Skip to contentSkip to conversation

kinit problem with cciss root device

PATCH Bjorn Helgaas 2 participants 4 messages Jun 6, 2006 Kbuild

lore
Share & tools
lore
#0 lore raw

kinit converts "root=/dev/cciss/c0d0p1" to "cciss.c0d0", which doesn't exist under /sys/block because register_disk() converts "cciss/c0d0" to "cciss!c0d0" (note "!", not ".").

I don't know whether it's the *right* fix, but the patch below makes things work. It still doesn't make kinit exactly match the register_disk() behavior because register_disk() only changes the first "/" in the string.

This was a nuisance to debug because when it failed, the only output I got was this:

...
Adding console on ttyS0 at I/O port 0x3f8 (options '115200')
Time: tsc clocksource has been installed.
Freeing unused kernel memory: 260k freed

    argc == 4
  Kernel panic - not syncing: Attempted to kill init!
    argv[0]: "/ini 

So you might consider something like the "drain output" hunk below, which allowed all the useful messages to get out.

usr/kinit/kinit.c +5 -0
Index: work-mm8/usr/kinit/kinit.c
===================================================================
@@ -317,5 +317,10 @@
 	if (mnt_sysfs)
 		umount2("/sys", 0);
 
+	/*
+	 * Allow time for messages to drain before kernel panics
+	 * because init is exiting.
+	 * */
+	sleep(10);
 	exit(ret);
 }
usr/kinit/name_to_dev.c +1 -1
Index: work-mm8/usr/kinit/name_to_dev.c
===================================================================
@@ -151,7 +151,7 @@
 
 	for (p = s; *p; p++)
 		if (*p == '/')
-			*p = '.';
+			*p = '!';
 	res = try_name(s, 0);
 	if (res)
 		return res;
#1 lore raw
Quoted 4 lines · Bjorn Helgaas

So you might consider something like the "drain output" hunk below, which allowed all the useful messages to get out.

Hm. That's rather ugly. Anyone knows if TCFLSB works on /dev/console?

Quoted 6 lines

Index: work-mm8/usr/kinit/kinit.c
===================================================================

usr/kinit/kinit.c line 326

Quoted 10 lines
usr/kinit/kinit.c +5 -0
@@ -317,5 +317,10 @@
 	if (mnt_sysfs)
 		umount2("/sys", 0);
 
+	/*
+	 * Allow time for messages to drain before kernel panics
+	 * because init is exiting.
+	 * */
+	sleep(10);
 	exit(ret);
 }

-hpa

#2 lore raw
Quoted 8 lines · H. Peter Anvin
Quoted 4 lines · Bjorn Helgaas

So you might consider something like the "drain output" hunk below, which allowed all the useful messages to get out.

Hm. That's rather ugly. Anyone knows if TCFLSB works on /dev/console?

That should of course have been TCSBRK(1), as in tcdrain().

-hpa

#3 lore raw
In reply to Bjorn Helgaas
Quoted 9 lines · Bjorn Helgaas

kinit converts "root=/dev/cciss/c0d0p1" to "cciss.c0d0", which doesn't exist under /sys/block because register_disk() converts "cciss/c0d0" to "cciss!c0d0" (note "!", not ".").

I don't know whether it's the *right* fix, but the patch below makes things work. It still doesn't make kinit exactly match the register_disk() behavior because register_disk() only changes the first "/" in the string.

That's still the right thing; obviously register_disk() will need to change if we ever have deeper trees.

Could you re-send that hunk as a proper patch with Signed-off-by: et al?

Thanks,

-hpa