(CR0)] & CR0_PG)
+ return true;
+ else
+ return false;
+}
+
+static uint8_t get_num_user_regs(struct target *t)
+{
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ return x86_32->cache->num_regs;
+}
+/* value of the CR0.PG (paging enabled) bit influences memory reads/writes */
+static int disable_paging(struct target *t)
+{
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ x86_32->pm_regs[I(CR0)] = x86_32->pm_regs[I(CR0)] & ~CR0_PG;
+ int err = x86_32->write_hw_reg(t, CR0, x86_32->pm_regs[I(CR0)], 0);
+ if (err != ERROR_OK) {
+ LOG_ERROR("%s error disabling paging", __func__);
+ return err;
+ }
+ return err;
+}
+
+static int enable_paging(struct target *t)
+{
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ x86_32->pm_regs[I(CR0)] = (x86_32->pm_regs[I(CR0)] | CR0_PG);
+ int err = x86_32->write_hw_reg(t, CR0, x86_32->pm_regs[I(CR0)], 0);
+ if (err != ERROR_OK) {
+ LOG_ERROR("%s error enabling paging", __func__);
+ return err;
+ }
+ return err;
+}
+
+static bool sw_bpts_supported(struct target *t)
+{
+ uint32_t tapstatus = get_tapstatus(t);
+ if (tapstatus & TS_SBP_BIT)
+ return true;
+ else
+ return false;
+}
+
+static int transaction_status(struct target *t)
+{
+ uint32_t tapstatus = get_tapstatus(t);
+ if ((TS_EN_PM_BIT | TS_PRDY_BIT) & tapstatus) {
+ LOG_ERROR("%s transaction error tapstatus = %08" PRIx32
+ , __func__, tapstatus);
+ return ERROR_FAIL;
+ } else {
+ return ERROR_OK;
+ }
+}
+
+static int submit_instruction(struct target *t, int num)
+{
+ int err = submit_instruction_pir(t, num);
+ if (err != ERROR_OK) {
+ LOG_ERROR("%s error submitting pir", __func__);
+ return err;
+ }
+ return err;
+}
+
+static int submit_reg_pir(struct target *t, int num)
+{
+ LOG_DEBUG("reg %s op=0x%016" PRIx64, regs[num].name, regs[num].op);
+ int err = submit_pir(t, regs[num].op);
+ if (err != ERROR_OK) {
+ LOG_ERROR("%s error submitting pir", __func__);
+ return err;
+ }
+ return err;
+}
+
+static int submit_instruction_pir(struct target *t, int num)
+{
+ LOG_DEBUG("%s op=0x%016" PRIx64, instructions[num].name,
+ instructions[num].op);
+ int err = submit_pir(t, instructions[num].op);
+ if (err != ERROR_OK) {
+ LOG_ERROR("%s error submitting pir", __func__);
+ return err;
+ }
+ return err;
+}
+
+/*
+ * PIR (Probe Mode Instruction Register), SUBMITPIR is an "IR only" TAP
+ * command; there is no corresponding data register
+ */
+static int submit_pir(struct target *t, uint64_t op)
+{
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+
+ uint8_t op_buf[8];
+ buf_set_u64(op_buf, 0, 64, op);
+ int flush = x86_32->flush;
+ x86_32->flush = 0;
+ scan.out[0] = WRPIR;
+ if (irscan(t, scan.out, NULL, LMT_IRLEN) != ERROR_OK)
+ return ERROR_FAIL;
+ if (drscan(t, op_buf, scan.out, PIR_SIZE) != ERROR_OK)
+ return ERROR_FAIL;
+ scan.out[0] = SUBMITPIR;
+ x86_32->flush = flush;
+ if (irscan(t, scan.out, NULL, LMT_IRLEN) != ERROR_OK)
+ return ERROR_FAIL;
+ jtag_add_sleep(DELAY_SUBMITPIR);
+ return ERROR_OK;
+}
+
+int lakemont_init_target(struct command_context *cmd_ctx, struct target *t)
+{
+ lakemont_build_reg_cache(t);
+ t->state = TARGET_RUNNING;
+ t->debug_reason = DBG_REASON_NOTHALTED;
+ return ERROR_OK;
