On Tue 14-06-22 00:20:23, Lecopzer Chen wrote:
> Sorry to jump in,
>
> This patch merged in 5.19-rc2,
> and I find that this commit breaks my QEMU bootup(can't boot to shell) on arm64.
>
> After revert this commit, everything goes back to normal.
>
> Please kindly help with it if you have any idea about it.
>
> for the log snippet:
>
> [ 3.755996][ T32] Unable to handle kernel paging request at virtual address ffffffc0f621e000
> [ 3.756657][ T32] Mem abort info:
> [ 3.756854][ T32] ESR = 0x0000000096000005
> [ 3.757224][ T32] EC = 0x25: DABT (current EL), IL = 32 bits
> [ 3.757501][ T32] SET = 0, FnV = 0
> [ 3.757685][ T32] EA = 0, S1PTW = 0
> [ 3.757837][ T32] FSC = 0x05: level 1 translation fault
> [ 3.758008][ T32] Data abort info:
> [ 3.758162][ T32] ISV = 0, ISS = 0x00000005
> [ 3.758393][ T32] CM = 0, WnR = 0
> [ 3.758643][ T32] swapper pgtable: 4k pages, 39-bit VAs, pgdp=0000000041619000
> [ 3.758813][ T32] [ffffffc0f621e000] pgd=0000000000000000, p4d=0000000000000000, pud=0000000000000000
> [ 3.765434][ T32] Internal error: Oops: 96000005 [#1] PREEMPT SMP
> [ 3.766394][ T32] Modules linked in:
> [ 3.766831][ T32] CPU: 1 PID: 32 Comm: kdevtmpfs Not tainted 5.19.0-rc2 #55
> [ 3.767292][ T32] Hardware name: linux,dummy-virt (DT)
> [ 3.767645][ T32] pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> [ 3.767827][ T32] pc : percpu_ref_get_many+0xf4/0x248
> [ 3.768344][ T32] lr : percpu_ref_get_many+0x70/0x248
> [ 3.768457][ T32] sp : ffffffc00a5cbb70
> [ 3.768534][ T32] pmr_save: 000000e0
> [ 3.768605][ T32] x29: ffffffc00a5cbb70 x28: 0000000000000000 x27: ffffffc00a3ca928
> [ 3.776825][ T32] x26: ffffffc00980c5d8 x25: ffffffc00980b688 x24: 0000000000000002
> [ 3.777005][ T32] x23: ffffffc00980b8a0 x22: ffffffc00980b698 x21: 0000000000000000
> [ 3.813382][ T32] x20: 0000000000000000 x19: ffffffc00a3cada0 x18: ffffffc00995ca70
> [ 3.814159][ T32] x17: ffffffc009bbeab0 x16: ffffffc0099fca70 x15: ffffffc0099fca70
> [ 3.814699][ T32] x14: ffffff80c03b4a40 x13: 00000000ffffffff x12: ffffff80c03b4a40
> [ 3.815745][ T32] x11: 0000000000000001 x10: 0000000000000001 x9 : ffffffc0f621e000
> [ 3.816383][ T32] x8 : ffffff80c03b4a40 x7 : ffffffc0084d192c x6 : 0000000000000000
> [ 3.817208][ T32] x5 : 0000000000000000 x4 : 0000000000000000 x3 : 0000000000000000
> [ 3.817618][ T32] x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffffffc0096eb800
> [ 3.818814][ T32] Call trace:
> [ 3.819360][ T32] percpu_ref_get_many+0xf4/0x248
> [ 3.820357][ T32] locked_inode_to_wb_and_lock_list+0x118/0x39c
> [ 3.821007][ T32] __mark_inode_dirty+0x2f0/0x4cc
> [ 3.821418][ T32] simple_setattr+0xbc/0xd4
> [ 3.821808][ T32] notify_change+0x590/0x800
> [ 3.822230][ T32] devtmpfs_work_loop+0x464/0x8c8
> [ 3.823063][ T32] sock_inuse_exit_net+0x0/0x2c
> [ 3.823520][ T32] kthread+0x22c/0x250
> [ 3.824159][ T32] ret_from_fork+0x10/0x20
> [ 3.824985][ T32] Code: 11000529 b9000909 d538d089 8b140129 (c85f7d2c)
> [ 3.826820][ T32] ---[ end trace 0000000000000000 ]---
> [ 3.829725][ T32] note: kdevtmpfs[32] exited with preempt_count 2
> [ 3.840608][ T32] ------------[ cut here ]------------
>
> for the assembly:
> ; asm(ALTERNATIVE("mrs %0, tpidr_el1",
> ffffffc008095048: 89 d0 38 d5 mrs x9, TPIDR_EL1
> ; this_cpu_add(*percpu_count, nr);
> ffffffc00809504c: 29 01 14 8b add x9, x9, x20
> ; PERCPU_OP(add, add, stadd)
> ***-->ffffffc008095050: 2c 7d 5f c8 ldxr x12, [x9]
>
> locked_inode_to_wb_and_lock_list
> -> wb_get(wb)
> -> percpu_ref_get(&wb->refcnt);
> -> percpu_ref_get_many
> ->
> if (__ref_is_percpu(ref, &percpu_count))
> ***--> this_cpu_add(*percpu_count, nr);
Thanks for report! I've checked now and the problem is that
locked_inode_to_wb_and_lock_list() now gets called for inodes in the
kdevtmpfs and it gets called very early (during driver_init()) which is
before noop_backing_dev_info gets initialized. I'll fix this...
Honza
--
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR
