Re: [Qemu-ppc] [PATCH for-4.0 0/3] target/ppc: Fix pseries.cap-ibs=workaround with TCG

[Greg Kurz]

On Mon, 25 Mar 2019 11:31:04 +1100
Suraj Jitindar Singh <address@hidden> wrote:

> On Fri, 2019-03-22 at 19:03 +0100, Greg Kurz wrote:
> > Since recent commit 2782ad4c4102 "target/ppc/spapr: Enable
> > mitigations by
> > default for pseries-4.0 machine type", some recent distros, eg.
> > fedora29,
> > fail to boot under TCG because of a kernel panic:  
> 
> Good catch! I noticed this as well but didn't get around to fixing it
> so thanks for doing it for me :D
> 

My pleasure :)

> > 
> > [    0.614425] Oops: Exception in kernel mode, sig: 4 [#1]
> > [    0.618832] LE SMP NR_CPUS=1024 NUMA pSeries
> > [    0.621868] Modules linked in:
> > [    0.624958] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.20.16-
> > 200.fc29.ppc64le #1
> > [    0.625978] NIP:  c00000000000bf00 LR: c00000000000e268 CTR:
> > 0000000000007fff
> > [    0.626572] REGS: c0000000015a3750 TRAP: 0700   Not
> > tainted  (4.20.16-200.fc29.ppc64le)
> > [    0.626835] MSR:  8000000002089033 <SF,VEC,EE,ME,IR,DR,RI,LE>  CR:
> > 44828222  XER: 00000000
> > [    0.628497] CFAR: c00000000000bee4 IRQMASK: 1 
> > [    0.628497] GPR00: c00000000001fbc8 c0000000015a39e0
> > c0000000015a6200 c000000001533450 
> > [    0.628497] GPR04: c00000003e5819d0 c00000003ff15f68
> > 0000000000000000 0000000022f63728 
> > [    0.628497] GPR08: 0000000000000000 0000000000007fff
> > 0000000000000000 0000000000000000 
> > [    0.628497] GPR12: 0000000000008000 c0000000018b0000
> > 000000003dc5fd20 0000000002e75e90 
> > [    0.628497] GPR16: 0000000002e75d40 c00000003e594d00
> > 000000003dc5fd20 0000000002e90b50 
> > [    0.628497] GPR20: 000000003e45e800 c000000001531a80
> > c000000001532100 0000000044828222 
> > [    0.628497] GPR24: c000000001533450 c000000001533450
> > c00000003e5819d0 c0000000010e51e0 
> > [    0.628497] GPR28: c0000000015d5c18 c000000001531a80
> > c00000003e580000 c000000001531a80 
> > [    0.631710] NIP [c00000000000bf00] flush_count_cache+0x120/0x2420
> > [    0.631905] LR [c00000000000e268] _switch+0x68/0x180
> > [    0.632585] Call Trace:
> > [    0.633490] [c0000000015a39e0] [c0000000015dbd58]
> > __cpu_online_mask+0x0/0x80 (unreliable)
> > [    0.634383] [c0000000015a3bc0] [c00000000001fbc8]
> > __switch_to+0x348/0x500
> > [    0.634614] [c0000000015a3c20] [c000000000c509dc]
> > __schedule+0x2bc/0xac0
> > [    0.634731] [c0000000015a3cf0] [c000000000c51648]
> > preempt_schedule_common+0x38/0x60
> > [    0.634852] [c0000000015a3d10] [c000000000c516d4]
> > _cond_resched+0x64/0x80
> > [    0.635527] [c0000000015a3d40] [c00000000011a190]
> > copy_process.isra.4.part.5+0xc90/0x1d20
> > [    0.635656] [c0000000015a3e40] [c00000000011b414]
> > _do_fork+0xd4/0x470
> > [    0.635772] [c0000000015a3eb0] [c00000000011b88c]
> > kernel_thread+0x3c/0x50
> > [    0.635891] [c0000000015a3ed0] [c000000000010b08]
> > rest_init+0x98/0xf8
> > [    0.636025] [c0000000015a3f00] [c000000000fe4084]
> > start_kernel+0x658/0x67c
> > [    0.636163] [c0000000015a3f90] [c00000000000b37c]
> > start_here_common+0x1c/0x520
> > [    0.636763] Instruction dump:
> > [    0.640925] 48000005 48000005 48000005 48000005 48000005 4800001c
> > 60000000 60000000 
> > [    0.641448] 60000000 60000000 60000000 60000000 <7d2803a6>
> > 39207fff 7d2903a6 4c400420 
> > [    0.648580] ---[ end trace 1dcd9494acdef8df ]---
> > [    0.649361] 
> > [    1.657870] Kernel panic - not syncing: Attempted to kill the idle
> > task!
> > 
> > The following error is also printed by QEMU:
> > 
> > Opcode 13 10 10 00 (4c400420) leaked temporaries
> > 
> > The root cause behind the panic is that the linux kernel uses for
> > spectre v2
> > mitigation a form of the bcctr instruction that we don't support.
> > This gets
> > triggered when passing cap-ibs=workaround machine option, which is
> > the default
> > since 2782ad4c4102.
> > 
> > The TCG temp leak comes from some missing tcg_temp_free()s on the
> > exception path.
> > 
> > This series fixes the leak and adds support for the invalid form of
> > bcctr.
> > Since this adds yet another user of PPC_SEGMENT_64B to discriminate
> > CPU
> > models that should expose the _new_ behaviour, the final patch
> > introduces
> > a helper for that purpose.
> > 
> > --
> > Greg
> > 
> > ---
> > 
> > Greg Kurz (3):
> >       target/ppc: Fix TCG temporary leaks in gen_bcond()
> >       target/ppc: Enable "decrement and test CTR" version of bcctr
> >       target/ppc: Consolidate 64-bit server processor detection in a
> > helper
> > 
> > 
> >  hw/ppc/ppc.c             |    2 +-
> >  target/ppc/cpu.h         |    6 +++++
> >  target/ppc/helper_regs.h |    2 +-
> >  target/ppc/translate.c   |   58 ++++++++++++++++++++++++++++++++--
> > ------------
> >  4 files changed, 48 insertions(+), 20 deletions(-)
> > 
> >