tools: Add hash table item batch operations support

Commit c3b11e5 (add bpf_map_lookup_and_delete_batch in bcc (#3234))
supports hash table use batch operation to solve race condition issue
and reduce items syscalls to improve performance.

Update tools with hash operations.

Improvement can be test with the belove commands:
  # filetop
  # syscount --syscall bpf -i 1
  # mkdir test; for i in `seq 1 100000`;do touch test/$i;done

Before patch filetop, syscall counts are more than 2k.
  [16:58:42]
  SYSCALL                   COUNT
  bpf                          55

  [16:58:43]
  SYSCALL                   COUNT
  bpf                        2735

  [16:58:44]
  SYSCALL                   COUNT
  bpf                        2263

After patch batch operations, result shows except 1 syscall for syscount
itself, filetop syscall count is reduced to 1.
  Tracing syscall 'bpf'... Ctrl+C to quit.
  [17:00:36]
  SYSCALL                   COUNT
  bpf                           1

  [17:00:37]
  SYSCALL                   COUNT
  bpf                           2

  [17:00:38]
  SYSCALL                   COUNT
  bpf                           2

Change-Id: Icc7f6d39f339bfc492ddcfc213ac91cdbdbe3583
Signed-off-by: Mickey Zhu <[email protected]>

Author: michael-chuh

Diff for: tools/argdist.py

@@ -470,6 +470,12 @@ def attach(self, bpf):
                         self._attach_k()
                 if self.entry_probe_required:
                         self._attach_entry_probe()
+                # Check whether hash table batch ops is supported
+                if self.type == "freq" and self.bpf.kernel_struct_has_field(
+                        b'bpf_map_ops', b'map_lookup_and_delete_batch') == 1:
+                    self.htab_batch_ops = True
+                else:
+                    self.htab_batch_ops = False
 
         def _v2s(self, v):
                 # Most fields can be converted with plain str(), but strings
@@ -510,7 +516,9 @@ def display(self, top):
                 if self.type == "freq":
                         print(self.label or self.raw_spec)
                         print("\t%-10s %s" % ("COUNT", "EVENT"))
-                        sdata = sorted(data.items(), key=lambda p: p[1].value)
+                        sdata = sorted(data.items_lookup_batch()
+                                if self.htab_batch_ops else data.items(),
+                                key=lambda p: p[1].value)
                         if top is not None:
                                 sdata = sdata[-top:]
                         for key, value in sdata:
@@ -531,6 +539,9 @@ def display(self, top):
                                 if not self.is_default_expr else "retval")
                         data.print_log2_hist(val_type=label)
                 if not self.cumulative:
+                    if self.htab_batch_ops:
+                        data.items_delete_batch()
+                    else:
                         data.clear()
 
         def __str__(self):

Diff for: tools/biopattern.py

@@ -105,6 +105,10 @@ def mkdev(major, minor):
 
 b = BPF(text=bpf_text)
 
+# check whether hash table batch ops is supported
+htab_batch_ops = True if BPF.kernel_struct_has_field(b'bpf_map_ops',
+        b'map_lookup_and_delete_batch') == 1 else False
+
 exiting = 0 if args.interval else 1
 counters = b.get_table("counters")
 
@@ -117,7 +121,8 @@ def mkdev(major, minor):
     except KeyboardInterrupt:
         exiting = 1
 
-    for k, v in counters.items():
+    for k, v in (counters.items_lookup_and_delete_batch()
+                if htab_batch_ops else counters.items()):
         total = v.random + v.sequential
         if total == 0:
             continue
@@ -133,7 +138,8 @@ def mkdev(major, minor):
             total,
             v.bytes / 1024))
 
-    counters.clear()
+    if not htab_batch_ops:
+        counters.clear()
 
     countdown -= 1
     if exiting or countdown == 0:

Diff for: tools/biotop.py

@@ -287,6 +287,10 @@
 else:
     b.attach_tracepoint(tp="block:block_io_done", fn_name="trace_req_completion_tp")
 
+# check whether hash table batch ops is supported
+htab_batch_ops = True if BPF.kernel_struct_has_field(b'bpf_map_ops',
+        b'map_lookup_and_delete_batch') == 1 else False
+
 print('Tracing... Output every %d secs. Hit Ctrl-C to end' % interval)
 
 # cache disk major,minor -> diskname
@@ -317,7 +321,8 @@
     # by-PID output
     counts = b.get_table("counts")
     line = 0
-    for k, v in reversed(sorted(counts.items(),
+    for k, v in reversed(sorted(counts.items_lookup_and_delete_batch()
+                                if htab_batch_ops else counts.items(),
                                 key=lambda counts: counts[1].bytes)):
 
         # lookup disk
@@ -336,7 +341,9 @@
         line += 1
         if line >= maxrows:
             break
-    counts.clear()
+
+    if not htab_batch_ops:
+        counts.clear()
 
     countdown -= 1
     if exiting or countdown == 0:

Diff for: tools/cachestat.py

@@ -140,6 +140,10 @@ def get_meminfo():
                     ("folio_account_dirtied", "account_page_dirtied"))
 b.attach_kprobe(event="mark_buffer_dirty", fn_name="do_count_mbd")
 
+# check whether hash table batch ops is supported
+htab_batch_ops = True if BPF.kernel_struct_has_field(b'bpf_map_ops',
+        b'map_lookup_and_delete_batch') == 1 else False
+
 # header
 if tstamp:
     print("%-8s " % "TIME", end="")
@@ -162,7 +166,9 @@ def get_meminfo():
         signal.signal(signal.SIGINT, signal_ignore)
 
     counts = b["counts"]
-    for k, v in sorted(counts.items(), key=lambda counts: counts[1].value):
+    for k, v in sorted(counts.items_lookup_and_delete_batch()
+                                if htab_batch_ops else counts.items(),
+                                key=lambda counts: counts[1].value):
         # partial string matches in case of .isra (necessary?)
         if k.nf == 0: # NF_APCL
             apcl = max(0, v.value)
@@ -197,7 +203,8 @@ def get_meminfo():
         print("%d %d %d %d %d %d %d\n" %
         (mpa, mbd, apcl, apd, total, misses, hits))
 
-    counts.clear()
+    if not htab_batch_ops:
+        counts.clear()
 
     # Get memory info
     mem = get_meminfo()

Diff for: tools/cachetop.py

@@ -64,7 +64,8 @@ def get_meminfo():
 def get_processes_stats(
         bpf,
         sort_field=DEFAULT_SORT_FIELD,
-        sort_reverse=False):
+        sort_reverse=False,
+        htab_batch_ops=False):
     '''
     Return a tuple containing:
     buffer
@@ -73,7 +74,8 @@ def get_processes_stats(
     '''
     counts = bpf.get_table("counts")
     stats = defaultdict(lambda: defaultdict(int))
-    for k, v in counts.items():
+    for k, v in (counts.items_lookup_batch()
+                if htab_batch_ops else counts.items()):
         stats["%d-%d-%s" % (k.pid, k.uid, k.comm.decode('utf-8', 'replace'))][k.nf] = v.value
     stats_list = []
 
@@ -129,7 +131,11 @@ def get_processes_stats(
     stats_list = sorted(
         stats_list, key=lambda stat: stat[sort_field], reverse=sort_reverse
     )
-    counts.clear()
+    if htab_batch_ops:
+        counts.items_delete_batch()
+    else:
+        counts.clear()
+
     return stats_list
 
 
@@ -220,6 +226,10 @@ def handle_loop(stdscr, args):
 
     exiting = 0
 
+    # check whether hash table batch ops is supported
+    htab_batch_ops = True if BPF.kernel_struct_has_field(b'bpf_map_ops',
+            b'map_lookup_and_delete_batch') == 1 else False
+
     while 1:
         s = stdscr.getch()
         if s == ord('q'):
@@ -245,7 +255,8 @@ def handle_loop(stdscr, args):
         process_stats = get_processes_stats(
             b,
             sort_field=sort_field,
-            sort_reverse=sort_reverse)
+            sort_reverse=sort_reverse,
+            htab_batch_ops=htab_batch_ops)
         stdscr.clear()
         stdscr.addstr(
             0, 0,

Diff for: tools/dirtop.py

@@ -197,6 +197,10 @@ def get_searched_ids(root_directories):
 b.attach_kprobe(event="vfs_read", fn_name="trace_read_entry")
 b.attach_kprobe(event="vfs_write", fn_name="trace_write_entry")
 
