[PATCH] nvme: limit max io queues as 1 in case of kdump kernel

[axboe@kernel.dk (Jens Axboe)]

On 12/6/18 6:57 PM, Ming Lei wrote:
> On Thu, Dec 06, 2018@06:39:44PM -0700, Jens Axboe wrote:
>> On 12/6/18 6:37 PM, Ming Lei wrote:
>>> On Thu, Dec 06, 2018@06:14:01PM -0700, Jens Axboe wrote:
>>>> On 12/6/18 6:13 PM, Ming Lei wrote:
>>>>> On Thu, Dec 06, 2018@08:31:00AM -0700, Jens Axboe wrote:
>>>>>> On 12/6/18 4:38 AM, Ming Lei wrote:
>>>>>>> NVMe PCI builds queue mapping before allocating tagset, in which
>>>>>>> set->nr_hw_queues is set as 1 in case of kdump kernel, so wrong
>>>>>>> queue mapping can be setup, and kernel panic[1] is observed during
>>>>>>> booting.
>>>>>>>
>>>>>>> This patch fixes the issue by setting max io queues as 1 under this
>>>>>>> situation.
>>>>>>>
>>>>>>> [1] kernel panic log
>>>>>>> [    4.438371] nvme nvme0: 16/0/0 default/read/poll queues
>>>>>>> [    4.443277] BUG: unable to handle kernel NULL pointer dereference at 0000000000000098
>>>>>>> [    4.444681] PGD 0 P4D 0
>>>>>>> [    4.445367] Oops: 0000 [#1] SMP NOPTI
>>>>>>> [    4.446342] CPU: 3 PID: 201 Comm: kworker/u33:10 Not tainted 4.20.0-rc5-00664-g5eb02f7ee1eb-dirty #459
>>>>>>> [    4.447630] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.10.2-2.fc27 04/01/2014
>>>>>>> [    4.448689] Workqueue: nvme-wq nvme_scan_work [nvme_core]
>>>>>>> [    4.449368] RIP: 0010:blk_mq_map_swqueue+0xfb/0x222
>>>>>>> [    4.450596] Code: 04 f5 20 28 ef 81 48 89 c6 39 55 30 76 93 89 d0 48 c1 e0 04 48 03 83 f8 05 00 00 48 8b 00 42 8b 3c 28 48 8b 43 58 48 8b 04 f8 <48> 8b b8 98 00 00 00 4c 0f a3 37 72 42 f0 4c 0f ab 37 66 8b b8 f6
>>>>>>> [    4.453132] RSP: 0018:ffffc900023b3cd8 EFLAGS: 00010286
>>>>>>> [    4.454061] RAX: 0000000000000000 RBX: ffff888174448000 RCX: 0000000000000001
>>>>>>> [    4.456480] RDX: 0000000000000001 RSI: ffffe8feffc506c0 RDI: 0000000000000001
>>>>>>> [    4.458750] RBP: ffff88810722d008 R08: ffff88817647a880 R09: 0000000000000002
>>>>>>> [    4.464580] R10: ffffc900023b3c10 R11: 0000000000000004 R12: ffff888174448538
>>>>>>> [    4.467803] R13: 0000000000000004 R14: 0000000000000001 R15: 0000000000000001
>>>>>>> [    4.469220] FS:  0000000000000000(0000) GS:ffff88817bac0000(0000) knlGS:0000000000000000
>>>>>>> [    4.471554] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>>>>>> [    4.472464] CR2: 0000000000000098 CR3: 0000000174e4e001 CR4: 0000000000760ee0
>>>>>>> [    4.474264] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>>>>>>> [    4.476007] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
>>>>>>> [    4.477061] PKRU: 55555554
>>>>>>> [    4.477464] Call Trace:
>>>>>>> [    4.478731]  blk_mq_init_allocated_queue+0x36a/0x3ad
>>>>>>> [    4.479595]  blk_mq_init_queue+0x32/0x4e
>>>>>>> [    4.480178]  nvme_validate_ns+0x98/0x623 [nvme_core]
>>>>>>> [    4.480963]  ? nvme_submit_sync_cmd+0x1b/0x20 [nvme_core]
>>>>>>> [    4.481685]  ? nvme_identify_ctrl.isra.8+0x70/0xa0 [nvme_core]
>>>>>>> [    4.482601]  nvme_scan_work+0x23a/0x29b [nvme_core]
>>>>>>> [    4.483269]  ? _raw_spin_unlock_irqrestore+0x25/0x38
>>>>>>> [    4.483930]  ? try_to_wake_up+0x38d/0x3b3
>>>>>>> [    4.484478]  ? process_one_work+0x179/0x2fc
>>>>>>> [    4.485118]  process_one_work+0x1d3/0x2fc
>>>>>>> [    4.485655]  ? rescuer_thread+0x2ae/0x2ae
>>>>>>> [    4.486196]  worker_thread+0x1e9/0x2be
>>>>>>> [    4.486841]  kthread+0x115/0x11d
>>>>>>> [    4.487294]  ? kthread_park+0x76/0x76
>>>>>>> [    4.487784]  ret_from_fork+0x3a/0x50
>>>>>>> [    4.488322] Modules linked in: nvme nvme_core qemu_fw_cfg virtio_scsi ip_tables
>>>>>>> [    4.489428] Dumping ftrace buffer:
>>>>>>> [    4.489939]    (ftrace buffer empty)
>>>>>>> [    4.490492] CR2: 0000000000000098
>>>>>>> [    4.491052] ---[ end trace 03cd268ad5a86ff7 ]---
>>>>>>>
>>>>>>> Cc: Jens Axboe <axboe at kernel.dk>
>>>>>>> Cc: David Milburn <dmilburn at redhat.com>
>>>>>>> Signed-off-by: Ming Lei <ming.lei at redhat.com>
>>>>>>> ---
>>>>>>>  drivers/nvme/host/pci.c | 9 +++++++++
>>>>>>>  1 file changed, 9 insertions(+)
>>>>>>>
>>>>>>> diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
>>>>>>> index 7732c4979a4e..86789921f463 100644
>>>>>>> --- a/drivers/nvme/host/pci.c
