Documentation/fault-injection/nvme-fault-injection.rst

   1 NVMe Fault Injection
   2 ====================
   3 Linux's fault injection framework provides a systematic way to support
   4 error injection via debugfs in the /sys/kernel/debug directory. When
   5 enabled, the default NVME_SC_INVALID_OPCODE with no retry will be
   6 injected into the nvme_end_request. Users can change the default status
   7 code and no retry flag via the debugfs. The list of Generic Command
   8 Status can be found in include/linux/nvme.h
   9
  10 Following examples show how to inject an error into the nvme.
  11
  12 First, enable CONFIG_FAULT_INJECTION_DEBUG_FS kernel config,
  13 recompile the kernel. After booting up the kernel, do the
  14 following.
  15
  16 Example 1: Inject default status code with no retry
  17 ---------------------------------------------------
  18
  19 ::
  20
  21   mount /dev/nvme0n1 /mnt
  22   echo 1 > /sys/kernel/debug/nvme0n1/fault_inject/times
  23   echo 100 > /sys/kernel/debug/nvme0n1/fault_inject/probability
  24   cp a.file /mnt
  25
  26 Expected Result::
  27
  28   cp: cannot stat ‘/mnt/a.file’: Input/output error
  29
  30 Message from dmesg::
  31
  32   FAULT_INJECTION: forcing a failure.
  33   name fault_inject, interval 1, probability 100, space 0, times 1
  34   CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.15.0-rc8+ #2
  35   Hardware name: innotek GmbH VirtualBox/VirtualBox,
  36   BIOS VirtualBox 12/01/2006
  37   Call Trace:
  38     <IRQ>
  39     dump_stack+0x5c/0x7d
  40     should_fail+0x148/0x170
  41     nvme_should_fail+0x2f/0x50 [nvme_core]
  42     nvme_process_cq+0xe7/0x1d0 [nvme]
  43     nvme_irq+0x1e/0x40 [nvme]
  44     __handle_irq_event_percpu+0x3a/0x190
  45     handle_irq_event_percpu+0x30/0x70
  46     handle_irq_event+0x36/0x60
  47     handle_fasteoi_irq+0x78/0x120
  48     handle_irq+0xa7/0x130
  49     ? tick_irq_enter+0xa8/0xc0
  50     do_IRQ+0x43/0xc0
  51     common_interrupt+0xa2/0xa2
  52     </IRQ>
  53   RIP: 0010:native_safe_halt+0x2/0x10
  54   RSP: 0018:ffffffff82003e90 EFLAGS: 00000246 ORIG_RAX: ffffffffffffffdd
  55   RAX: ffffffff817a10c0 RBX: ffffffff82012480 RCX: 0000000000000000
  56   RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000
  57   RBP: 0000000000000000 R08: 000000008e38ce64 R09: 0000000000000000
  58   R10: 0000000000000000 R11: 0000000000000000 R12: ffffffff82012480
  59   R13: ffffffff82012480 R14: 0000000000000000 R15: 0000000000000000
  60     ? __sched_text_end+0x4/0x4
  61     default_idle+0x18/0xf0
  62     do_idle+0x150/0x1d0
  63     cpu_startup_entry+0x6f/0x80
  64     start_kernel+0x4c4/0x4e4
  65     ? set_init_arg+0x55/0x55
  66     secondary_startup_64+0xa5/0xb0
  67     print_req_error: I/O error, dev nvme0n1, sector 9240
  68   EXT4-fs error (device nvme0n1): ext4_find_entry:1436:
  69   inode #2: comm cp: reading directory lblock 0
  70
  71 Example 2: Inject default status code with retry
  72 ------------------------------------------------
  73
  74 ::
