On Mon, 24 Mar 2014 16:57:18 -0400, Don Zickus wrote:
> In order for the c2c tool to work correctly, it needs to properly
> sort all the records on uniquely identifiable data addresses. These
> unique addresses are converted from virtual addresses provided by the
> hardware into a kernel address using an mmap2 record as the decoder.
>
> Once a unique address is converted, we can sort on them based on
> various rules. Then it becomes clear which address are overlapping
> with each other across mmap regions or pid spaces.
>
> This patch just creates the rules and inserts the records into a
> sort entry for safe keeping until later patches process them.
>
> The general sorting rule is:
>
> o group cpumodes together
> o if (nonzero major/minor number - ie mmap'd areas)
> o sort on major, minor, inode, inode generation numbers
> o else if cpumode is not kernel
> o sort on pid
> o sort on data addresses
>
> I also hacked in the concept of 'color'. The purpose of that bit is to
> provides hints later when processing these records that indicate a new unique
> address has been encountered. Because later processing only checks the data
> addresses, there can be a theoretical scenario that similar sequential data
> addresses (when walking the rbtree) could be misinterpreted as overlapping
> when in fact they are not.
>
> Sample output: (perf report --stdio --physid-mode)
>
> Overhead Data Address Source Address Command: Pid Tid Major Minor Inode Inode Gen
> ........ ...................... ........................ ................. ..... ..... ..... ....... .........
> 18.93% [k] 0xffffc900139c40b0 [k] igb_update_stats kworker/0:1: 257 257 0 0 0 0
> 7.63% [k] 0xffff88082e6cf0a8 [k] watchdog_timer_fn swapper: 0 0 0 0 0 0
> 1.86% [k] 0xffff88042ef94700 [k] _raw_spin_lock swapper: 0 0 0 0 0 0
> 1.77% [k] 0xffff8804278afa50 [k] __switch_to swapper: 0 0 0 0 0 0
>
> V4: add manpage entry in perf-report
>
> V3: split out the sorting into unique entries. This makes it look
> far less ugly
> create a new 'physid mode' to group all the sorting rules together
> (mimics the mem-mode)
What is 'physid' then? I guess you meant physical id but it seems
unique id or unique map id looks like a better fit IMHO.
>
> Signed-off-by: Don Zickus <[email protected]>
> ---
> tools/perf/Documentation/perf-report.txt | 23 +++
> tools/perf/builtin-report.c | 20 ++-
> tools/perf/util/hist.c | 27 ++-
> tools/perf/util/hist.h | 8 +
> tools/perf/util/sort.c | 294 +++++++++++++++++++++++++++++++
> tools/perf/util/sort.h | 13 ++
> 6 files changed, 381 insertions(+), 4 deletions(-)
>
> diff --git a/tools/perf/Documentation/perf-report.txt b/tools/perf/Documentation/perf-report.txt
> index 8eab8a4..01391b0 100644
> --- a/tools/perf/Documentation/perf-report.txt
> +++ b/tools/perf/Documentation/perf-report.txt
> @@ -95,6 +95,23 @@ OPTIONS
> And default sort keys are changed to comm, dso_from, symbol_from, dso_to
> and symbol_to, see '--branch-stack'.
>
> + If --physid-mode option is used, following sort keys are also
> + available:
> + daddr, iaddr, pid, tid, major, minor, inode, inode_gen.
> +
> + - daddr: data address (sorted based on major, minor, inode and inode
> + generation numbers if shared, otherwise pid)
By "if shared", did you mean "for shared file mapping"?
> + - iaddr: instruction address
> + - pid: command and pid of the task
> + - tid: tid of the task
> + - major: major number of mapped location (0 if not mapped)
> + - minor: minor number of mapped location (0 if not mapped)
> + - inode: inode number of mapped location (0 if not mapped)
> + - inode_gen: inode generation number of mapped location (0 if not mapped)
s/if not mapped/if not file-mapped/ ?
> +
> + And default sort keys are changed to daddr, iaddr, pid, tid, major,
> + minor, inode and inode_gen, see '--physid-mode'.
> +
> -p::
> --parent=<regex>::
> A regex filter to identify parent. The parent is a caller of this
> @@ -223,6 +240,12 @@ OPTIONS
> branch stacks and it will automatically switch to the branch view mode,
> unless --no-branch-stack is used.
>
> +--physid-mode::
> + Use the data addresses sampled using perf record -d and combine them
> + with the mmap'd area region where they are located. This helps identify
> + which data addresses collide with similar addresses in another process
> + space. See --sort for output choices.
> +
> --objdump=<path>::
> Path to objdump binary.
>
> diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
> index c87412b..093f5ad 100644
> --- a/tools/perf/builtin-report.c
> +++ b/tools/perf/builtin-report.c
> @@ -49,6 +49,7 @@ struct report {
> bool show_threads;
> bool inverted_callchain;
> bool mem_mode;
> + bool physid_mode;
> bool header;
> bool header_only;
> int max_stack;
> @@ -241,7 +242,7 @@ static int process_sample_event(struct perf_tool *tool,
> ret = report__add_branch_hist_entry(rep, &al, sample, evsel);
> if (ret < 0)
> pr_debug("problem adding lbr entry, skipping event\n");
> - } else if (rep->mem_mode == 1) {
> + } else if ((rep->mem_mode == 1) || (rep->physid_mode)) {
As you can see rep->mem_mode is also a boolean field. Please change it
like:
} else if (rep->mem_mode || rep->physid_mode) {
> ret = report__add_mem_hist_entry(rep, &al, sample, evsel);
> if (ret < 0)
> pr_debug("problem adding mem entry, skipping event\n");
> @@ -746,6 +747,7 @@ int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
> OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
> "Disable symbol demangling"),
> OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
> + OPT_BOOLEAN(0, "physid-mode", &report.physid_mode, "physid access profile"),
> OPT_CALLBACK(0, "percent-limit", &report, "percent",
> "Don't show entries under that percent", parse_percent_limit),
> OPT_END()
> @@ -817,6 +819,22 @@ repeat:
> sort_order = "local_weight,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked";
> }
>
> + if (report.physid_mode) {
> + if ((sort__mode == SORT_MODE__BRANCH) ||
> + (sort__mode == SORT_MODE__MEMORY)) {
> + pr_err("branch or memory and physid mode incompatible\n");
> + goto error;
> + }
> + sort__mode = SORT_MODE__PHYSID;
> +
> + /*
> + * if no sort_order is provided, then specify
> + * branch-mode specific order
s/branch-mode/physid-mode/
It looks mem-mode has the same copy-n-paste problem.
