Re: domU crash when resuming from suspend - Xen 4.22

["Marek Marczykowski-Górecki" <marmarek@invisiblethingslab.com>]

[-- Attachment #1: Type: text/plain, Size: 9461 bytes --]

On Sun, Jul 19, 2026 at 08:31:28AM +0100, Frediano Ziglio wrote:
> On Sun, 19 Jul 2026 at 03:00, Marek Marczykowski-Górecki
> <marmarek@invisiblethingslab.com> wrote:
> >
> > On Sat, Jul 18, 2026 at 04:03:29PM +0200, Marek Marczykowski-Górecki wrote:
> > > On Fri, Jul 17, 2026 at 09:22:47PM +0100, Frediano Ziglio wrote:
> > > > On Tue, 14 Jul 2026 at 18:12, Marek Marczykowski-Górecki
> > > > <marmarek@invisiblethingslab.com> wrote:
> > > > >
> > > > > On Fri, Jun 19, 2026 at 01:24:09PM +0200, Marek Marczykowski-Górecki wrote:
> > > > > > Hi,
> > > > > >
> > > > > > I'm continuing testing 4.22 and found this:
> > > > > >
> > > > > > When a domU is suspended (even without host suspend) it crashes on
> > > > > > resume with:
> > > > > >
> > > > > >     [   24.374629] BUG: kernel NULL pointer dereference, address: 0000000000000020
> > > > > >     [   24.374647] #PF: supervisor read access in kernel mode
> > > > > >     [   24.374657] #PF: error_code(0x0000) - not-present page
> > > > > >     [   24.374668] PGD 0 P4D 0
> > > > > >     [   24.374675] Oops: Oops: 0000 [#1] SMP NOPTI
> > > > > >     [   24.374685] CPU: 0 UID: 0 PID: 19 Comm: migration/0 Not tainted 6.18.31-1.qubes.13.fc41.x86_64 #1 PREEMPT(full)
> > > > > >     [   24.374705] Stopper: multi_cpu_stop+0x0/0x140 <- __stop_cpus.constprop.0+0x6f/0xb0
> > > > > >     [   24.374723] RIP: 0010:multi_cpu_stop+0x63/0x140
> > > > > >     [   24.374734] Code: 49 0f a3 1c 24 41 0f 92 c6 c7 44 24 04 00 00 00 00 31 ed 49 8d 5f 24 eb 2d 41 83 fd 02 74 6e 41 83 fd 03 74 49 f0 ff 0b 75 13 <41> 8b 47 20 41 8b 77 10 83 c0 01 41 89 77 24 41 89 47 20 41 83 fd
> > > > > >     [   24.374764] RSP: 0018:ffffc900000abe30 EFLAGS: 00010046
> > > > > >     [   24.374775] RAX: 0000000000000000 RBX: ffffc90000a13de4 RCX: 0000000000000404
> > > > > >     [   24.374790] RDX: 0000000000000040 RSI: ffffffffffffffff RDI: 0000000000770ef0
> > > > > >     [   24.374801] RBP: 0000000000000002 R08: ffffc900000abc80 R09: ffffc900000abc8c
> > > > > >     [   24.374813] R10: 0000000000000000 R11: ffff88801e81cdc0 R12: ffffffff81835ac0
> > > > > >     [   24.374825] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
> > > > > >     [   24.374837] FS:  0000000000000000(0000) GS:ffff88809b369000(0000) knlGS:0000000000000000
> > > > > >     [   24.374850] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > > > >     [   24.374860] CR2: 0000000000000020 CR3: 000000000d6a2002 CR4: 0000000000770ef0
> > > > > >     [   24.374873] PKRU: 00000000
> > > > > >     [   24.374877] Call Trace:
> > > > > >     [   24.374884]  <TASK>
> > > > > >     [   24.374891]  ? __pfx_multi_cpu_stop+0x10/0x10
> > > > > >     [   24.374900]  cpu_stopper_thread+0xa3/0x170
> > > > > >     [   24.374908]  ? __pfx_smpboot_thread_fn+0x10/0x10
> > > > > >     [   24.374917]  smpboot_thread_fn+0xf3/0x220
> > > > > >     [   24.374925]  kthread+0xfc/0x240
> > > > > >     [   24.374933]  ? __pfx_kthread+0x10/0x10
> > > > > >     [   24.374940]  ? __pfx_kthread+0x10/0x10
> > > > > >     [   24.374949]  ret_from_fork+0x158/0x170
> > > > > >     [   24.374957]  ? __pfx_kthread+0x10/0x10
> > > > > >     [   24.374965]  ret_from_fork_asm+0x1a/0x30
> > > > > >     [   24.374973]  </TASK>
> > > > > >     [   24.374978] Modules linked in: snd_seq_dummy snd_hrtimer snd_seq snd_seq_device snd_timer snd soundcore cfg80211 rfkill vfat fat xenfs binfmt_misc nft_reject_inet nf_reject_ipv4 nf_reject_ipv6 nft_reject nft_ct nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 nf_tables intel_rapl_msr intel_rapl_common intel_uncore_frequency_common intel_pmc_ssram_telemetry intel_vsec polyval_clmulni ghash_clmulni_intel xen_netfront xen_privcmd xen_gntdev xen_gntalloc xen_blkback xen_evtchn fuse loop nfnetlink ip_tables overlay xen_blkfront
> > > > > >     [   24.375064] CR2: 0000000000000020
> > > > > >     [   24.375071] ---[ end trace 0000000000000000 ]---
