Skip to contentSkip to conversation

x86: fix to parse multiboot command line passed by latest grub

PATCH Superseded Wei Yongjun 5 participants 10 messages Dec 7, 2009 – Dec 8, 2009

lore
Share & tools
lore
#0 lore raw

latest grub had changed to "don't pass filename in multiboot command line".

The old cmdline format is: "module-name options..." The new cmdline format is: "options..."

So xen + grub2 always loss the first option, because xen will skip the first option.

Usually, the module-name is not start with char [a-zA-Z], such as /boot/vmlinuz, (hd0,1)/boot/vmlinuz, so this patch added isalpha() test of the first char of module-name/option to function cmdline_cook(), try to handle both the old and new cmdline format.

xen/arch/x86/setup.c +2 -0
@@ -423,6 +423,8 @@
     p = p ? : "";
     while ( *p == ' ' )
         p++;
+    if (isalpha(*p))
+        return p;
     while ( (*p != ' ') && (*p != '\0') )
         p++;
     while ( *p == ' ' )
#1 lore raw

PATCH 1/1 x86: fix to parse multiboot command line passed by latest grub

latest grub had changed to "don't pass filename in multiboot command line".

The old cmdline format is: "module-name options..." The new cmdline format is: "options..."

So xen + grub2 always loss the first option, because xen will skip the first option.

Usually, the module-name is not start with char [a-zA-Z], such as /boot/vmlinuz, (hd0,1)/boot/vmlinuz, so this patch added isalpha() test of the first char of modulename/option to function cmdline_cook(), try to handle both the old and new cmdline format.

xen/arch/x86/setup.c +2 -0
@@ -423,6 +423,8 @@
     p = p ? : "";
     while ( *p == ' ' )
         p++;
+    if (isalpha(*p))
+        return p;
     while ( (*p != ' ') && (*p != '\0') )
         p++;
     while ( *p == ' ' )
#2 lore raw
Quoted 6 lines · Wei Yongjun

latest grub had changed to "don't pass filename in multiboot command line".

The old cmdline format is: "module-name options..." The new cmdline format is: "options..."

Quoted 8 lines

So xen + grub2 always loss the first option, because xen will skip the first option.

Usually, the module-name is not start with char [a-zA-Z], such as /boot/vmlinuz, (hd0,1)/boot/vmlinuz, so this patch added isalpha() test of the first char of modulename/option to function cmdline_cook(), try to handle both the old and new cmdline format.

Xen doesn't actually complain about unknown options, does it? So why not simply stop stripping the first option altogether? Under grub1 Xen will just ignore it and under grub2 it won't be passed in.

Ian.

Quoted 22 lines
xen/arch/x86/setup.c +2 -0
@@ -423,6 +423,8 @@
     p = p ? : "";
     while ( *p == ' ' )
         p++;
+    if (isalpha(*p))
+        return p;
     while ( (*p != ' ') && (*p != '\0') )
         p++;
     while ( *p == ' ' )

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#3 lore raw
Quoted 18 lines · Ian Campbell
Quoted 6 lines · Wei Yongjun

latest grub had changed to "don't pass filename in multiboot command line".

The old cmdline format is: "module-name options..." The new cmdline format is: "options..."

Quoted 9 lines

So xen + grub2 always loss the first option, because xen will skip the first option.

Usually, the module-name is not start with char [a-zA-Z], such as /boot/vmlinuz, (hd0,1)/boot/vmlinuz, so this patch added isalpha() test of the first char of modulename/option to function cmdline_cook(), try to handle both the old and new cmdline format.

Quoted 4 lines

Xen doesn't actually complain about unknown options, does it? So why not simply stop stripping the first option altogether? Under grub1 Xen will just ignore it and under grub2 it won't be passed in.

That it is OK. But the module-name will pass to the dom0, the cmdline of dom0 like this
under grub1:

# cat /proc/cmdline
/boot/vmlinuz-2.6.31.6-xen
root=UUID=575f3be9-0a50-4d62-a36d-800e85a7aa70 ro console=tty0

Other way, I will send a patch as your advice^_^.

Thanks

Quoted 30 lines

Ian.

