Skip to contentSkip to conversation

: vm86 and xen kernels

PATCH In review Juan Quintela 2 participants 4 messages Mar 7, 2006 – Mar 8, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 : vm86 and xen kernels

Hi
vm86() call don't work at all on xen kernels, and it didn't worked
either with kernels previous to the TSS removal. VM86 requires TSS support, then it is better to disable it. Specially because it didn't work. This patch disables it for xen kernels.

Patch didn't apply because init/Kconfig don't exist on the sparse tree.

(And no, I don't know either why an x86 option is on that file :)

init/Kconfig renamed from init/Kconfig.orig +1 -1
diff -uNp linux-2.6.15.noarch/init/Kconfig.orig linux-2.6.15.noarch/init/Kconfig
@@ -224,7 +224,7 @@ config UID16
 	  This enables the legacy 16-bit UID syscall wrappers.
 
 config VM86
-	depends X86
+	depends X86 && !XEN
 	default y
 	bool "Enable VM86 support" if EMBEDDED
 	help

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

#1 lore raw
Quoted 8 lines · Juan Quintela

vm86() call don't work at all on xen kernels, and it didn't worked
either with kernels previous to the TSS removal. VM86 requires TSS support, then it is better to disable it. Specially because it didn't work. This patch disables it for xen kernels.

Patch didn't apply because init/Kconfig don't exist on the sparse tree.

(And no, I don't know either why an x86 option is on that file :)

We test vm86 support in our regression tests, so it does seem to be working.

-- Keir

#2 lore raw
Quoted 7 lines · Keir Fraser
Quoted 5 lines · Juan Quintela

vm86() call don't work at all on xen kernels, and it didn't worked
either with kernels previous to the TSS removal. VM86 requires TSS support, then it is better to disable it. Specially because it didn't work. This patch disables it for xen kernels.

Quoted 6 lines
Quoted 3 lines

Patch didn't apply because init/Kconfig don't exist on the sparse tree.

(And no, I don't know either why an x86 option is on that file :)

We test vm86 support in our regression tests, so it does seem to be working.

After further investigation, it shows that _sometimes_ it works. Problem is the definition of "sometimes". In any FC5 system, when kudzu tries to execute bios at 0xc00000 with vm86, kudzu dies with a segmentation fault.

It fails with xen-unstable kernels, i.e. it is not related to any fedora kernel patch, with xen-unstable configuration :(
It happens the same with yesterday xen-unstable kernels, but I don't have unmodified kernels installed just now. This kudzu version has been modified to print the arguments that are passed to vm86.

quintela$ uname -a
Linux gnomo.mitica 2.6.16-rc3-xen0 #1 Tue Mar 7 12:09:03 CET 2006 i686 athlon i386 GNU/Linux
quintela$ ~/work/kudzu-1.2.33/kudzu

ERROR - You must be root to run kudzu.
quintela$ sudo ~/work/kudzu-1.2.33/kudzu
ebx: 0
0
0
0
0
0
4f00
0
0
0
0

sudo strace ~/work/kudzu-1.2.33/kudzu

<lots of output removed>

write(1, "gs: 0 gsh:0\n", 12gs: 0 gsh:0
) = 12
vm86old(0x806f82c <unfinished ...>
--- SIGSEGV (Segmentation fault) @ 0 (0) ---
+++ killed by SIGSEGV +++
Process 3518 detached

This is completelly reproducible, and has happened I think that always.

Later, Juan.

#3 lore raw
Quoted 4 lines · Juan Quintela

After further investigation, it shows that _sometimes_ it works. Problem is the definition of "sometimes". In any FC5 system, when kudzu tries to execute bios at 0xc00000 with vm86, kudzu dies with a segmentation fault.

Is it trying to execute something at physical address 12MB? That would be unlikely to work.

-- Keir