Hi lianbo & guanyou,
On Thu, Nov 21, 2024 at 8:56 PM Guanyou Chen <chenguanyou9338(a)gmail.com> wrote:
Hi Tao
> 1. What is the root cause, why cpu offline state will lead to mapping
> errors? You didn't make this clear in your commit message.
The back CPU context wrong order When the front CPU is offline.
exp:
CPU0, CPU1 is offline, CPU2 is online, CPU2's nt_prstatus_percpu from corefile note
prstatus[0], because i, j not equal.
> 2. Is this issue cpu arch specific? Since the example you gave is only
> arm64 specific, however the code change is generic for all archs. So
> I'm a bit worried about whether it will make the result incorrect for
> other archs. In addition, I didn't reproduce the mapping error issue
> on my x86_64 machine, so I cannot verify the patch locally.
N/A
> What command line you trigger will get the following outputs?
$ crash vmlinux vmcore -d 1
crash> help -r
Thanks,
Guanyou
Tao Liu <ltao(a)redhat.com> 于2024年11月21日周四 13:42写道:
>
> Hi Guanyou,
>
> Thanks for the fix. However I have a few questions for this patch:
>
> On Fri, Nov 1, 2024 at 10:16 PM Guanyou Chen <chenguanyou9338(a)gmail.com>
wrote:
> >
> > Hi Lianbo, Tao
> >
> > When CPUs are in an offline state, it can lead to mapping errors.
>
> 1. What is the root cause, why cpu offline state will lead to mapping
> errors? You didn't make this clear in your commit message.
> 2. Is this issue cpu arch specific? Since the example you gave is only
> arm64 specific, however the code change is generic for all archs. So
> I'm a bit worried about whether it will make the result incorrect for
> other archs. In addition, I didn't reproduce the mapping error issue
> on my x86_64 machine, so I cannot verify the patch locally.
>
> > We need to map them to the correct positions one by one.
> >
>
> 3. Please make the commit log tidy. I know you want to diff the
> outputs of "before" and "after", but we only need the info which
are
> highly related to this issue. e.g:
>
> > Before:
>
> What command line you trigger will get the following outputs?
>
> > n_namesz: 5 ("CPU2")
> > n_descsz: 392
> > n_type: 1 (NT_PRSTATUS)
> > si.signo: 0 si.code: 0 si.errno: 0
> > cursig: 0 sigpend: 0 sighold: 0
> > pid: 3 ppid: 0 pgrp: 0 sid:0
> > utime: 0.000000 stime: 0.000000
> > cutime: 0.000000 cstime: 0.000000
>
> I'm not sure if the above info is related to this issue?
>
> > X0: ffffffc000fc8818 X1: 0000000000000000 X2: ffffffc000fc84c8
> > X3: 0000000000000000 X4: ffffffc0405e37bf X5: ffffffc00a07372f
> > X6: 322e34323320205b X7: 545b5d3539383334 X8: ffffffc000fc2f0c
> > X9: 89fece0a9ef8cb00 X10: c0000001001f75f4 X11: 00000001001f75f4
> > X12: 0000000000000003 X13: 00000000000005f4 X14: ffffffc009eb1210
> > X15: 0000000000000004 X16: 000000002a4cec24 X17: 000000002a4cec24
> > X18: ffffffc009e7d140 X19: ffffffc00a04c670 X20: 0000000000000000
> > X21: 0000000000000000 X22: ffffff8027f22280 X23: 0000000000000009
> > X24: 0000000000000007 X25: ffffffc009f839c0 X26: ffffffc0090f87f8
> > X27: 0000000000000000 X28: ffffff80454f3840 X29: ffffffc0405e3b60
> > LR: ffffffc0080e57fc SP: ffffffc0405e3b60 PC: ffffffc000fc2f84
>
> We don't need a list of all regs, a few lines of which can indicate
> the output mismatch for "Before" and "After" is enough.
