qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] pxe boot problems


From: Laszlo Ersek
Subject: Re: [Qemu-devel] pxe boot problems
Date: Thu, 30 Jan 2014 12:38:28 +0100
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/20131118 Thunderbird/17.0.11

On 01/30/14 07:37, Dietmar Maurer wrote:
>> Does it work with TCG? 
> 
> It simply hangs a bit later if I use TCG, without any output on the console.

Strange. How recent qemu happens this with? The relevant emulation code
(under "ljmp Ev" in "target-i386/translate.c") has been changed as
recently as commit 78261634 (not in any release yet).

> But It works perfectly when I switch back to the pxe-XX.rom files.
> 
>> Also, can you try with a NIC model different from the default e1000?
> 
> same behavior with e1000, rtl8139, pcnet

These do match my results.

Please allow me to summarize the rest of the thread:

- New builds of iPXE contain funny jmp instructions.

- They are only present in the qemu tree in the efi-*.rom files,
  the pxe-*.rom builds date back to much earlier.

- When running the funny jmp instructions in a KVM guest, you either
  need "unrestricted_guest" support from the host CPU (check the
  "/sys/module/kvm_intel/parameters/unrestricted_guest" file when
  kvm-intel.ko is inserted), *or* you need to ask KVM to emulate invalid
  guest state, by passing "emulate_invalid_guest_state=1" to
  kvm-intel.ko -- check your module options under /etc/modprobe.d.
  (Note that you should rebuild the initramfs with dracut if you change
  those options.)

- In the latter case (ie. unrestricted_guest==0 &&
  emulate_invalid_guest_state==1), you will still run into an emulation
  problem on a current RHEL-6 host *later* (a different jmp insn in
  the iPXE builds). I filed RHBZ#1059496 for this and posted the
  backport last night. Gleb's upstream patches in question are e35b7b9c
  and ea79849d.

Laszlo



reply via email to

[Prev in Thread] Current Thread [Next in Thread]