linux-kernel - [PATCH] firmware: dmi: Optimize dmi

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: <20170727083128.7ab14de9@endymion>
Date:   Thu, 27 Jul 2017 08:31:28 +0200
From:   Jean Delvare <jdelvare@...e.de>
To:     LKML <linux-kernel@...r.kernel.org>
Cc:     Jani Nikula <jani.nikula@...ux.intel.com>,
        Daniel Vetter <daniel.vetter@...el.com>,
        "Rafael J. Wysocki" <rafael.j.wysocki@...el.com>,
        Jeff Garzik <jgarzik@...hat.com>,
        Hans de Goede <hdegoede@...hat.com>
Subject: [PATCH] firmware: dmi: Optimize dmi_matches

Function dmi_matches can me made a bit faster:

* The documented purpose of dmi_initialized is to catch too early
  calls to dmi_check_system(). I'm not fully convinced it justifies
  slowing down the initialization of all systems out there, but at
  least the check should not have been moved from dmi_check_system()
  to dmi_matches(). dmi_matches() is being called for every entry of
  the table passed to dmi_check_system(), causing the same redundant
  check to be performed again and again. So move it back to
  dmi_check_system(), reverting this specific portion of commit
  d7b1956fed33 ("DMI: Introduce dmi_first_match to make the interface
  more flexible").

* Don't check for the exact_match flag again when we already know its
  value.

Signed-off-by: Jean Delvare <jdelvare@...e.de>
Cc: Jani Nikula <jani.nikula@...ux.intel.com>
Cc: Daniel Vetter <daniel.vetter@...el.com>
Cc: Rafael J. Wysocki <rafael.j.wysocki@...el.com>
Cc: Jeff Garzik <jgarzik@...hat.com>
---
Regarding dmi_initialized, I don't think it makes sense to check for
a possible bad initialization order at run time on every system when
it is all decided at build time. If a developer introduces a new call
to dmi_check_system() and it is too early in the initialization
sequence, I believe he/she would notice upon first testing, and a
comment to his/her intention in the source code would serve the same
purpose without the worldwide performance penalty. Would anyone
object to such a change?

 drivers/firmware/dmi_scan.c |   19 +++++++++++--------
 1 file changed, 11 insertions(+), 8 deletions(-)

--- linux-4.12.orig/drivers/firmware/dmi_scan.c	2017-07-27 08:15:38.283519194 +0200
+++ linux-4.12/drivers/firmware/dmi_scan.c	2017-07-27 08:26:23.013053058 +0200
@@ -784,19 +784,20 @@ static bool dmi_matches(const struct dmi
 {
 	int i;
 
-	WARN(!dmi_initialized, KERN_ERR "dmi check: not initialized yet.\n");
-
 	for (i = 0; i < ARRAY_SIZE(dmi->matches); i++) {
 		int s = dmi->matches[i].slot;
 		if (s == DMI_NONE)
 			break;
 		if (dmi_ident[s]) {
-			if (!dmi->matches[i].exact_match &&
-			    strstr(dmi_ident[s], dmi->matches[i].substr))
-				continue;
-			else if (dmi->matches[i].exact_match &&
-				 !strcmp(dmi_ident[s], dmi->matches[i].substr))
-				continue;
+			if (dmi->matches[i].exact_match) {
+				if (!strcmp(dmi_ident[s],
+					    dmi->matches[i].substr))
+					continue;
+			} else {
+				if (strstr(dmi_ident[s],
+					   dmi->matches[i].substr))
+					continue;
+			}
 		}
 
 		/* No match */
@@ -832,6 +833,8 @@ int dmi_check_system(const struct dmi_sy
 	int count = 0;
 	const struct dmi_system_id *d;
 
+	WARN(!dmi_initialized, KERN_ERR "dmi check: not initialized yet.\n");
+
 	for (d = list; !dmi_is_end_of_table(d); d++)
 		if (dmi_matches(d)) {
 			count++;

-- 
Jean Delvare
SUSE L3 Support