> > > > > >     [   24.375080] RIP: 0010:multi_cpu_stop+0x63/0x140
> > > > > >     [   24.375089] Code: 49 0f a3 1c 24 41 0f 92 c6 c7 44 24 04 00 00 00 00 31 ed 49 8d 5f 24 eb 2d 41 83 fd 02 74 6e 41 83 fd 03 74 49 f0 ff 0b 75 13 <41> 8b 47 20 41 8b 77 10 83 c0 01 41 89 77 24 41 89 47 20 41 83 fd
> > > > > >     [   24.375115] RSP: 0018:ffffc900000abe30 EFLAGS: 00010046
> > > > > >     [   24.375124] RAX: 0000000000000000 RBX: ffffc90000a13de4 RCX: 0000000000000404
> > > > > >     [   24.375142] RDX: 0000000000000040 RSI: ffffffffffffffff RDI: 0000000000770ef0
> > > > > >     [   24.375155] RBP: 0000000000000002 R08: ffffc900000abc80 R09: ffffc900000abc8c
> > > > > >     [   24.375170] R10: 0000000000000000 R11: ffff88801e81cdc0 R12: ffffffff81835ac0
> > > > > >     [   24.375184] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
> > > > > >     [   24.375200] FS:  0000000000000000(0000) GS:ffff88809b369000(0000) knlGS:0000000000000000
> > > > > >     [   24.375214] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > > > >     [   24.375226] CR2: 0000000000000020 CR3: 000000000d6a2002 CR4: 0000000000770ef0
> > > > > >     [   24.375241] PKRU: 00000000
> > > > > >     [   24.375247] Kernel panic - not syncing: Fatal exception
> > > > > >     [   25.417403] Shutting down cpus with NMI
> > > > > >     [   25.417447] Kernel Offset: disabled
> > > > > >
> > > > > > There is no change on the Linux kernel side (neither dom0 nor domU), the
> > > > > > only change is updating Xen 4.19.4 -> 4.22-rc2 (seems to be already
> > > > > > broken on -rc1 too). This happens for both HVM and PVH domains. Haven't
> > > > > > tried PV domU (but dom0 survives host suspend).
> > > > >
> > > > > Any ideas?
> > > > >
> > > >
> > > > Hi,
> > > >    not many ideas but while testing my libxenguest changes I found
> > > > similar failures using xl create/save/restore. But I was testing more
> > > > PV guests.
> > > > Does 4.21 work ?
> > >
> > > It seems 4.21 also has this issue:
> > > https://gitlab.com/xen-project/people/marmarek/xen/-/jobs/15414211523
> > > But 4.20 is okay:
> > > https://gitlab.com/xen-project/people/marmarek/xen/-/jobs/15414227095
> > >
> > > I guess I can bisect it further.
> >
> > So, bisect says the first bad commit is:
> >
> >     commit 9f892f84c27970840d255aef7e2851b762967de7
> >     Author: Andrew Cooper <andrew.cooper3@citrix.com>
> >     Date:   Mon Dec 30 11:49:14 2024 +0000
> >
> >         x86/domctl: Stop using XLAT_cpu_user_regs()
> >
> > My test isn't 100% reliable, but I retested parent of this commit a
> > couple of times and it was good every time.
> >
> > I wanted to test reverting it on top of 4.21, but there is a lot of
> > FRED-related work on top that makes such revert quite hard, to say the
> > least...
> >
> 
> The generated XLAT_cpu_user_regs was
> 
> #define XLAT_cpu_user_regs(_d_, _s_) do { \
>     (_d_)->__DECL_REG_LO8(b) = (_s_)->__DECL_REG_LO8(b); \
>     (_d_)->__DECL_REG_LO8(c) = (_s_)->__DECL_REG_LO8(c); \
>     (_d_)->__DECL_REG_LO8(d) = (_s_)->__DECL_REG_LO8(d); \
>     (_d_)->__DECL_REG_LO16(si) = (_s_)->__DECL_REG_LO16(si); \
>     (_d_)->__DECL_REG_LO16(di) = (_s_)->__DECL_REG_LO16(di); \
>     (_d_)->__DECL_REG_LO16(bp) = (_s_)->__DECL_REG_LO16(bp); \
>     (_d_)->__DECL_REG_LO8(a) = (_s_)->__DECL_REG_LO8(a); \
>     (_d_)->error_code = (_s_)->error_code; \
>     (_d_)->entry_vector = (_s_)->entry_vector; \
>     (_d_)->__DECL_REG_LO16(ip) = (_s_)->__DECL_REG_LO16(ip); \
>     (_d_)->cs = (_s_)->cs; \
>     (_d_)->saved_upcall_mask = (_s_)->saved_upcall_mask; \
>     (_d_)->__DECL_REG_LO16(flags) = (_s_)->__DECL_REG_LO16(flags); \
>     (_d_)->__DECL_REG_LO16(sp) = (_s_)->__DECL_REG_LO16(sp); \
>     (_d_)->ss = (_s_)->ss; \
>     (_d_)->es = (_s_)->es; \
>     (_d_)->ds = (_s_)->ds; \
>     (_d_)->fs = (_s_)->fs; \
>     (_d_)->gs = (_s_)->gs; \
> } while (0)
> 
> registers are copied, but additional fields (error_code, entry_vector,
> saved_upcall_mask).
> Maybe you could try to copy them again.
> Also, just to be on the safe side, you could memset the structure
> where "Backing memory is pre-zeroed" comments are.