+}
+
+int lakemont_init_arch_info(struct target *t, struct x86_32_common *x86_32)
+{
+ x86_32->submit_instruction = submit_instruction;
+ x86_32->transaction_status = transaction_status;
+ x86_32->read_hw_reg = read_hw_reg;
+ x86_32->write_hw_reg = write_hw_reg;
+ x86_32->sw_bpts_supported = sw_bpts_supported;
+ x86_32->get_num_user_regs = get_num_user_regs;
+ x86_32->is_paging_enabled = is_paging_enabled;
+ x86_32->disable_paging = disable_paging;
+ x86_32->enable_paging = enable_paging;
+ return ERROR_OK;
+}
+
+int lakemont_poll(struct target *t)
+{
+ /* LMT1 PMCR register currently allows code breakpoints, data breakpoints,
+ * single stepping and shutdowns to be redirected to PM but does not allow
+ * redirecting into PM as a result of SMM enter and SMM exit
+ */
+ uint32_t ts = get_tapstatus(t);
+
+ if (ts == 0xFFFFFFFF && t->state != TARGET_DEBUG_RUNNING) {
+ /* something is wrong here */
+ LOG_ERROR("tapstatus invalid - scan_chain serialization or locked JTAG access issues");
+ /* TODO: Give a hint that unlocking is wrong or maybe a
+ * 'jtag arp_init' helps
+ */
+ t->state = TARGET_DEBUG_RUNNING;
+ return ERROR_OK;
+ }
+
+ if (t->state == TARGET_HALTED && (!(ts & TS_PM_BIT))) {
+ LOG_INFO("target running for unknown reason");
+ t->state = TARGET_RUNNING;
+ }
+
+ if (t->state == TARGET_RUNNING &&
+ t->state != TARGET_DEBUG_RUNNING) {
+
+ if ((ts & TS_PM_BIT) && (ts & TS_PMCR_BIT)) {
+
+ LOG_DEBUG("redirect to PM, tapstatus=%08" PRIx32, get_tapstatus(t));
+
+ t->state = TARGET_DEBUG_RUNNING;
+ if (save_context(t) != ERROR_OK)
+ return ERROR_FAIL;
+ if (halt_prep(t) != ERROR_OK)
+ return ERROR_FAIL;
+ t->state = TARGET_HALTED;
+ t->debug_reason = DBG_REASON_UNDEFINED;
+
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ uint32_t eip = buf_get_u32(x86_32->cache->reg_list[EIP].value, 0, 32);
+ uint32_t dr6 = buf_get_u32(x86_32->cache->reg_list[DR6].value, 0, 32);
+ uint32_t hwbreakpoint = (uint32_t)-1;
+
+ if (dr6 & DR6_BRKDETECT_0)
+ hwbreakpoint = 0;
+ if (dr6 & DR6_BRKDETECT_1)
+ hwbreakpoint = 1;
+ if (dr6 & DR6_BRKDETECT_2)
+ hwbreakpoint = 2;
+ if (dr6 & DR6_BRKDETECT_3)
+ hwbreakpoint = 3;
+
+ if (hwbreakpoint != (uint32_t)-1) {
+ uint32_t dr7 = buf_get_u32(x86_32->cache->reg_list[DR7].value, 0, 32);
+ uint32_t type = dr7 & (0x03 << (DR7_RW_SHIFT + hwbreakpoint*DR7_RW_LEN_SIZE));
+ if (type == DR7_BP_EXECUTE) {
+ LOG_USER("hit hardware breakpoint (hwreg=%d) at 0x%08" PRIx32, hwbreakpoint, eip);
+ } else {
+ uint32_t address = 0;
+ switch (hwbreakpoint) {
+ default:
+ case 0:
+ address = buf_get_u32(x86_32->cache->reg_list[DR0].value, 0, 32);
+ break;
+ case 1:
+ address = buf_get_u32(x86_32->cache->reg_list[DR1].value, 0, 32);
+ break;
+ case 2:
+ address = buf_get_u32(x86_32->cache->reg_list[DR2].value, 0, 32);
+ break;
+ case 3:
+ address = buf_get_u32(x86_32->cache->reg_list[DR3].value, 0, 32);
+ break;
+ }
+ LOG_USER("hit '%s' watchpoint for 0x%08" PRIx32 " (hwreg=%d) at 0x%08" PRIx32,
+ type == DR7_BP_WRITE ? "write" : "access", address,
+ hwbreakpoint, eip);
+ }
+ t->debug_reason = DBG_REASON_BREAKPOINT;
+ } else {
+ /* Check if the target hit a software breakpoint.
