Add 4 more cases for Intel PT.

1. Add Intel PT sanpshot mode test.
2. Add Intel PT exclude user trace mode test.
3. Add Intel PT exclude kernel trace mode test.
4. Add Intel PT disable branch trace mode test.

Signed-off-by: Ammy Yi <ammy.yi@intel.com>
---
 runtest/tracing                            |  4 ++
 testcases/kernel/tracing/pt_test/pt_test.c | 79 +++++++++++++++++++++++++-----
 2 files changed, 71 insertions(+), 12 deletions(-)

Hi!
> +/* 10/2018 Add full trace check test case */
> +/* 2/2019  Add snapshot/disable branch/user/kernel trace */

We keep the history in git, please do not add comments like these.

>  #include <sched.h>
>  #include <stdlib.h>
>  #include <stdio.h>
> @@ -40,22 +42,36 @@ int fde = -1;
>  //map head and size
>  uint64_t **bufm;
>  long buhsz;
> -
> -static uint64_t **create_map(int fde, long bufsize)
> +char *str_mode;
> +char *str_exclude_info;
> +char *str_branch_flag;
> +int mode = 1;
> +int exclude_info = 3;
> +int branch_flag = 1;
> +
> +static uint64_t **create_map(int fde, long bufsize, int flag)
                                                           ^
							   Can we call
							   this rw_flag?

>  {
>  	uint64_t **buf_ev;
> +	int pro_flag;
>  	struct perf_event_mmap_page *pc;
>  
>  	buf_ev = SAFE_MALLOC(2*sizeof(uint64_t *));
>  	buf_ev[0] = NULL;
>  	buf_ev[1] = NULL;
> +	if (flag == 1) {
> +		tst_res(TINFO, "memory will be r/w!");
                                ^
				Can you be more specific here?

You map one of the buffers read only if the flag is not set but the
option is called full/snapshot trace, it does not explain the difference
much.

> +		pro_flag = PROT_READ | PROT_WRITE;
> +	} else {
> +		tst_res(TINFO, "memory will be r only!");
> +		pro_flag = PROT_READ;
> +	}
>  	buf_ev[0] = SAFE_MMAP(NULL, INTEL_PT_MEMSIZE, PROT_READ | PROT_WRITE,
>  							MAP_SHARED, fde, 0);
>  
>  	pc = (struct perf_event_mmap_page *)buf_ev[0];
>  	pc->aux_offset = INTEL_PT_MEMSIZE;
>  	pc->aux_size = bufsize;
> -	buf_ev[1] = SAFE_MMAP(NULL, bufsize, PROT_READ | PROT_WRITE,
> +	buf_ev[1] = SAFE_MMAP(NULL, bufsize, pro_flag,
>  					MAP_SHARED, fde, INTEL_PT_MEMSIZE);
>  	return buf_ev;
>  }
> @@ -89,7 +105,7 @@ static void del_map(uint64_t **buf_ev, long bufsize)
>  	free(buf_ev);
>  }
>  
> -static void intel_pt_full_trace_check(void)
> +static void intel_pt_trace_check(void)
>  {
>  	uint64_t aux_head = 0;
>  	struct perf_event_mmap_page *pmp;
> @@ -108,7 +124,19 @@ static void intel_pt_full_trace_check(void)
>  		return;
>  	}
>  
> -	tst_res(TPASS, "perf trace full mode is passed!");
> +	tst_res(TPASS, "perf trace test is passed!");
> +}
> +
> +static void parse_options(void)
> +{
> +	if (tst_parse_int(str_mode, &mode, 0, 1))
> +		tst_brk(TBROK, "Invalid mode '%s'", str_mode);
> +
> +	if (tst_parse_int(str_exclude_info, &exclude_info, 1, 2))
> +		tst_brk(TBROK, "Invalid exclude info '%s'", str_exclude_info);
> +
> +	if (tst_parse_int(str_branch_flag, &branch_flag, 0, 1))
> +		tst_brk(TBROK, "Invalid branch flag '%s'", str_branch_flag);
>  }
>  
>  static void setup(void)
> @@ -116,6 +144,9 @@ static void setup(void)
>  	struct perf_event_attr attr = {};
>  
>  	buhsz = 2 * PAGESIZE;
> +
> +	parse_options();
> +
>  	if (access(INTEL_PT_PATH, F_OK)) {
>  		tst_brk(TCONF,
>  			"Requires Intel Core 5th+ generation (Broadwell and newer)"
> @@ -130,9 +161,24 @@ static void setup(void)
>  	attr.config	= BIT(intel_pt_pmu_value(INTEL_PT_FORMAT_TSC)) |
>  				BIT(intel_pt_pmu_value(INTEL_PT_FORMAT_NRT));
>  	attr.size	= sizeof(struct perf_event_attr);
> -	attr.exclude_kernel		= 0;
> -	attr.exclude_user		= 0;
> -	attr.mmap			= 1;
> +	attr.mmap	= 1;

Removing whitespaces here is kind of useless change.

> +	if (branch_flag == 0) {
> +		tst_res(TINFO, "Intel PT will disable branch trace!");
                                                                  ^
				Why does all the messages include
				exclamation mark? These are just info
				messages anyways.

> +		attr.config |= 1;
> +	}
> +
> +	attr.exclude_kernel	= 0;
> +	attr.exclude_user	= 0;
> +
> +	if (exclude_info == 1) {
> +		tst_res(TINFO, "Intel PT will exclude user trace!");
> +		attr.exclude_user = 1;
> +	}
> +
> +	if (exclude_info == 2) {
> +		tst_res(TINFO, "Intel PT will exclude kernel trace!");
> +		attr.exclude_kernel = 1;
> +	}
>  
>  	/* only get trace for own pid */
>  	fde = tst_syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0);
> @@ -142,7 +188,7 @@ static void setup(void)
>  		return;
>  	}
>  	bufm = NULL;
> -	bufm = create_map(fde, buhsz);
> +	bufm = create_map(fde, buhsz, mode);
>  
>  }
>  
> @@ -154,8 +200,17 @@ static void cleanup(void)
>  	del_map(bufm, buhsz);
>  }
>  
> +static struct tst_option options[] = {
> +	{"m:", &str_mode, "-m different mode, as full trace or snapshot trace"},

This parameter is binary switch so there is no point having it as a flag
with a value. We can just pass "m" instead of "m:" there then do if ()
on the string pointer passed here.

> +	{"e:", &str_exclude_info, "-e exclude info, 1->user, 2->kernel"},

It would be a bit more user-friendly if we passed strings "user" and
"kernel" insteead of 1 and 2 as the flag parameters.

> +	{"b:", &str_branch_flag, "-b if disable branch trace"},

Here as well.

> +	{NULL, NULL, NULL}
> +};
> +
> +
>  static struct tst_test test = {
> -	.test_all = intel_pt_full_trace_check,
> +	.test_all = intel_pt_trace_check,
> +	.options = options,
>  	.min_kver = "4.1",
>  	.setup = setup,
>  	.cleanup = cleanup,
> -- 
> 2.14.1
> 
> 
> -- 
> Mailing list info: https://lists.linux.it/listinfo/ltp