This does not help :(

Specifically:
1. Applied on top of staging-4.21:
  - build: https://gitlab.com/xen-project/people/marmarek/xen/-/commits/188ca612c51a57ef54b917c833eaeb59666994c9
  - test (fails): https://gitlab.com/xen-project/people/marmarek/xen/-/pipelines/2688813821
2. Applied directly on top of 9f892f84c27970840d255aef7e2851b762967de7:
  - build: https://gitlab.com/xen-project/people/marmarek/xen/-/commits/dd3dee35fc4fcd6ba284f3f5fa78b333c472888e
  - test (fails): https://gitlab.com/xen-project/people/marmarek/xen/-/pipelines/2688832398

Yet, parent of 9f892f84c27970840d255aef7e2851b762967de7 still passes:
 - build: https://gitlab.com/xen-project/people/marmarek/xen/-/commits/dbcbbed4e9dc25faa211d359c2f04a9c70f087c9
 - test (pass): https://gitlab.com/xen-project/people/marmarek/xen/-/pipelines/2688835512

Hint about reading the above: I use two branches, one for building
(branch "test") and one for running the test (branch "automation-uki-grub-only"),
which takes build artifacts from the build branch. I do that to not
rebase test scripts. It's a bit opaque, but the console log of the test
contains "Latest ChangeSet" fields telling what actually was tested.

-- 
Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]