> + */
> + if (sort_order == default_sort_order)
> + sort_order = "daddr,iaddr,pid,tid,major,minor,inode,inode_gen";
So if the 'daddr' key already checks major, minor, inode and inode_gen
by itself why do we need to add those sort keys again?
> + }
> +
> if (setup_sorting() < 0) {
> parse_options_usage(report_usage, options, "s", 1);
> goto error;
[SNIP]
> +static int64_t
> +sort__physid_daddr_cmp(struct hist_entry *left, struct hist_entry *right)
> +{
> + u64 l, r;
> + struct map *l_map = left->mem_info->daddr.map;
> + struct map *r_map = right->mem_info->daddr.map;
> +
> + /* store all NULL mem maps at the bottom */
> + /* shouldn't even need this check, should have stubs */
> + if (!left->mem_info->daddr.map || !right->mem_info->daddr.map)
> + return 1;
You might want to use 'return cmp_null(l_map, r_map);' here.
> +
> + /* group event types together */
> + if (left->cpumode > right->cpumode) return -1;
> + if (left->cpumode < right->cpumode) return 1;
> +
> + /*
> + * Addresses with no major/minor numbers are assumed to be
> + * anonymous in userspace. Sort those on pid then address.
> + *
> + * The kernel and non-zero major/minor mapped areas are
> + * assumed to be unity mapped. Sort those on address then pid.
> + */
> +
> + if (l_map->maj || l_map->min || l_map->ino || l_map->ino_generation) {
> + /* mmapped areas */
> +
> + if (l_map->maj > r_map->maj) return -1;
> + if (l_map->maj < r_map->maj) return 1;
> +
> + if (l_map->min > r_map->min) return -1;
> + if (l_map->min < r_map->min) return 1;
> +
> + if (l_map->ino > r_map->ino) return -1;
> + if (l_map->ino < r_map->ino) return 1;
> +
> + if (l_map->ino_generation > r_map->ino_generation) return -1;
> + if (l_map->ino_generation < r_map->ino_generation) return 1;
> +
> + } else if (left->cpumode != PERF_RECORD_MISC_KERNEL) {
> + /* userspace anonymous */
> + if (left->thread->pid_ > right->thread->pid_) return -1;
> + if (left->thread->pid_ < right->thread->pid_) return 1;
> + }
> +
> + /* hack to mark similar regions, 'right' is new entry */
> + right->color = TRUE;
I don't understand the logic behind the 'color'. It seems just mark
every samples except first one on a same file (or same pid for anon map)
indicating that those accesses are for distinct maps, right?
I don't know how it could help to distinguish whether an access is for a
same map or different map. For the userspace anon map case, why doesn't
it check the start addresses of l_map and r_map?
I'm feeling ignorant.. :-(
> +
> + /* al_addr does all the right addr - start + offset calculations */
> + l = left->mem_info->daddr.al_addr;
> + r = right->mem_info->daddr.al_addr;
> +
> + if (l > r) return -1;
> + if (l < r) return 1;
> +
> + /* sanity check the maps; only mmaped areas should have different maps */
> + if ((left->mem_info->daddr.map != right->mem_info->daddr.map) &&
> + !right->mem_info->daddr.map->maj && !right->mem_info->daddr.map->min)
> + pr_debug("physid_cmp: Similar entries have different maps\n");
> +
> + return 0;
> +}
[SNIP]
> @@ -1104,6 +1382,22 @@ int sort_dimension__add(const char *tok)
> return 0;
> }
>
> + for (i = 0; i < ARRAY_SIZE(physid_sort_dimensions); i++) {
> + struct sort_dimension *sd = &physid_sort_dimensions[i];
> +
> + if (strncasecmp(tok, sd->name, strlen(tok)))
> + continue;
> +
> + if (sort__mode != SORT_MODE__PHYSID)
> + return -EINVAL;
> +
> + if (sd->entry == &sort_physid_daddr)
> + sort__has_sym = 1;
I think it's not needed. The sort__has_sym is for doing annotate during
report/top session and it only works for symbol (i.e. function) basis.
Thanks,
Namhyung
> +
> + __sort_dimension__add(sd, i + __SORT_PHYSID_MODE);
> + return 0;
> + }
> +
> return -ESRCH;
> }
On Wed, Apr 09, 2014 at 02:21:49PM +0900, Namhyung Kim wrote:
> > create a new 'physid mode' to group all the sorting rules together
> > (mimics the mem-mode)
>
> What is 'physid' then? I guess you meant physical id but it seems
> unique id or unique map id looks like a better fit IMHO.
I suspect this is legacy naming; they used to do this using physical
addresses.