+ * ! Watch out: EIP is currently pointing after the breakpoint opcode
+ */
+ struct breakpoint *bp = NULL;
+ bp = breakpoint_find(t, eip-1);
+ if (bp != NULL) {
+ t->debug_reason = DBG_REASON_BREAKPOINT;
+ if (bp->type == BKPT_SOFT) {
+ /* The EIP is now pointing the the next byte after the
+ * breakpoint instruction. This needs to be corrected.
+ */
+ buf_set_u32(x86_32->cache->reg_list[EIP].value, 0, 32, eip-1);
+ x86_32->cache->reg_list[EIP].dirty = 1;
+ x86_32->cache->reg_list[EIP].valid = 1;
+ LOG_USER("hit software breakpoint at 0x%08" PRIx32, eip-1);
+ } else {
+ /* it's not a hardware breakpoint (checked already in DR6 state)
+ * and it's also not a software breakpoint ...
+ */
+ LOG_USER("hit unknown breakpoint at 0x%08" PRIx32, eip);
+ }
+ } else {
+
+ /* There is also the case that we hit an breakpoint instruction,
+ * which was not set by us. This needs to be handled be the
+ * application that introduced the breakpoint.
+ */
+
+ LOG_USER("unknown break reason at 0x%08" PRIx32, eip);
+ }
+ }
+
+ return target_call_event_callbacks(t, TARGET_EVENT_HALTED);
+ }
+ }
+ return ERROR_OK;
+}
+
+int lakemont_arch_state(struct target *t)
+{
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+
+ LOG_USER("target halted due to %s at 0x%08" PRIx32 " in %s mode",
+ debug_reason_name(t),
+ buf_get_u32(x86_32->cache->reg_list[EIP].value, 0, 32),
+ (buf_get_u32(x86_32->cache->reg_list[CR0].value, 0, 32) & CR0_PE) ? "protected" : "real");
+
+ return ERROR_OK;
+}
+
+int lakemont_halt(struct target *t)
+{
+ if (t->state == TARGET_RUNNING) {
+ t->debug_reason = DBG_REASON_DBGRQ;
+ if (do_halt(t) != ERROR_OK)
+ return ERROR_FAIL;
+ return ERROR_OK;
+ } else {
+ LOG_ERROR("%s target not running", __func__);
+ return ERROR_FAIL;
+ }
+}
+
+int lakemont_resume(struct target *t, int current, uint32_t address,
+ int handle_breakpoints, int debug_execution)
+{
+ struct breakpoint *bp = NULL;
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+
+ if (check_not_halted(t))
+ return ERROR_TARGET_NOT_HALTED;
+ /* TODO lakemont_enable_breakpoints(t); */
+ if (t->state == TARGET_HALTED) {
+
+ /* running away for a software breakpoint needs some special handling */
+ uint32_t eip = buf_get_u32(x86_32->cache->reg_list[EIP].value, 0, 32);
+ bp = breakpoint_find(t, eip);
+ if (bp != NULL /*&& bp->type == BKPT_SOFT*/) {
+ /* the step will step over the breakpoint */
+ if (lakemont_step(t, 0, 0, 1) != ERROR_OK) {
+ LOG_ERROR("%s stepping over a software breakpoint at 0x%08" PRIx32 " "
+ "failed to resume the target", __func__, eip);
+ return ERROR_FAIL;
+ }
+ }
+
+ /* if breakpoints are enabled, we need to redirect these into probe mode */
+ struct breakpoint *activeswbp = t->breakpoints;
+ while (activeswbp != NULL && activeswbp->set == 0)
+ activeswbp = activeswbp->next;
+ struct watchpoint *activehwbp = t->watchpoints;
+ while (activehwbp != NULL && activehwbp->set == 0)
+ activehwbp = activehwbp->next;
+ if (activeswbp != NULL || activehwbp != NULL)
+ buf_set_u32(x86_32->cache->reg_list[PMCR].value, 0, 32, 1);
+ if (do_resume(t) != ERROR_OK)
+ return ERROR_FAIL;
+ } else {
+ LOG_USER("target not halted");
+ return ERROR_FAIL;
+ }
+ return ERROR_OK;
+}
+
+int lakemont_step(struct target *t, int current,
+ uint32_t address, int handle_breakpoints)
+{
+ struct x86_32_common *x86_32 = target_to_x86_32(t);
+ uint32_t eflags = buf_get_u32(x86_32->cache->reg_list[EFLAGS].value, 0, 32);
+ uint32_t eip = buf_get_u32(x86_32->cache->reg_list[EIP].value, 0, 32);
+ uint32_t pmcr = buf_get_u32(x86_32->cache->reg_list[PMCR].value, 0, 32);
+ struct breakpoint *bp = NULL;
+ int retval = ERROR_OK;
+ uint32_t tapstatus = 0;
+
+ if (check_not_halted(t))
+ return ERROR_TARGET_NOT_HALTED;
+ bp = breakpoint_find(t, eip);
+ if (retval == ERROR_OK && bp != NULL/*&& bp->type == BKPT_SOFT*/) {
+ /* TODO: This should only be done for software breakpoints.
+ * Stepping from hardware breakpoints should be possible with the resume flag
+ * Needs testing.
+ */
+ retval = x86_32_common_remove_breakpoint(t, bp);
+ }
+
+ /* Set EFLAGS[TF] and PMCR[IR], exit pm and wait for PRDY# */
+ LOG_DEBUG("modifying PMCR = %d and EFLAGS = %08" PRIx32, pmcr, eflags);
+ eflags = eflags | (EFLAGS_TF | EFLAGS_RF);
+ buf_set_u32(x86_32->cache->reg_list[EFLAGS].value, 0, 32, eflags);
+ buf_set_u32(x86_32->cache->reg_list[PMCR].value, 0, 32, 1);
+ LOG_DEBUG("EFLAGS [TF] [RF] bits set=%08" PRIx32 ", PMCR=%d, EIP=%08" PRIx32,
+ eflags, pmcr, eip);
+
+ tapstatus = get_tapstatus(t);
+
+ t->debug_reason = DBG_REASON_SINGLESTEP;
+ t->state = TARGET_DEBUG_RUNNING;
+ if (restore_context(t) != ERROR_OK)
+ return ERROR_FAIL;
+ if (exit_probemode(t) != ERROR_OK)
+ return ERROR_FAIL;
+
+ target_call_event_callbacks(t, TARGET_EVENT_RESUMED);
+
+ tapstatus = get_tapstatus(t);
+ if (tapstatus & (TS_PM_BIT | TS_EN_PM_BIT | TS_PRDY_BIT | TS_PMCR_BIT)) {
+ /* target has stopped */
+ if (save_context(t) != ERROR_OK)
+ return ERROR_FAIL;
+ if (halt_prep(t) != ERROR_OK)
+ return ERROR_FAIL;
+ t->state = TARGET_HALTED;
+
+ LOG_USER("step done from EIP 0x%08" PRIx32 " to 0x%08" PRIx32, eip,
+ buf_get_u32(x86_32->cache->reg_list[EIP].value, 0, 32));
+ target_call_event_callbacks(t, TARGET_EVENT_HALTED);
+ } else {
+ /* target didn't stop
+ * I hope the poll() will catch it, but the deleted breakpoint is gone
+ */
+ LOG_ERROR("%s target didn't stop after executing a single step", __func__);
+ t->state = TARGET_RUNNING;
+ return ERROR_FAIL;
+ }
+
+ /* try to re-apply the breakpoint, even of step failed
+ * TODO: When a bp was set, we should try to stop the target - fix the return above
+ */
+ if (bp != NULL/*&& bp->type == BKPT_SOFT*/) {
+ /* TODO: This should only be done for software breakpoints.
+ * Stepping from hardware breakpoints should be possible with the resume flag
+ * Needs testing.
+ */
+ retval = x86_32_common_add_breakpoint(t, bp);
+ }
+
+ return retval;
+}
+
+/* TODO - implement resetbreak fully through CLTAP registers */
+int lakemont_reset_assert(struct target *t)
+{
+ LOG_DEBUG("-");
+ return ERROR_OK;
+}
+
+int lakemont_reset_deassert(struct target *t)
+{
+ LOG_DEBUG("-");
+ return ERROR_OK;
+}