Re: nvme crash - Re: linux-next: Tree for Aug 13

[John Garry <john.garry@huawei.com>]

On 14/08/2020 13:08, Christoph Hellwig wrote:
>>>> [148.455065]__sg_alloc_table_from_pages+0xec/0x238
>>>> [148.459931]sg_alloc_table_from_pages+0x18/0x28
>>>> [148.464541]iommu_dma_alloc+0x474/0x678
>>>> [148.468455]dma_alloc_attrs+0xd8/0xf0
>>>> [148.472193]nvme_alloc_queue+0x114/0x160 [nvme]
>>>> [148.476798]nvme_reset_work+0xb34/0x14b4 [nvme]
>>>> [148.481407]process_one_work+0x1e8/0x360
>>>> [148.485405]worker_thread+0x44/0x478
>>>> [148.489055]kthread+0x150/0x158
>>>> [148.492273]ret_from_fork+0x10/0x34
>>>> [148.495838] Code: f94002c3 6b01017f 540007c2 11000486 (f8645aa5)
>>>> [148.501921] ---[ end trace 89bb2b72d59bf925 ]---
>>>>
>>>> Anything to worry about? I guess not since we're in the merge window, but
>>>> mentioning just in case ...
>> I bisected, and this patch looks to fix it (note the comments below the
>> '---'):
>>
>>  From 263891a760edc24b901085bf6e5fe2480808f86d Mon Sep 17 00:00:00 2001
>> From: John Garry<john.garry@huawei.com>
>> Date: Fri, 14 Aug 2020 12:45:18 +0100
>> Subject: [PATCH] nvme-pci: Use u32 for nvme_dev.q_depth
>>
>> Recently nvme_dev.q_depth was changed from int to u16 type.
>>
>> This falls over for the queue depth calculation in nvme_pci_enable(),
>> where NVME_CAP_MQES(dev->ctrl.cap) + 1 may overflow, as NVME_CAP_MQES()
>> gives a 16b number also. That happens for me, and this is the result:
> Oh, interesting.  Please also switch the module option parsing to
> use kstrtou32 and param_set_uint and send this as a formal patch.
> 

I'm doing it now.

BTW, as for the DMA/sg scatterlist code, it so happens in this case that 
we try the dma alloc for size=0 in nvme_alloc_queue() - I know an 
allocation for size=0 makes no sense, but couldn't we bit a bit more robust?

Cheers,
John



_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu