Skip to contentSkip to conversation

setleds: add option to reset state to the BIOS default

PATCH In review Petr Tesarik 2 participants 4 messages Jul 27, 2012 – Aug 2, 2012

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 setleds: add option to reset state to the BIOS default

Hello folks,

please consider the following patch for kbd.

As you probably know, the Linux kernel resets the LED states to all off on boot. However, many users would like to keep the BIOS NumLock setting, and others get confused after they press one of the lock keys in the boot loader and it gets reset again when Linux initializes the terminal.

Because of that, many distros have used various ways to "improve" user
experience: setting NumLock on by default, making OS-specific configuration options, or reading the BIOS area. The last option seems best to me, because it usually also preserves whatever you did in the boot loader. Unfortunately, there's no simple utility to read the state, so distros sometimes do insane things to get it (e.g. run hwinfo and grep only for the LED states and then run setleds with the appropriate arguments). Obviously, this approach isn't exactly fast...

Since setleds must be always used in the end, the most efficient solution is to add an option that resets the LED states to what they were before the Linux kernel booted.

---
 src/setleds.c |   61 ++++++++++++++++++++++++++++++++++++++++++++++++++++++--
 1 files changed, 58 insertions(+), 3 deletions(-)
src/setleds.c +58 -3
index 3577aee..3defdea 100644
@@ -11,18 +11,29 @@
 #include <fcntl.h>
 #include <linux/kd.h>
 #include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <unistd.h>
 #include "nls.h"
 #include "version.h"
 
+#if defined(__i386__) || defined(__x86_64__)
+# define HAVE_BIOS	1
+# define BIOS_KBD_ADDR	0x497
+# define BIOS_LED_SCR	0x01
+# define BIOS_LED_NUM	0x02
+# define BIOS_LED_CAP	0x04
+#endif
+
 static void attr_noreturn
 usage(void)
 {
     fprintf(stderr, _(
 "Usage:\n"
-"	setleds [-v] [-L] [-D] [-F] [[+|-][ num | caps | scroll %s]]\n"
+"	setleds [-v] [-L] [-D] [-F] [[+|-][ num | caps | scroll %s]%s]\n"
 "Thus,\n"
 "	setleds +caps -num\n"
 "will set CapsLock, clear NumLock and leave ScrollLock unchanged.\n"
+"%s"
 "The settings before and after the change (if any) are reported\n"
 "when the -v option is given or when no change is requested.\n"
 "Normally, setleds influences the vt flag settings\n"
@@ -32,9 +43,15 @@ usage(void)
 "that a subsequent reset will not change the flags.\n"
 ),
 #ifdef __sparc__
-    "| compose "
+    "| compose ",
+#else
+    "",
+#endif
+#ifdef HAVE_BIOS
+    " | bios ",
+"Specify \"bios\" to reset state to the BIOS default.\n"
 #else
-    ""
+    "", ""
 #endif
     );
     exit(1);
@@ -152,6 +169,37 @@ sunsetleds(arg_state char *cur_leds) {
 #endif
 }
 
+#ifdef HAVE_BIOS
+static void
+biosgetleds(char *cur_leds) {
+    int memfd;
+    long pagesz, mapoff;
+    char *map, bios_state;
+
+    memfd = open("/dev/mem", O_RDONLY);
+    if (memfd < 0) {
+	perror("/dev/mem");
+	fprintf(stderr, _("Error opening /dev/mem.\n"));
+	exit(1);
+    }
+    pagesz = sysconf(_SC_PAGESIZE);
+    mapoff = BIOS_KBD_ADDR & ~(pagesz-1);
+    map = mmap(NULL, pagesz, PROT_READ, MAP_SHARED, memfd, mapoff);
+    if (map == MAP_FAILED) {
+	perror("/dev/mem");
+	fprintf(stderr, _("Error mapping /dev/mem.\n"));
+	exit(1);
+    }
+    bios_state = map[BIOS_KBD_ADDR - mapoff];
+    *cur_leds =
+	(bios_state & BIOS_LED_SCR ? LED_SCR : 0) |
+	(bios_state & BIOS_LED_NUM ? LED_NUM : 0) |
+	(bios_state & BIOS_LED_CAP ? LED_CAP : 0);
+    munmap(map, pagesz);
+    close(memfd);
+}
+#endif	/* HAVE_BIOS */
+
 int
 main(int argc, char **argv) {
     int optL = 0, optD = 0, optF = 0, verbose = 0;
@@ -241,6 +289,13 @@ main(int argc, char **argv) {
 
