x86_64: fix earlyprintk=...,keep regression
PATCH 1/1 x86_64: fix earlyprintk=...,keep regression
Subject: [patch] x86_64: fix earlyprintk=...,keep regression
From: Ingo Molnar <[email protected]>
the following cleanup patch:
commit 2c8c0e6b8d7700a990da8d24eff767f9ca223b96
Author: Andi Kleen <[email protected]>
Date: Tue Sep 26 10:52:32 2006 +0200
[PATCH] Convert x86-64 to early param
Instead of hackish manual parsing
broke the earlyprintk=...,keep feature. This patch restores that functionality. Tested on x86_64. Must-have for v2.6.19, no risk.
--- arch/x86_64/kernel/early_printk.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86_64/kernel/early_printk.c +1 -1
@@ -224,7 +224,7 @@ static int __init setup_early_printk(chareturn 0 ; early_console_initialized =1 ; -if (!strcmp (buf,"keep" )) +if (strstr (buf,"keep" )) keep_early =1 ;if (!strncmp (buf,"serial" ,6 )) {
Quoted 7 lines · Ingo Molnar
the following cleanup patch:
commit
2c8c0e6b8d7700a990da8d24eff767f9ca223b96
Author: Andi Kleen <[email protected]>
Date: Tue Sep 26 10:52:32 2006 +0200
Quoted 6 lines
[PATCH] Convert x86-64 to early param
Instead of hackish manual parsing
broke the earlyprintk=...,keep feature. This patch restores that functionality. Tested on
x86_64. Must-have for v2.6.19, no risk.
Hmm.
Review
Quoted 5 lines
(quoted diff) +1
-1- if (!strcmp(buf,"keep")) + if (strstr(buf, "keep")) keep_early = 1; if (!strncmp(buf, "serial", 6)) {
This is pretty ugly. The whole reason for the bug in the first place was that "keep" was tested differently from all the other cases, and now you just perpetuate that.
All the other strings are tested with "!strncmp(..)", so I think this one should too. No?
Something like the appended (where I made the patch a bit bigger by just making it visually look the same too and adding the "} else if .." thing.
Or is there some reason you really _want_ "keep" to be different? If so, it should probably be commented on.
Linus
---
arch/x86_64/kernel/early_printk.c +2 -3
@@ -224,10 +224,9 @@ static int __init setup_early_printk(char *buf)return 0 ; early_console_initialized =1 ; -if (!strcmp (buf,"keep" )) +if (!strncmp (buf,"keep" ,4 )) { keep_early =1 ; - -if (!strncmp (buf,"serial" ,6 )) { + }else if (!strncmp (buf,"serial" ,6 )) {early_serial_init (buf +6 ); early_console = &early_serial_console; }else if (!strncmp (buf,"ttyS" ,4 )) {
Quoted 3 lines · Linus Torvalds
Or is there some reason you really _want_ "keep" to be different? If so, it should probably be commented on.
Hmm. Looking at the commit that broke, the "strstr()" was there originally, so in that sense your patch is obviously the minimal and safe one. So I'll apply it as-is after all, but I'd be even happier if somebody cleaned this up a bit and sent a patch after I do 2.6.19 (which may well be later today, I think the time has come).
Linus
Quoted 4 lines
Or is there some reason you really _want_ "keep" to be different? If so, it should probably be commented on.
It's just that keep is the only option that can only be at the end, all others can be followed by more, so it
made minor sense to use strcmp() to match the \0 too. But using strncmp everywhere is fine too since the syntax checking on these things is always quite weak and it probably doesn't make much difference either way.
I would have gone with Ingo's fix, but if you prefer strncmp..?
-Andi
arch/x86_64/kernel/early_printk.c static int __init setup_early_printk(char *buf) · line 231
Quoted 8 lines · Linus Torvalds
arch/x86_64/kernel/early_printk.c +2
-3-if (!strcmp (buf,"keep" )) +if (!strncmp (buf,"keep" ,4 )) { keep_early =1 ; - -if (!strncmp (buf,"serial" ,6 )) { + }else if (!strncmp (buf,"serial" ,6 )) {early_serial_init (buf +6 ); early_console = &early_serial_console;
nope, that doesnt work, because the function call is a one-time thing via the early_console_initialized flag. Nor does this keep compatibility with the 2.6.18 API, my existing boot-entries:
root (hd0,4)
kernel /boot/bzImage-x64 root=/dev/hda5 \
earlyprintk=serial,ttyS0,115200,keep console=ttyS0,115200 console=ttystopped working.
I agree that the parameter parsing here is a bit hacky, but my patch restores the original behavior, so i think that's the best option for now.
Ingo