> 2023年12月21日 14:19，Yu Kuai <yukuai1@xxxxxxxxxxxxxxx> 写道： > > Hi, > > 在 2023/12/21 12:49, Coly Li 写道: >> This reverts commit dd6291c506490c195620b394dc96763675e7e5f4. >> With this patch, a kernel oops triggered when creating a md device, >> [ 311.224353][ T3545] BUG: unable to handle page fault for address: 000003e800030d40 >> [ 311.314951][ T3545] #PF: supervisor read access in kernel mode >> [ 311.384748][ T3545] #PF: error_code(0x0000) - not-present page >> [ 311.454538][ T3545] PGD 12be1c067 P4D 0 >> [ 311.501451][ T3545] Oops: 0000 [#1] PREEMPT SMP NOPTI >> [ 311.561888][ T3545] CPU: 19 PID: 3545 Comm: modprobe [snipped...] >> [ 311.869958][ T3545] RIP: 0010:string+0x48/0xe0 >> [ 311.923116][ T3545] Code: 3b 45 89 d1 45 31 c0 49 01 f9 66 45 85 d2 75 1a eb 1f 48 39 f7 73 02 88 07 48 83 c7 01 41 83 c0 01 48 83 c2 01 4c 39 cf 74 07 <0f> b6 02 84 c0 75 e1 48 89 f2 44 89 c6 e9 c6 e3 ff ff 48 c7 c0 3d >> [ 312.156194][ T3545] RSP: 0018:ffa000000b877a70 EFLAGS: 00010086 >> [ 312.227025][ T3545] RAX: 000003e80002fd40 RBX: ffa000000b877b86 RCX: ffff0a00ffffff04 >> [ 312.320737][ T3545] RDX: 000003e800030d40 RSI: ffa000000b877b68 RDI: ffa000000b877b86 >> [ 312.414449][ T3545] RBP: ffa000000b877b48 R08: 0000000000000000 R09: ffa000010b877b85 >> [ 312.508160][ T3545] R10: ffffffffffffffff R11: 0000000000000040 R12: ffa000000b877b68 >> [ 312.601873][ T3545] R13: ffffffff99c221fa R14: 0000000000000008 R15: ffffffff99c221fa >> [ 312.695583][ T3545] FS: 00007fea7a856740(0000) GS:ff11000fffd80000(0000) knlGS:0000000000000000 >> [ 312.800733][ T3545] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 >> [ 312.877805][ T3545] CR2: 000003e800030d40 CR3: 0000000123790001 CR4: 0000000000771ee0 >> [ 312.971518][ T3545] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 >> [ 313.065229][ T3545] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 >> [ 313.158940][ T3545] PKRU: 55555554 >> [ 313.199610][ T3545] Call Trace: >> [ 313.237162][ T3545] <TASK> >> [ 313.270554][ T3545] ? __die+0x23/0x70 >> [ 313.315391][ T3545] ? page_fault_oops+0x14d/0x490 >> [ 313.372701][ T3545] ? update_load_avg+0x7e/0x7d0 >> [ 313.428972][ T3545] ? exc_page_fault+0x71/0x160 >> [ 313.484203][ T3545] ? asm_exc_page_fault+0x26/0x30 >> [ 313.542555][ T3545] ? string+0x48/0xe0 >> [ 313.588426][ T3545] vsnprintf+0x2d5/0x5a0 >> [ 313.637417][ T3545] vprintk_store+0x15e/0x4b0 >> [ 313.690567][ T3545] ? schedule_timeout+0x147/0x160 >> [ 313.748918][ T3545] ? wait_for_completion_killable+0x1a6/0x1d0 >> [ 313.819750][ T3545] vprintk_emit+0xc9/0x230 >> [ 313.870823][ T3545] _printk+0x5c/0x80 >> [ 313.915657][ T3545] sysctl_err+0x6a/0x90 >> [ 313.963610][ T3545] ? __kmalloc+0x4d/0x150 >> [ 314.013639][ T3545] __register_sysctl_table+0x144/0x7d0 >> [ 314.077192][ T3545] ? kmalloc_trace+0x2a/0xa0 >> [ 314.130341][ T3545] md_init+0xd2/0xff0 [snipped...] >> [ 314.228226][ T3545] ? __pfx_md_init+0x10/0x10 [snipped...] >> [ 314.333383][ T3545] do_one_initcall+0x47/0x220 >> [ 314.387576][ T3545] ? kmalloc_trace+0x2a/0xa0 >> [ 314.440726][ T3545] do_init_module+0x60/0x240 >> [ 314.493878][ T3545] __do_sys_finit_module+0xac/0x120 >> [ 314.554308][ T3545] do_syscall_64+0x5d/0x90 >> [ 314.605380][ T3545] ? ksys_lseek+0x66/0xb0 >> [ 314.655411][ T3545] ? syscall_exit_to_user_mode+0x2b/0x40 >> [ 314.721042][ T3545] ? do_syscall_64+0x6c/0x90 >> [ 314.774194][ T3545] ? exit_to_user_mode_prepare+0x142/0x1f0 >> [ 314.841906][ T3545] ? syscall_exit_to_user_mode+0x2b/0x40 >> [ 314.907535][ T3545] ? do_syscall_64+0x6c/0x90 >> [ 314.960685][ T3545] ? exc_page_fault+0x71/0x160 >> [ 315.015917][ T3545] entry_SYSCALL_64_after_hwframe+0x72/0xdc >> [ 315.084667][ T3545] RIP: 0033:0x7fea79f161bd >> The last NULL element in raid_table[] is necessary, after reverting this > Hi Kuai, > Based on commit message, avoid last NULL element is exactly what [1] > did, if this is not true, can you explame more how sysctl_err() is > called from md_init()? I can't find this by code review, and I think > maybe it's better to fix this in sysctl error path. > I feel you are right! The test was based on a backport of stable tree, and the register_sysctl() related code was not included. After look at the changes of sysctl, I feel the oops should go away after taking the sysctl changes. Thanks, and please ignore the noise. Coly Li [snipped]
