linux-kernel - [PATCH] perf: save RAM when filtering events

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <20190922164730.GA16336@home0.donatello.akdev.xyz>
Date:   Sun, 22 Sep 2019 12:47:30 -0400
From:   Alex <alex@...ev.xyz>
To:     linux-kernel@...r.kernel.org
Cc:     namhyung@...nel.org, jolsa@...hat.com,
        alexander.shishkin@...ux.intel.com, mark.rutland@....com,
        acme@...nel.org, mingo@...hat.com, peterz@...radead.org
Subject: [PATCH] perf: save RAM when filtering events

The code mentioned wanting to use an implementation of log10() in the dvb
drivers. I am not aware why that specific implementation is mentioned.
I used the implementation found in `math.h`.

I tested this patch passing 200 syscalls to filter on a `perf trace -e ${syscalls} ls`
call. You can see the commands used and a snipped of the valgrind results here:
https://termbin.com/k4of

Before: 61,910,110 bytes allocated
After:  61,907,045 bytes allocated

perf used to allocate space in excess to build the filtering expressions.
now perf only allocates the space necessary and not more.

Signed-off-by: Alex Diaz <alex@...ev.xyz>
---
 tools/perf/util/string.c  | 25 ++++++++++++++++++++-----
 tools/perf/util/string2.h |  2 ++
 2 files changed, 22 insertions(+), 5 deletions(-)

diff --git a/tools/perf/util/string.c b/tools/perf/util/string.c
index 52603876c548..4c3913a9e7e7 100644
--- a/tools/perf/util/string.c
+++ b/tools/perf/util/string.c
@@ -3,6 +3,7 @@
 #include <linux/kernel.h>
 #include <linux/string.h>
 #include <stdlib.h>
+#include <math.h>

 #include <linux/ctype.h>

@@ -209,15 +210,29 @@ int strtailcmp(const char *s1, const char *s2)
         return 0;
 }

-char *asprintf_expr_inout_ints(const char *var, bool in, size_t nints, int *ints)
+size_t calc_expr_buffer_size(const char *var, size_t nints, int *ints)
 {
         /*
-        * FIXME: replace this with an expression using log10() when we
-        * find a suitable implementation, maybe the one in the dvb drivers...
+        * Calculate the buffer for the expression:
          *
-        * "%s == %d || " = log10(MAXINT) * 2 + 8 chars for the operators
+        * "%s == %d || "
+        * length: strlen(var) + strlen(" == ") + log10(n) + strlen(" || ") + 1
          */
-       size_t size = nints * 28 + 1; /* \0 */
+       size_t size = 0;
+       size_t num_len = 0;
+       const size_t var_len = strlen(var);
+
+       for (size_t i = 0; i < nints; ++i) {
+               num_len = (ints[i] == 0) ? 1 : log10(ints[i]);
+               size += var_len + num_len + 9;
+       }
+
+       return size;
+}
+
+char *asprintf_expr_inout_ints(const char *var, bool in, size_t nints, int *ints)
+{
+       size_t size = calc_expr_buffer_size(var, nints, ints);
         size_t i, printed = 0;
         char *expr = malloc(size);

diff --git a/tools/perf/util/string2.h b/tools/perf/util/string2.h
index 708805f5573e..28fbc782ad54 100644
--- a/tools/perf/util/string2.h
+++ b/tools/perf/util/string2.h
@@ -20,6 +20,8 @@ static inline bool strisglob(const char *str)
 }
 int strtailcmp(const char *s1, const char *s2);

+size_t calc_expr_buffer_size(const char *var, size_t nints, int *ints);
+
 char *asprintf_expr_inout_ints(const char *var, bool in, size_t nints, int *ints);

 static inline char *asprintf_expr_in_ints(const char *var, size_t nints, int *ints)
--
2.21.0