  75
  76   mount /dev/nvme0n1 /mnt
  77   echo 1 > /sys/kernel/debug/nvme0n1/fault_inject/times
  78   echo 100 > /sys/kernel/debug/nvme0n1/fault_inject/probability
  79   echo 1 > /sys/kernel/debug/nvme0n1/fault_inject/status
  80   echo 0 > /sys/kernel/debug/nvme0n1/fault_inject/dont_retry
  81
  82   cp a.file /mnt
  83
  84 Expected Result::
  85
  86   command success without error
  87
  88 Message from dmesg::
  89
  90   FAULT_INJECTION: forcing a failure.
  91   name fault_inject, interval 1, probability 100, space 0, times 1
  92   CPU: 1 PID: 0 Comm: swapper/1 Not tainted 4.15.0-rc8+ #4
  93   Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
  94   Call Trace:
  95     <IRQ>
  96     dump_stack+0x5c/0x7d
  97     should_fail+0x148/0x170
  98     nvme_should_fail+0x30/0x60 [nvme_core]
  99     nvme_loop_queue_response+0x84/0x110 [nvme_loop]
 100     nvmet_req_complete+0x11/0x40 [nvmet]
 101     nvmet_bio_done+0x28/0x40 [nvmet]
 102     blk_update_request+0xb0/0x310
 103     blk_mq_end_request+0x18/0x60
 104     flush_smp_call_function_queue+0x3d/0xf0
 105     smp_call_function_single_interrupt+0x2c/0xc0
 106     call_function_single_interrupt+0xa2/0xb0
 107     </IRQ>
 108   RIP: 0010:native_safe_halt+0x2/0x10
 109   RSP: 0018:ffffc9000068bec0 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff04
 110   RAX: ffffffff817a10c0 RBX: ffff88011a3c9680 RCX: 0000000000000000
 111   RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000
 112   RBP: 0000000000000001 R08: 000000008e38c131 R09: 0000000000000000
 113   R10: 0000000000000000 R11: 0000000000000000 R12: ffff88011a3c9680
 114   R13: ffff88011a3c9680 R14: 0000000000000000 R15: 0000000000000000
 115     ? __sched_text_end+0x4/0x4
 116     default_idle+0x18/0xf0
 117     do_idle+0x150/0x1d0
 118     cpu_startup_entry+0x6f/0x80
 119     start_secondary+0x187/0x1e0
 120     secondary_startup_64+0xa5/0xb0
 121
 122 Example 3: Inject an error into the 10th admin command
 123 ------------------------------------------------------
 124
 125 ::
 126
 127   echo 100 > /sys/kernel/debug/nvme0/fault_inject/probability
 128   echo 10 > /sys/kernel/debug/nvme0/fault_inject/space
 129   echo 1 > /sys/kernel/debug/nvme0/fault_inject/times
 130   nvme reset /dev/nvme0
 131
 132 Expected Result::
 133
 134   After NVMe controller reset, the reinitialization may or may not succeed.
 135   It depends on which admin command is actually forced to fail.
 136
 137 Message from dmesg::
 138
 139   nvme nvme0: resetting controller
 140   FAULT_INJECTION: forcing a failure.
 141   name fault_inject, interval 1, probability 100, space 1, times 1
 142   CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.2.0-rc2+ #2
 143   Hardware name: MSI MS-7A45/B150M MORTAR ARCTIC (MS-7A45), BIOS 1.50 04/25/2017
 144   Call Trace:
 145    <IRQ>
 146    dump_stack+0x63/0x85
 147    should_fail+0x14a/0x170
 148    nvme_should_fail+0x38/0x80 [nvme_core]
 149    nvme_irq+0x129/0x280 [nvme]
 150    ? blk_mq_end_request+0xb3/0x120
 151    __handle_irq_event_percpu+0x84/0x1a0
 152    handle_irq_event_percpu+0x32/0x80
 153    handle_irq_event+0x3b/0x60
 154    handle_edge_irq+0x7f/0x1a0
 155    handle_irq+0x20/0x30
 156    do_IRQ+0x4e/0xe0
 157    common_interrupt+0xf/0xf
 158    </IRQ>
 159   RIP: 0010:cpuidle_enter_state+0xc5/0x460
 160   Code: ff e8 8f 5f 86 ff 80 7d c7 00 74 17 9c 58 0f 1f 44 00 00 f6 c4 02 0f 85 69 03 00 00 31 ff e8 62 aa 8c ff fb 66 0f 1f 44 00 00 <45> 85 ed 0f 88 37 03 00 00 4c 8b 45 d0 4c 2b 45 b8 48 ba cf f7 53
 161   RSP: 0018:ffffffff88c03dd0 EFLAGS: 00000246 ORIG_RAX: ffffffffffffffdc
 162   RAX: ffff9dac25a2ac80 RBX: ffffffff88d53760 RCX: 000000000000001f
 163   RDX: 0000000000000000 RSI: 000000002d958403 RDI: 0000000000000000
 164   RBP: ffffffff88c03e18 R08: fffffff75e35ffb7 R09: 00000a49a56c0b48
 165   R10: ffffffff88c03da0 R11: 0000000000001b0c R12: ffff9dac25a34d00
 166   R13: 0000000000000006 R14: 0000000000000006 R15: ffffffff88d53760
 167    cpuidle_enter+0x2e/0x40
 168    call_cpuidle+0x23/0x40
 169    do_idle+0x201/0x280
 170    cpu_startup_entry+0x1d/0x20
 171    rest_init+0xaa/0xb0
 172    arch_call_rest_init+0xe/0x1b
 173    start_kernel+0x51c/0x53b
 174    x86_64_start_reservations+0x24/0x26
 175    x86_64_start_kernel+0x74/0x77
 176    secondary_startup_64+0xa4/0xb0
 177   nvme nvme0: Could not set queue count (16385)
 178   nvme nvme0: IO queues not created