     while(--argc) {
 	ap = *++argv;
+#ifdef HAVE_BIOS
+	if (!strcmp(ap, "bios")) {
+	  biosgetleds(&nval);
+	  ndef = LED_NUM | LED_CAP | LED_SCR;
+	  goto nxtarg;
+	}
+#endif
 	sign = 1;		/* by default: set */
 	if(*ap == '+')
 	  ap++;
#1 lore raw

src/setleds.c line 20

Quoted 2 lines · Petr Tesarik
src/setleds.c +2 -0
+#if defined(__i386__) || defined(__x86_64__)
+# define HAVE_BIOS	1

I think it's strange condition.
How your code will work on the EFI ?

Signature · 1 line
Rgrds, legion
#2 lore raw

Hi Alexey,

glad you didn't smash the idea from an ideological standpoint.

Dne Ne 29. července 2012 23:10:19 Alexey Gladkov napsal(a):

Quoted 5 lines

src/setleds.c line 20

Quoted 2 lines · Petr Tesarik
src/setleds.c +2 -0
+#if defined(__i386__) || defined(__x86_64__)
+# define HAVE_BIOS	1

I think it's strange condition.

It is, indeed. For one thing, some Itanium machines also have the standard PC BIOS Data Area (BDA), and they don't get this new feature. Unfortunately, some (early?) SGI Altix machines don't have physical memory at 0x400, and they don't handle accesses to non-existent RAM very well (read: freeze or reboot). OTOH detecting the machine model and maintaining a blacklist or whitelist of well-known IA-64 systems in setleds sounds like overkill, especially for a dying platform. So, I didn't include them.

Quoted 1 line

How your code will work on the EFI ?

Well, UEFI will work just fine, same with Intel's EFI, because they include a Compatibility Support Module (CSM), which (among other things) initializes legacy BDA. Yes, some machines don't do that properly (IIRC Phoenix issued an errata a few years ago exactly to fix LED states in the BDA).

Even Apple includes a CSM for their EFI implementation...

I would be a bit more concerned about embedded hardware like the one I used to have in one of my ancient D-Link routers in the early 2000's, which emulated an x86 on top of ICplus Corp. IC175. However, platforms like this usually run in a well-controlled environment, so the firmware writer will probably not even attemp to use this new setleds feature.

In any case, even if the BDA is not initialized properly, the worst thing that happens is that the LEDs are set incorrectly when you run "setleds bios".

Petr Tesarik
SUSE Linux

#3 lore raw
In reply to Petr Tesarik
Quoted 18 lines · Petr Tesarik

Hello folks,

please consider the following patch for kbd.

As you probably know, the Linux kernel resets the LED states to all off on boot. However, many users would like to keep the BIOS NumLock setting, and others get confused after they press one of the lock keys in the boot loader and it gets reset again when Linux initializes the terminal.

Because of that, many distros have used various ways to "improve" user
experience: setting NumLock on by default, making OS-specific configuration options, or reading the BIOS area. The last option seems best to me, because it usually also preserves whatever you did in the boot loader. Unfortunately, there's no simple utility to read the state, so distros sometimes do insane things to get it (e.g. run hwinfo and grep only for the LED states and then run setleds with the appropriate arguments). Obviously, this approach isn't exactly fast...

Quoted 5 lines

Since setleds must be always used in the end, the most efficient solution is to add an option that resets the LED states to what they were before the Linux kernel booted.

It looks like what you propose has already been discussed several years ago:

http://article.gmane.org/gmane.linux.kernel/493362

I also found that some code is already in the kernel:

http://git.kernel.org/?p=linux/kernel/git/torvalds/linux.git;a=commitdiff;h=b2d0b7a061bfddd27155c7dcd53f365d9dc0c7c3

As far as I could learn, it is possible to query the LEDs on the USB keyboard, but for PS/2 this is not possible.

