hla_target: allow non-intrusive profiling on cortex-m

Leverages the existing work that added profiling via DWT_PCSR.

hla_target doesn't have direct access to the mem_ap for doing a bulk
repeated read, but simply reading the DWT_PCSR register repeatedly is
still ~2 order of magnitude faster than halt/resume.

Change-Id: Ibe451aa95143694398370fdad6939cfb6191d56f
Signed-off-by: Karl Palsson <karlp@tweak.net.au>
Reviewed-on: http://openocd.zylin.com/4220
Tested-by: jenkins
Reviewed-by: Paul Fertser <fercerpav@gmail.com>

diff --git a/src/target/cortex_m.c b/src/target/cortex_m.c
index 988323b042132592169aa7723a775f1cf44e58ce..24063a7e6af06dc49bbf083cb0c72a6f05226dcc 100644 (file)
--- a/src/target/cortex_m.c
+++ b/src/target/cortex_m.c
@@ -1707,7 +1707,7 @@ void cortex_m_deinit_target(struct target *target)
        free(cortex_m);
 }
 
-static int cortex_m_profiling(struct target *target, uint32_t *samples,
+int cortex_m_profiling(struct target *target, uint32_t *samples,
                              uint32_t max_num_samples, uint32_t *num_samples, uint32_t seconds)
 {
        struct timeval timeout, now;
@@ -1749,13 +1749,18 @@ static int cortex_m_profiling(struct target *target, uint32_t *samples,
 
        for (;;) {
                if (use_pcsr) {
-                       uint32_t read_count = max_num_samples - sample_count;
-                       if (read_count > 1024)
-                               read_count = 1024;
-                       retval = mem_ap_read_buf_noincr(armv7m->debug_ap,
+                       if (armv7m && armv7m->debug_ap) {
+                               uint32_t read_count = max_num_samples - sample_count;
+                               if (read_count > 1024)
+                                       read_count = 1024;
+
+                               retval = mem_ap_read_buf_noincr(armv7m->debug_ap,
                                                        (void *)&samples[sample_count],
                                                        4, read_count, DWT_PCSR);
-                       sample_count += read_count;
+                               sample_count += read_count;
+                       } else {
+                               target_read_u32(target, DWT_PCSR, &samples[sample_count++]);
+                       }
                } else {
                        target_poll(target);
                        if (target->state == TARGET_HALTED) {

diff --git a/src/target/cortex_m.h b/src/target/cortex_m.h
index dbf9ac8a7f00f707b1a0c46332208cc3d35eb307..9500acc1d7d6363035bba13e078950dc3c02581a 100644 (file)
--- a/src/target/cortex_m.h
+++ b/src/target/cortex_m.h
@@ -213,5 +213,7 @@ void cortex_m_enable_breakpoints(struct target *target);
 void cortex_m_enable_watchpoints(struct target *target);
 void cortex_m_dwt_setup(struct cortex_m_common *cm, struct target *target);
 void cortex_m_deinit_target(struct target *target);
+int cortex_m_profiling(struct target *target, uint32_t *samples,
+       uint32_t max_num_samples, uint32_t *num_samples, uint32_t seconds);
 
 #endif /* OPENOCD_TARGET_CORTEX_M_H */

diff --git a/src/target/hla_target.c b/src/target/hla_target.c
index 78dc8c5128228ef12e22c5636eeb80ae28f3d62a..a3e683597bc0126e886783819931ad751ac84ca2 100644 (file)
--- a/src/target/hla_target.c
+++ b/src/target/hla_target.c
@@ -814,4 +814,5 @@ struct target_type hla_target = {
        .remove_breakpoint = cortex_m_remove_breakpoint,
        .add_watchpoint = cortex_m_add_watchpoint,
        .remove_watchpoint = cortex_m_remove_watchpoint,
+       .profiling = cortex_m_profiling,
 };