+# check whether hash table batch ops is supported
+htab_batch_ops = True if BPF.kernel_struct_has_field(b'bpf_map_ops',
+        b'map_lookup_and_delete_batch') == 1 else False
+
 DNAME_INLINE_LEN = 32  # linux/dcache.h
 
 print('Tracing... Output every %d secs. Hit Ctrl-C to end' % interval)
@@ -235,7 +239,8 @@ def sort_fn(counts):
     writes = {}
     reads_Kb = {}
     writes_Kb = {}
-    for k, v in reversed(sorted(counts.items(),
+    for k, v in reversed(sorted(counts.items_lookup_and_delete_batch()
+                                if htab_batch_ops else counts.items(),
                                 key=sort_fn)):
         # If it's the first time we see this inode
         if k.inode_id not in reads:
@@ -259,7 +264,8 @@ def sort_fn(counts):
         if line >= maxrows:
             break
 
-    counts.clear()
+    if not htab_batch_ops:
+        counts.clear()
 
     countdown -= 1
     if exiting or countdown == 0:

Diff for: tools/filetop.py

@@ -167,6 +167,10 @@
 b.attach_kprobe(event="vfs_read", fn_name="trace_read_entry")
 b.attach_kprobe(event="vfs_write", fn_name="trace_write_entry")
 
+# check whether hash table batch ops is supported
+htab_batch_ops = True if BPF.kernel_struct_has_field(b'bpf_map_ops',
+        b'map_lookup_and_delete_batch') == 1 else False
+
 DNAME_INLINE_LEN = 32  # linux/dcache.h
 
 print('Tracing... Output every %d secs. Hit Ctrl-C to end' % interval)
@@ -198,7 +202,8 @@ def sort_fn(counts):
     # by-TID output
     counts = b.get_table("counts")
     line = 0
-    for k, v in reversed(sorted(counts.items(),
+    for k, v in reversed(sorted(counts.items_lookup_and_delete_batch()
+                                if htab_batch_ops else counts.items(),
                                 key=sort_fn)):
         name = k.name.decode('utf-8', 'replace')
         if k.name_len > DNAME_INLINE_LEN:
@@ -213,7 +218,9 @@ def sort_fn(counts):
         line += 1
         if line >= maxrows:
             break
-    counts.clear()
+
+    if not htab_batch_ops:
+        counts.clear()
 
     countdown -= 1
     if exiting or countdown == 0:

Diff for: tools/klockstat.py

@@ -413,7 +413,9 @@ def display(sort, maxs, totals, counts):
     global missing_stacks
     global has_enomem
 
-    for k, v in sorted(sort.items(), key=lambda sort: sort[1].value, reverse=True)[:args.locks]:
+    for k, v in sorted(sort.items_lookup_batch()
+                       if htab_batch_ops else sort.items(),
+                       key=lambda sort: sort[1].value, reverse=True)[:args.locks]:
         missing_stacks += int(stack_id_err(k.value))
         has_enomem      = has_enomem or (k.value == -errno.ENOMEM)
 
@@ -458,6 +460,10 @@ def display(sort, maxs, totals, counts):
         b.attach_kretprobe(event="mutex_lock", fn_name="mutex_lock_return")
         b.attach_kprobe(event="mutex_lock", fn_name="mutex_lock_enter")
 
+# check whether hash table batch ops is supported
+htab_batch_ops = True if BPF.kernel_struct_has_field(b'bpf_map_ops',
+        b'map_lookup_and_delete_batch') == 1 else False
+
 enabled = b.get_table("enabled");
 
 stack_traces = b.get_table("stack_traces")
@@ -508,12 +514,20 @@ def display(sort, maxs, totals, counts):
         break;
 
     stack_traces.clear()
-    aq_counts.clear()
-    aq_maxs.clear()
-    aq_totals.clear()
-    hl_counts.clear()
-    hl_maxs.clear()
-    hl_totals.clear()
+    if htab_batch_ops:
+        aq_counts.items_delete_batch()
+        aq_maxs.items_delete_batch()
+        aq_totals.items_delete_batch()
+        hl_counts.items_delete_batch()
+        hl_maxs.items_delete_batch()
+        hl_totals.items_delete_batch()
+    else:
+        aq_counts.clear()
+        aq_maxs.clear()
+        aq_totals.clear()
+        hl_counts.clear()
+        hl_maxs.clear()
+        hl_totals.clear()
 
 if missing_stacks > 0:
     enomem_str = " Consider increasing --stack-storage-size."