>>>>>>> +++ b/drivers/nvme/host/pci.c
>>>>>>> @@ -31,6 +31,7 @@
>>>>>>>  #include <linux/io-64-nonatomic-lo-hi.h>
>>>>>>>  #include <linux/sed-opal.h>
>>>>>>>  #include <linux/pci-p2pdma.h>
>>>>>>> +#include <linux/crash_dump.h>
>>>>>>>  
>>>>>>>  #include "nvme.h"
>>>>>>>  
>>>>>>> @@ -253,6 +254,14 @@ static inline void _nvme_check_size(void)
>>>>>>>  
>>>>>>>  static unsigned int max_io_queues(void)
>>>>>>>  {
>>>>>>> +	/*
>>>>>>> +	 * blk-mq may set set->nr_hw_queues as 1 during allocating tagset
>>>>>>> +	 * in case of kdump kernel. However, queue mapping can be setup
>>>>>>> +	 * before allocating tagset, so limit the max io queues as 1 if
>>>>>>> +	 * running from kdump kernel.
>>>>>>> +	 */
>>>>>>> +	if (is_kdump_kernel())
>>>>>>> +		return 1;
>>>>>>>  	return num_possible_cpus() + write_queues + poll_queues;
>>>>>>>  }
>>>>>>
>>>>>> Hmm, can we do this in blk-mq instead? Seems kind of shady to have to
>>>>>> be aware of this in the driver. Something like the below? Totally
>>>>>> untested.
>>>>>>
>>>>>> diff --git a/block/blk-mq.c b/block/blk-mq.c
>>>>>> index 900550594651..38beb0de5088 100644
>>>>>> --- a/block/blk-mq.c
>>>>>> +++ b/block/blk-mq.c
>>>>>> @@ -3028,9 +3028,9 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
>>>>>>  	 * memory constrained environment. Limit us to 1 queue and
>>>>>>  	 * 64 tags to prevent using too much memory.
>>>>>>  	 */
>>>>>> -	if (is_kdump_kernel()) {
>>>>>> -		set->nr_hw_queues = 1;
>>>>>> +	if (is_kdump_kernel() && set->nr_hw_queues > 1) {
>>>>>>  		set->queue_depth = min(64U, set->queue_depth);
>>>>>> +		blk_mq_update_nr_hw_queues(set, 1);
>>>>>
>>>>> It can't make a difference.
>>>>>
>>>>> The queue mapping is built via pci_alloc_irq_vectors_affinity() in nvme
>>>>> reset work fn. However, .map_queues(nvme_pci_map_queues()) won't trigger
>>>>> nvme reset, so blk_mq_update_nr_hw_queues() still uses the old built
>>>>> mapping.
>>>>>
>>>>> That might be an generic issue with blk_mq_update_nr_hw_queues(), but
>>>>> I guess it might not trigger a kernel panic when it is called via nvme_dev_add().
>>>>
>>>> See the later one, it worked for me. This one was untested, as mentioned.
>>>
>>> Looks there is still the panic in my test, as I mentioned, the old mapping
>>> can't be updated by the patch, and all other hw queues(index > 0) can be
>>> retrieved via CPU index.
>>
>> Which one did you try? Worked for me. Not saying there can't be an issue...
> 
> The later one with single queue/single queue map, follows the patch I
> applied exactly:
> 
> [mingl at hp-dl380g10-01 kernel]$ git diff
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index 900550594651..2f0725f3f59c 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -2810,6 +2810,9 @@ static void blk_mq_realloc_hw_ctxs(struct blk_mq_tag_set *set,
>   */
>  static unsigned int nr_hw_queues(struct blk_mq_tag_set *set)
>  {
> +      if (/*is_kdump_kernel()*/1)
> +            return 1;
> +
>         if (set->nr_maps == 1)
>                 return nr_cpu_ids;
>  
> @@ -3028,9 +3031,11 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
>          * memory constrained environment. Limit us to 1 queue and
>          * 64 tags to prevent using too much memory.
>          */
> -       if (is_kdump_kernel()) {
> +       if (/*is_kdump_kernel()*/1) {
>                 set->nr_hw_queues = 1;
>                 set->queue_depth = min(64U, set->queue_depth);
> +               //blk_mq_update_nr_hw_queues(set, 1);
> +               set->nr_maps = 1;
>         }
> 
> The former patch doesn't work too.
> 
> The issue is that we only allocate one hw_ctx instance in case of kdump,
> however, we may map to other hctx by the old mapping, then kernel panic
> is triggered on cpumask_test_cpu(i, hctx->cpumask)) in
> blk_mq_map_swqueue().
> 
>> If we are going to modify the driver to fix it, it should be by some
>> exported function. It seems kind of nasty to have implied kdump specific
>> knowledge in the driver.
> 
> I agree it is just one workaround, and I actually mentioned that in Red Hat
> internal BZ.
> 
> Another approach may be to use blk_mq_map_queues() to re-build the
> mapping by passing .nr_queues as 1.

Yeah, I think that is a better approach.

-- 
Jens Axboe