Quoted 22 lines
xen/arch/x86/setup.c +2 -0
@@ -423,6 +423,8 @@
     p = p ? : "";
     while ( *p == ' ' )
         p++;
+    if (isalpha(*p))
+        return p;
     while ( (*p != ' ') && (*p != '\0') )
         p++;
     while ( *p == ' ' )

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#4 lore raw

PATCH v2 1/1 x86: fix to parse multiboot command line passed by latest grub

In reply to Ian Campbell

latest grub had changed to "don't pass filename in multiboot command line".

The old cmdline format is: "module-name options..." The new cmdline format is: "options..."

So xen + grub2 always loss the first option, because xen will skip the first option.

Since xen doesn't actually complain about unknown options, so simply stop stripping the first option altogether, under grub1 Xen will just ignore it and under grub2 it won't be passed in.

xen/arch/x86/setup.c +0 -4
@@ -423,10 +423,6 @@
     p = p ? : "";
     while ( *p == ' ' )
         p++;
-    while ( (*p != ' ') && (*p != '\0') )
-        p++;
-    while ( *p == ' ' )
-        p++;
     return p;
 }
#5 lore raw
In reply to Wei Yongjun

I recommend just sticking a dummy first arg when writing a command line for grub2, and living with it. Or get grub2 guys to give us a flag to detect this stupid compatibility breakage. I don't really want hacks like this.

-- Keir

Quoted 31 lines · Wei Yongjun

latest grub had changed to "don't pass filename in multiboot command line".

The old cmdline format is: "module-name options..." The new cmdline format is: "options..."

So xen + grub2 always loss the first option, because xen will skip the first option.

Usually, the module-name is not start with char [a-zA-Z], such as /boot/vmlinuz, (hd0,1)/boot/vmlinuz, so this patch added isalpha() test of the first char of module-name/option to function cmdline_cook(), try to handle both the old and new cmdline format.

xen/arch/x86/setup.c +2 -0
@@ -423,6 +423,8 @@
     p = p ? : "";
     while ( *p == ' ' )
         p++;
+    if (isalpha(*p))
+        return p;
     while ( (*p != ' ') && (*p != '\0') )
         p++;
     while ( *p == ' ' )
#6 lore raw
Quoted 2 lines · Keir Fraser, le Mon 07 Dec 2009 09:57:41 +0000, a écrit :

I recommend just sticking a dummy first arg when writing a command line for grub2, and living with it.

That's what we did in GNU Mach.

Quoted 2 lines

Or get grub2 guys to give us a flag to detect
this stupid compatibility breakage.

They said multiboot never asserted anything about the command line and thus implementation were allowed to vary without notice. The rationale for grub's behavior change is that the user has to migrate from grub1 syntax to grub2 syntax anyway.

Quoted 1 line

I don't really want hacks like this.

It was out of question to commit things like this in GNU Mach as well :)

Samuel

#7 lore raw
Quoted 9 lines · Samuel Thibault
Quoted 2 lines · Keir Fraser, le Mon 07 Dec 2009 09:57:41 +0000, a écrit :

Or get grub2 guys to give us a flag to detect
this stupid compatibility breakage.

They said multiboot never asserted anything about the command line and thus implementation were allowed to vary without notice. The rationale for grub's behavior change is that the user has to migrate from grub1 syntax to grub2 syntax anyway.

They broke it because they could? :-)

#8 lore raw
Quoted 6 lines · Simon Horman, le Tue 08 Dec 2009 09:00:00 +1100, a écrit :
Quoted 5 lines · Samuel Thibault
Quoted 2 lines · Keir Fraser, le Mon 07 Dec 2009 09:57:41 +0000, a écrit :

Or get grub2 guys to give us a flag to detect
this stupid compatibility breakage.

Quoted 6 lines
Quoted 4 lines

They said multiboot never asserted anything about the command line and thus implementation were allowed to vary without notice. The rationale for grub's behavior change is that the user has to migrate from grub1 syntax to grub2 syntax anyway.

They broke it because they could? :-)

Yes...

Samuel

#9 lore raw
In reply to Simon Horman
Quoted 9 lines · Simon Horman
Quoted 7 lines
Quoted 2 lines

Or get grub2 guys to give us a flag to detect
this stupid compatibility breakage.

They said multiboot never asserted anything about the command line and thus implementation were allowed to vary without notice. The rationale for grub's behavior change is that the user has to migrate from grub1 syntax to grub2 syntax anyway.

They broke it because they could? :-)

Yes, it's both stupid and arrogant. Oh well.

-- Keir