Re: [PATCH] accel/tcg: Complete cpu initialization before registration

[Eric Auger]

Hi Richard,

Commit 4e4fa6c12d ("accel/tcg: Complete cpu initialization before
registration") seems to cause a regression on one kvm unit test:

FAIL debug-wp-migration (terminated on SIGSEGV)

This can be reproduced with upstream kernel, qemu and kvm unit test.

Seems the change in accel/tcg/translate-all.c is the cause of the SIGSEV
(the removal of the allocation of jc (CPUJumpCache)).

If I restore
    if (unlikely(jc == NULL)) {
        jc = g_new0(CPUJumpCache, 1);
        jc = qatomic_xchg(&cpu->tb_jmp_cache, jc);
        assert(jc == NULL);
        return;
    }
I don't get the crash anymore.

What I fail to understand is why this code is called with a kvm
accelerated qemu (the test runs by default with kvm).


#0  0x000002aaab41ee94 in tcg_flush_jmp_cache (cpu=cpu@entry=0x2aaac391910)
at ../accel/tcg/translate-all.c:1581
#1  0x000002aaab423458 in tlb_flush_by_mmuidx_async_work
(cpu=0x2aaac391910, data=...)
at ../accel/tcg/cputlb.c:360
#2  0x000002aaaae0b1d0 in process_queued_cpu_work
(cpu=cpu@entry=0x2aaac391910) at ../cpus-common.c:351
<augere> (cpu=cpu@entry=0x2aaac391910)

#0  0x000002aaab423658 in tlb_flush_by_mmuidx (cpu=0x2aaac391910,
idxmap=4095)
at ../accel/tcg/cputlb.c:377
#1  0x000002aaab4236e8 in tlb_flush (cpu=cpu@entry=0x2aaac391910) at
../accel/tcg/cputlb.c:391
#2  0x000002aaab1500f0 in vmsa_ttbr_write
(env=0x2aaac393850, ri=0x2aaac3c90e0, value=2154950976315703518) at
../target/arm/helper.c:3784
#3  0x000002aaab14e5a8 in write_raw_cp_reg
(env=env@entry=0x2aaac393850, ri=ri@entry=0x2aaac3c90e0,
v=v@entry=2154950976315703518)
at ../target/arm/helper.c:96
#4  0x000002aaab153f1c in write_list_to_cpustate
(cpu=cpu@entry=0x2aaac391910)
at ../target/arm/helper.c:191
#5  0x000002aaab20f24c in kvm_arm_reset_vcpu
(cpu=cpu@entry=0x2aaac391910) at ../target/arm/kvm.c:634
#6  0x000002aaab147cbc in arm_cpu_reset (dev=0x2aaac391910) at
../target/arm/cpu.c:522

Thanks

Eric