>
> >
> > CPU 0: [OFFLINE]
> > CPU 1: [OFFLINE]
> > CPU 2:
> > X0: 0000000000000000 X1: 0000000000000000 X2: 0000000000000000
> > X3: 000000000003fcbc X4: 0000000000000001 X5: 0000000000000000
> > X6: 0000000000000000 X7: 0000000000000000 X8: 00000000ffffffff
> > X9: ffffffc009e6ae48 X10: ffffffc009e6ae20 X11: 0000000000000000
> > X12: 0000000000000002 X13: 0000000000000004 X14: 0000000000000000
> > X15: 0000000000004000 X16: 00000000f90f05f6 X17: 00000000f90f05f6
> > X18: 0000000000000000 X19: 0000000000000002 X20: ffffffc009e3b008
> > X21: ffffffc00a01d020 X22: ffffffc009f798f0 X23: 0000000060001000
> > X24: 0000000000000000 X25: 0000000000000000 X26: 0000000000000000
> > X27: 0000000000000000 X28: ffffff8111eecb00 X29: ffffffc008003f50
> > LR: ffffffc00802df88 SP: ffffffc008003f40 PC: ffffffc00802df94
> > PSTATE: 024003c5 FPVALID: 00000000
> >
> > After:
> > CPU 2:
> > X0: ffffffc000fc8818 X1: 0000000000000000 X2: ffffffc000fc84c8
> > X3: 0000000000000000 X4: ffffffc0405e37bf X5: ffffffc00a07372f
> > X6: 322e34323320205b X7: 545b5d3539383334 X8: ffffffc000fc2f0c
> > X9: 89fece0a9ef8cb00 X10: c0000001001f75f4 X11: 00000001001f75f4
> > X12: 0000000000000003 X13: 00000000000005f4 X14: ffffffc009eb1210
> > X15: 0000000000000004 X16: 000000002a4cec24 X17: 000000002a4cec24
> > X18: ffffffc009e7d140 X19: ffffffc00a04c670 X20: 0000000000000000
> > X21: 0000000000000000 X22: ffffff8027f22280 X23: 0000000000000009
> > X24: 0000000000000007 X25: ffffffc009f839c0 X26: ffffffc0090f87f8
> > X27: 0000000000000000 X28: ffffff80454f3840 X29: ffffffc0405e3b60
>
> > LR: ffffffc0080e57fc SP: ffffffc0405e3b60 PC: ffffffc000fc2f84
> > PSTATE: 600000c5 FPVALID: 00000000
> >
> > crash> bt
> > PID: 15959 TASK: ffffff80454f3840 CPU: 2 COMMAND:
"AnrConsumer"
> > [ffffffc0405e3b60] ipanic at ffffffc000fc2f80 [mrdump]
> > [ffffffc0405e3b70] atomic_notifier_call_chain at ffffffc0080e57f8
> > [ffffffc0405e3c30] panic at ffffffc008f734d0
> > [ffffffc0405e3c80] sysrq_handle_crash at ffffffc0087f3c18
> > [ffffffc0405e3c90] __handle_sysrq at ffffffc0087f3798
> > [ffffffc0405e3ce0] write_sysrq_trigger at ffffffc0087f49c0
> > [ffffffc0405e3d00] proc_reg_write at ffffffc00842e4b8
> > [ffffffc0405e3d80] vfs_write at ffffffc008381eb4
> > [ffffffc0405e3dd0] ksys_write at ffffffc008382200
> > [ffffffc0405e3e10] __arm64_sys_write at ffffffc00838228c
> > [ffffffc0405e3e20] invoke_syscall at ffffffc00802efe0
> > [ffffffc0405e3e40] el0_svc_common at ffffffc00802eef4
> > [ffffffc0405e3e70] do_el0_svc at ffffffc00802ede8
> > [ffffffc0405e3e80] el0_svc at ffffffc008f7a7d0
> > [ffffffc0405e3ea0] el0t_64_sync_handler at ffffffc008f7a758
> > [ffffffc0405e3fe0] el0t_64_sync at ffffffc00801157c
>
> The stacktrace is meaningless for this patch, we can cut this off.