Quoted 110 lines
---
 src/setleds.c |   61 ++++++++++++++++++++++++++++++++++++++++++++++++++++++--
 1 files changed, 58 insertions(+), 3 deletions(-)
src/setleds.c +58 -3
index 3577aee..3defdea 100644
@@ -11,18 +11,29 @@
 #include <fcntl.h>
 #include <linux/kd.h>
 #include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <unistd.h>
 #include "nls.h"
 #include "version.h"
 
+#if defined(__i386__) || defined(__x86_64__)
+# define HAVE_BIOS	1
+# define BIOS_KBD_ADDR	0x497
+# define BIOS_LED_SCR	0x01
+# define BIOS_LED_NUM	0x02
+# define BIOS_LED_CAP	0x04
+#endif
+
 static void attr_noreturn
 usage(void)
 {
     fprintf(stderr, _(
 "Usage:\n"
-"	setleds [-v] [-L] [-D] [-F] [[+|-][ num | caps | scroll %s]]\n"
+"	setleds [-v] [-L] [-D] [-F] [[+|-][ num | caps | scroll %s]%s]\n"
 "Thus,\n"
 "	setleds +caps -num\n"
 "will set CapsLock, clear NumLock and leave ScrollLock unchanged.\n"
+"%s"
 "The settings before and after the change (if any) are reported\n"
 "when the -v option is given or when no change is requested.\n"
 "Normally, setleds influences the vt flag settings\n"
@@ -32,9 +43,15 @@ usage(void)
 "that a subsequent reset will not change the flags.\n"
 ),
 #ifdef __sparc__
-    "| compose "
+    "| compose ",
+#else
+    "",
+#endif
+#ifdef HAVE_BIOS
+    " | bios ",
+"Specify \"bios\" to reset state to the BIOS default.\n"
 #else
-    ""
+    "", ""
 #endif
     );
     exit(1);
@@ -152,6 +169,37 @@ sunsetleds(arg_state char *cur_leds) {
 #endif
 }
 
+#ifdef HAVE_BIOS
+static void
+biosgetleds(char *cur_leds) {
+    int memfd;
+    long pagesz, mapoff;
+    char *map, bios_state;
+
+    memfd = open("/dev/mem", O_RDONLY);
+    if (memfd < 0) {
+	perror("/dev/mem");
+	fprintf(stderr, _("Error opening /dev/mem.\n"));
+	exit(1);
+    }
+    pagesz = sysconf(_SC_PAGESIZE);
+    mapoff = BIOS_KBD_ADDR & ~(pagesz-1);
+    map = mmap(NULL, pagesz, PROT_READ, MAP_SHARED, memfd, mapoff);
+    if (map == MAP_FAILED) {
+	perror("/dev/mem");
+	fprintf(stderr, _("Error mapping /dev/mem.\n"));
+	exit(1);
+    }
+    bios_state = map[BIOS_KBD_ADDR - mapoff];
+    *cur_leds =
+	(bios_state & BIOS_LED_SCR ? LED_SCR : 0) |
+	(bios_state & BIOS_LED_NUM ? LED_NUM : 0) |
+	(bios_state & BIOS_LED_CAP ? LED_CAP : 0);
+    munmap(map, pagesz);
+    close(memfd);
+}
+#endif	/* HAVE_BIOS */
+
 int
 main(int argc, char **argv) {
     int optL = 0, optD = 0, optF = 0, verbose = 0;
@@ -241,6 +289,13 @@ main(int argc, char **argv) {
 
     while(--argc) {
 	ap = *++argv;
+#ifdef HAVE_BIOS
+	if (!strcmp(ap, "bios")) {
+	  biosgetleds(&nval);
+	  ndef = LED_NUM | LED_CAP | LED_SCR;
+	  goto nxtarg;
+	}
+#endif
 	sign = 1;		/* by default: set */
 	if(*ap == '+')
 	  ap++;
Signature · 1 line
Rgrds, legion