Ever since v2.6.36 there's a conflict between OLPC support and Lguest guest support: a kernel image can't both support booting OLPC x86 hardware and booting as an Lguest guest. Booting a kernel image that tries to do both as an Lguest guest will fail with lguest: Reinjecting trap 13 for fault at 0x1000062: Invalid argument
This means that people wanting to boot an Lguest guest on 32 bits x86 needed to build their own kernel image without OLPC support since, it seems, Fedora 15. So let's stop pretending to support being an Lguest guest. People wanting to have that will now not only have to disable CONFIG_OLPC but also enable CONFIG_LGUEST_GUEST.
(A second order effect is that both CONFIG_VIRTIO and CONFIG_VIRTIO_CONSOLE will change from 'y' to 'm' in our builds. Assuming a sane build system no one should notice. We'll see.)
Signed-off-by: Paul Bolle pebolle@tiscali.nl --- config-x86-generic | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
Tested on F22. Applies cleanly to master (F25?) but I have not even tried building that.
The conflict between these two options is known for years now. See https://lists.ozlabs.org/pipermail/lguest/2013-April/001975.html .
(There's also a conflict with early microcode loading, see https://lists.ozlabs.org/pipermail/lguest/2013-May/002001.html . That one can actually be worked around by booting the guest with the dis_ucode_ldr kernel parameter.)
diff --git a/config-x86-generic b/config-x86-generic index 13de9462e556..c6c02779399f 100644 --- a/config-x86-generic +++ b/config-x86-generic @@ -283,7 +283,7 @@ CONFIG_KVM_INTEL=m CONFIG_KVM_AMD=m CONFIG_KVM_DEVICE_ASSIGNMENT=y CONFIG_LGUEST=m -CONFIG_LGUEST_GUEST=y +# CONFIG_LGUEST_GUEST is not set
CONFIG_HYPERVISOR_GUEST=y CONFIG_PARAVIRT=y
On Thu, Apr 14, 2016 at 6:48 AM, Paul Bolle pebolle@tiscali.nl wrote:
Ever since v2.6.36 there's a conflict between OLPC support and Lguest guest support: a kernel image can't both support booting OLPC x86 hardware and booting as an Lguest guest. Booting a kernel image that tries to do both as an Lguest guest will fail with lguest: Reinjecting trap 13 for fault at 0x1000062: Invalid argument
This means that people wanting to boot an Lguest guest on 32 bits x86 needed to build their own kernel image without OLPC support since, it seems, Fedora 15. So let's stop pretending to support being an Lguest guest. People wanting to have that will now not only have to disable CONFIG_OLPC but also enable CONFIG_LGUEST_GUEST.
I'd probably disable CONFIG_OLPC too. I'm planning on submitting a Change for F25 to drop the non-PAE kernel entirely, which will in effect drop OLPC support as well.
(A second order effect is that both CONFIG_VIRTIO and CONFIG_VIRTIO_CONSOLE will change from 'y' to 'm' in our builds. Assuming a sane build system no one should notice. We'll see.)
Signed-off-by: Paul Bolle pebolle@tiscali.nl
config-x86-generic | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
Tested on F22. Applies cleanly to master (F25?) but I have not even tried building that.
The conflict between these two options is known for years now. See https://lists.ozlabs.org/pipermail/lguest/2013-April/001975.html .
(There's also a conflict with early microcode loading, see https://lists.ozlabs.org/pipermail/lguest/2013-May/002001.html . That one can actually be worked around by booting the guest with the dis_ucode_ldr kernel parameter.)
lguest is a toy, and one that doesn't work very well it seems. Probably OK to disable it for now in any case. Let me think it over.
josh
diff --git a/config-x86-generic b/config-x86-generic index 13de9462e556..c6c02779399f 100644 --- a/config-x86-generic +++ b/config-x86-generic @@ -283,7 +283,7 @@ CONFIG_KVM_INTEL=m CONFIG_KVM_AMD=m CONFIG_KVM_DEVICE_ASSIGNMENT=y CONFIG_LGUEST=m -CONFIG_LGUEST_GUEST=y +# CONFIG_LGUEST_GUEST is not set
CONFIG_HYPERVISOR_GUEST=y CONFIG_PARAVIRT=y -- 2.4.11 _______________________________________________ kernel mailing list kernel@lists.fedoraproject.org http://lists.fedoraproject.org/admin/lists/kernel@lists.fedoraproject.org
Hi Josh,
On do, 2016-04-14 at 07:09 -0400, Josh Boyer wrote:
lguest is a toy, and one that doesn't work very well it seems. Probably OK to disable it for now in any case. Let me think it over.
What I wasn't sure about is whether we should bother setting CONFIG_LGUEST (ie, the host side support). I kept it enabled. Perhaps that could be dropped too.
(That in my opinion Lguest guest support offers little value should be clear.)
Thanks,
Paul Bolle
Hi Josh,
On do, 2016-04-14 at 13:17 +0200, Paul Bolle wrote:
On do, 2016-04-14 at 07:09 -0400, Josh Boyer wrote:
lguest is a toy, and one that doesn't work very well it seems. Probably OK to disable it for now in any case. Let me think it over.
What I wasn't sure about is whether we should bother setting CONFIG_LGUEST (ie, the host side support). I kept it enabled. Perhaps that could be dropped too.
Nothing seems to have happened after this. So I'm guessing you decided to keep Lguest guest support enabled (perhaps because you expect to drop OLPC support instead).
Did I guess right?
Paul Bolle
On Mon, May 30, 2016 at 4:54 AM, Paul Bolle pebolle@tiscali.nl wrote:
Hi Josh,
On do, 2016-04-14 at 13:17 +0200, Paul Bolle wrote:
On do, 2016-04-14 at 07:09 -0400, Josh Boyer wrote:
lguest is a toy, and one that doesn't work very well it seems. Probably OK to disable it for now in any case. Let me think it over.
What I wasn't sure about is whether we should bother setting CONFIG_LGUEST (ie, the host side support). I kept it enabled. Perhaps that could be dropped too.
Nothing seems to have happened after this. So I'm guessing you decided to keep Lguest guest support enabled (perhaps because you expect to drop OLPC support instead).
Did I guess right?
No, I just forgot about it. I'll drop LGUEST today.
josh
On (Thu) 14 Apr 2016 [12:48:59], Paul Bolle wrote:
Ever since v2.6.36 there's a conflict between OLPC support and Lguest guest support: a kernel image can't both support booting OLPC x86 hardware and booting as an Lguest guest. Booting a kernel image that tries to do both as an Lguest guest will fail with lguest: Reinjecting trap 13 for fault at 0x1000062: Invalid argument
This means that people wanting to boot an Lguest guest on 32 bits x86 needed to build their own kernel image without OLPC support since, it seems, Fedora 15. So let's stop pretending to support being an Lguest guest. People wanting to have that will now not only have to disable CONFIG_OLPC but also enable CONFIG_LGUEST_GUEST.
(A second order effect is that both CONFIG_VIRTIO and CONFIG_VIRTIO_CONSOLE will change from 'y' to 'm' in our builds. Assuming a sane build system no one should notice. We'll see.)
CONFIG_VIRTIO_CONSOLE was =y because anaconda couldn't load the module at startup (but I think it's been fixed since a while now). It doesn't make sense to keep this at =y, so even explicitly marking both VIRTIO and VIRTIO_CONSOLE to =m is advised.
Amit
kernel@lists.fedoraproject.org
-
Amit Shah -
Josh Boyer -
Paul Bolle