Date: Sat, 16 Feb 2008 19:18:50 +0100 From: Juergen Lock <nox@jelal.kn-bremen.de> To: Kostik Belousov <kostikbel@gmail.com> Cc: emulation@FreeBSD.org Subject: Re: qemu-system-amd64 and freebsd guest Message-ID: <20080216181850.GA35839@saturn.kn-bremen.de> In-Reply-To: <20080216164331.GJ57756@deviant.kiev.zoral.com.ua> References: <20080216164331.GJ57756@deviant.kiev.zoral.com.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, Feb 16, 2008 at 06:43:31PM +0200, Kostik Belousov wrote:
> My recent attempts to boot CURRENT amd64 on QEMU 0.9.1 resulted in the
> consistent kernel panic in the first instruction of the root_bus_ocnfigure()
> function. Looking around, I noted that this instruction is the second one
> executed after "sti" in the amd64/amd64/autoconf.c:configure().
> CPU executes
> sti
> call root_bus_configure
> pushq %rbp <- General Protection Fault there
>
> It makes me suspect that fault happen when hardware interrupt is delivered.
> Indeed, looking around qemu source code and setting breakpoints in QEMU,
> I noted that hw interrupt is delivered, with into = -1. Calculation of
> the IDT index in the QEMU' target-i386/helper.c:do_interrupt64() as
> intno*16 and comparision with unsigned limit looked as overflow and GPF
> was delivered instead.
>
> Then, it appears that -1 comes from cpu_get_pic_interrupt() when
> apic_accept_pic_intr() returns false. This happen, in particular, when
> corresponding lapic vector is masked. The situation is legitimate, and
> I think that QEMU shall not deliver an interrupt in this case.
>
> The following patch allows me to boot the amd64 freebsd on qemu 0.9.1
> from ports.
>
> --- cpu-exec.c.orig 2008-02-16 18:23:53.134009488 +0200
> +++ cpu-exec.c 2008-02-16 18:24:47.127662872 +0200
> @@ -452,13 +452,15 @@
> svm_check_intercept(SVM_EXIT_INTR);
> env->interrupt_request &= ~(CPU_INTERRUPT_HARD | CPU_INTERRUPT_VIRQ);
> intno = cpu_get_pic_interrupt(env);
> - if (loglevel & CPU_LOG_TB_IN_ASM) {
> - fprintf(logfile, "Servicing hardware INT=0x%02x\n", intno);
> - }
> - do_interrupt(intno, 0, 0, 0, 1);
> - /* ensure that no TB jump will be modified as
> - the program flow was changed */
> - BREAK_CHAIN;
> + if (intno != -1) {
> + if (loglevel & CPU_LOG_TB_IN_ASM) {
> + fprintf(logfile, "Servicing hardware INT=0x%02x\n", intno);
> + }
> + do_interrupt(intno, 0, 0, 0, 1);
> + /* ensure that no TB jump will be modified as
> + the program flow was changed */
> + BREAK_CHAIN;
> + }
> #if !defined(CONFIG_USER_ONLY)
> } else if ((interrupt_request & CPU_INTERRUPT_VIRQ) &&
> (env->eflags & IF_MASK) && !(env->hflags & HF_INHIBIT_IRQ_MASK)) {
Hey, thanks for looking at this one! :) Maybe you should post the patch
on the qemu list so the right ppl can look at it, and see if it is correct:
http://lists.gnu.org/mailman/listinfo/qemu-devel
Thanx,
Juergen
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20080216181850.GA35839>