>
> > PC: 00000077c798ca28 LR: 00000077a82e19f4 SP: 000000761c517af0
> > X29: 000000761c517b00 X28: 000000761c517db8 X27: 000000761c517c90
> > X26: 000000761c517c98 X25: 000000761c517bf9 X24: 000000761c519000
> > X23: 000000761c517be1 X22: 0000000000000001 X21: 00000000000003e3
> > X20: 000000761c517c11 X19: 000000761c517bf8 X18: 0000007568224000
> > X17: 00000077c798ca20 X16: 00000077c79b2ae0 X15: b4000077202cc480
> > X14: 0000000000000000 X13: 000000761c517a70 X12: ffffff80ffffffd0
> > X11: 000000761c517a40 X10: 0000000000000001 X9: 0000000000000000
> > X8: 0000000000000040 X7: 7f7f7f7f7f7f7f7f X6: 0000000000000010
> > X5: 000000761c517c0c X4: ffffffffffffffff X3: ffffffffffffffff
> > X2: 0000000000000001 X1: 000000761c517c11 X0: 00000000000003e3
> > ORIG_X0: 00000000000003e3 SYSCALLNO: 40 PSTATE: 00001000
> >
> ditto.
>
> Maybe a short summary after the output paste, telling us which info
> should reviewers pay attention to.
>
> Thanks,
> Tao Liu
>
> > Signed-off-by: Guanyou.Chen <chenguanyou(a)xiaomi.com>
> > ---
> > netdump.c | 6 +++---
> > 1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/netdump.c b/netdump.c
> > index b4e2a5c..8ea5159 100644
> > --- a/netdump.c
> > +++ b/netdump.c
> > @@ -75,7 +75,7 @@ void
> > map_cpus_to_prstatus(void)
> > {
> > void **nt_ptr;
> > - int online, i, j, nrcpus;
> > + int online, i, nrcpus;
> > size_t size;
> >
> > if (pc->flags2 & QEMU_MEM_DUMP_ELF) /* notes exist for all cpus */
> > @@ -100,9 +100,9 @@ map_cpus_to_prstatus(void)
> > */
> > nrcpus = (kt->kernel_NR_CPUS ? kt->kernel_NR_CPUS : NR_CPUS);
> >
> > - for (i = 0, j = 0; i < nrcpus; i++) {
> > + for (i = 0; i < nrcpus; i++) {
> > if (in_cpu_map(ONLINE_MAP, i) &&
machdep->is_cpu_prstatus_valid(i)) {
> > - nd->nt_prstatus_percpu[i] = nt_ptr[j++];
> > + nd->nt_prstatus_percpu[i] = nt_ptr[i];
I re-think about the code change, it seems to me the variable 'j'
doesn't make any sense here.
nt_ptr = (void **)GETBUF(size);
BCOPY(nd->nt_prstatus_percpu, nt_ptr, size);
BZERO(nd->nt_prstatus_percpu, size);
The data of nt_ptr buffer is copied from nd->nt_prstatus_percpu
buffer, then nd->nt_prstatus_percpu buffer is cleared. If cpu 0 1 2 4
is online and 3 is offline, nd->nt_prstatus_percpu[4] = nt_ptr[3];
that is, copy the original cpu3's info into current cpu4's buf. This
cpu shift operation doesn't make any sense to me,
nd->nt_prstatus_percpu[4] = nt_ptr[4] should make sense in this case.
By searching the code, I didn't see nd->nt_prstatus_percpu[cpu] except
for any cpu shifts.
However I cannot find out the git history of why variable 'j' was
introduced. So this patch looks correct to me, though I haven't run
the overall regression test against it. What's your opinion on the
variable 'j'? In addition, I can do a regression test if needed.
@Lianbo Jiang
There is another similar code in diskdump.c, which seems need to be
updated as well.
Thanks,
Tao Liu
> > nd->num_prstatus_notes =
> > MAX(nd->num_prstatus_notes, i+1);
> > }
> > --
> > 2.34.1
> >
> > Guanyou.
> > Thanks.
>