On Fri, 6 Dec 2024 at 03:11, Jerome Forissier jerome.forissier@linaro.org wrote:
When converting a U-Boot trace records file to ftrace function graph format for use by trace-cmd ('proftool -f funcgraph dump-ftrace'), the time associated to each function is incorrectly saved in microseconds instead of nanoseconds. Multuply the value by 1000 to fix the issue.
With this change, the trace-cmd output looks consistent. Here is an example with udelay(25) added to mem_malloc_init() as a test case:
$ ./tools/proftool -m System.map -t /tmp/trace.bin -f funcgraph \ dump-ftrace -o /tmp/trace.dat $ trace-cmd report /tmp/trace.dat >/tmp/trace.log $ vi /tmp/trace.log [...] u-boot-1 [000] 6.719659: funcgraph_entry: | mem_malloc_init() { u-boot-1 [000] 6.719659: funcgraph_entry: | udelay() { u-boot-1 [000] 6.719660: funcgraph_entry: | schedule() { u-boot-1 [000] 6.719660: funcgraph_entry: | cyclic_run() { u-boot-1 [000] 6.719660: funcgraph_entry: 1.000 us | cyclic_get_list(); u-boot-1 [000] 6.719661: funcgraph_exit: 1.000 us | } u-boot-1 [000] 6.719661: funcgraph_exit: 1.000 us | } u-boot-1 [000] 6.719661: funcgraph_entry: | __udelay() { u-boot-1 [000] 6.719662: funcgraph_entry: 0.000 us | usec_to_tick(); u-boot-1 [000] 6.719687: funcgraph_exit: + 26.000 us | } u-boot-1 [000] 6.719687: funcgraph_exit: + 28.000 us | } u-boot-1 [000] 6.719687: funcgraph_entry: # 37971.000 us | memset(); u-boot-1 [000] 6.757658: funcgraph_exit: # 37999.000 us | } u-boot-1 [000] 6.757658: funcgraph_exit: # 38000.000 us | }
In the above dump, the udelay() call is reported as taking 26 us which is consistent with the timestamps (6.719687 - 6.719659 = 0.000026). Without this patch we would have "0.026 us" instead of "+ 26.000 us".
Signed-off-by: Jerome Forissier jerome.forissier@linaro.org
tools/proftool.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
Reviewed-by: Simon Glass sjg@chromium.org