Diff for: tools/netqtop.py

@@ -63,7 +63,8 @@ def print_table(table, qnum):
     tGroup = [0,0,0,0,0]
     tpkt = 0
     tlen = 0
-    for k, v in table.items():
+    for k, v in (table.items_lookup_batch()
+                if htab_batch_ops else table.items()):
         qids += [k.value]
         tlen += v.total_pkt_len
         tpkt += v.num_pkt
@@ -142,14 +143,20 @@ def print_result(b):
     print("TX")
     table = b['tx_q']
     print_table(table, tx_num)
-    b['tx_q'].clear()
+    if htab_batch_ops:
+        b['tx_q'].items_delete_batch()
+    else:
+        b['tx_q'].clear()
 
     # --------- print rx queues ---------------
     print("")
     print("RX")
     table = b['rx_q']
     print_table(table, rx_num)
-    b['rx_q'].clear()
+    if htab_batch_ops:
+        b['rx_q'].items_delete_batch()
+    else:
+        b['rx_q'].clear()
     if args.throughput:
         print("-"*95)
     else:
@@ -205,6 +212,10 @@ def print_result(b):
 
 ################## start tracing ##################
 b = BPF(src_file = EBPF_FILE)
+# --- check whether hash table batch ops is supported ---
+htab_batch_ops = True if BPF.kernel_struct_has_field(b'bpf_map_ops',
+        b'map_lookup_and_delete_batch') == 1 else False
+
 # --------- set hash array --------
 devname_map = b['name_map']
 _name = Devname()

Diff for: tools/slabratetop.py

@@ -199,6 +199,9 @@
 
 # initialize BPF
 b = BPF(text=bpf_text)
+# check whether hash table batch ops is supported
+htab_batch_ops = True if BPF.kernel_struct_has_field(b'bpf_map_ops',
+        b'map_lookup_and_delete_batch') == 1 else False
 
 print('Tracing... Output every %d secs. Hit Ctrl-C to end' % interval)
 
@@ -222,14 +225,16 @@
     # by-TID output
     counts = b.get_table("counts")
     line = 0
-    for k, v in reversed(sorted(counts.items(),
+    for k, v in reversed(sorted(counts.items_lookup_and_delete_batch()
+                                if htab_batch_ops else counts.items(),
                                 key=lambda counts: counts[1].size)):
         printb(b"%-32s %6d %10d" % (k.name, v.count, v.size))
 
         line += 1
         if line >= maxrows:
             break
-    counts.clear()
+    if not htab_batch_ops:
+        counts.clear()
 
     countdown -= 1
     if exiting or countdown == 0:

Diff for: tools/stackcount.py

@@ -319,6 +319,9 @@ def run(self):
             print("Tracing %d functions for \"%s\"... Hit Ctrl-C to end." %
                   (self.probe.matched, self.args.pattern))
         b = self.probe.bpf
+        # check whether hash table batch ops is supported
+        htab_batch_ops = True if BPF.kernel_struct_has_field(b'bpf_map_ops',
+                         b'map_lookup_and_delete_batch') == 1 else False
         exiting = 0 if self.args.interval else 1
         seconds = 0
         while True:
@@ -340,7 +343,8 @@ def run(self):
             counts = self.probe.bpf["counts"]
             stack_traces = self.probe.bpf["stack_traces"]
             self.comm_cache = {}
-            for k, v in sorted(counts.items(),
+            for k, v in sorted(counts.items_lookup_and_delete_batch()
+                               if htab_batch_ops else counts.items(),
                                key=lambda counts: counts[1].value):
                 user_stack = [] if k.user_stack_id < 0 else \
                     stack_traces.walk(k.user_stack_id)
@@ -368,7 +372,8 @@ def run(self):
                     if not self.args.pid and k.tgid != 0xffffffff:
                         self._print_comm(k.name, k.tgid)
                     print("    %d\n" % v.value)
-            counts.clear()
+            if not htab_batch_ops:
+                counts.clear()
 
             if exiting:
                 if not self.args.folded: