Re: [bug] __blk_mq_run_hw_queue suspicious rcu usage

From: "Lendacky, Thomas" <Thomas.Lendacky@amd.com>
To: David Rientjes <rientjes@google.com>,
	Christoph Hellwig <hch@lst.de>, Keith Busch <kbusch@kernel.org>,
	Jens Axboe <axboe@kernel.dk>
Cc: "Singh, Brijesh" <brijesh.singh@amd.com>,
	"x86@kernel.org" <x86@kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Ming Lei <ming.lei@redhat.com>,
	"iommu@lists.linux-foundation.org"
	<iommu@lists.linux-foundation.org>,
	Peter Gonda <pgonda@google.com>, Jianxiong Gao <jxgao@google.com>
Subject: Re: [bug] __blk_mq_run_hw_queue suspicious rcu usage
Date: Tue, 17 Sep 2019 18:41:02 +0000	[thread overview]
Message-ID: <1d74607e-37f7-56ca-aba3-5a3bd7a68561@amd.com> (raw)
In-Reply-To: <alpine.DEB.2.21.1909171121300.151243@chino.kir.corp.google.com>

On 9/17/19 1:23 PM, David Rientjes wrote:
> On Mon, 16 Sep 2019, David Rientjes wrote:
> 
>> Brijesh and Tom, we currently hit this any time we boot an SEV enabled 
>> Ubuntu 18.04 guest; I assume that guest kernels, especially those of such 
>> major distributions, are expected to work with warnings and BUGs when 
>> certain drivers are enabled.
>>
>> If the vmap purge lock is to remain a mutex (any other reason that 
>> unmapping aliases can block?) then it appears that allocating a dmapool 
>> is the only alternative.  Is this something that you'll be addressing 
>> generically or do we need to get buy-in from the maintainers of this 
>> specific driver?
>>
> 
> We've found that the following applied on top of 5.2.14 suppresses the 
> warnings.
> 
> Christoph, Keith, Jens, is this something that we could do for the nvme 
> driver?  I'll happily propose it formally if it would be acceptable.
> 
> Thanks!
> 
> 
> diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
> --- a/drivers/nvme/host/pci.c
> +++ b/drivers/nvme/host/pci.c
> @@ -1613,7 +1613,8 @@ static int nvme_alloc_admin_tags(struct nvme_dev *dev)
>  		dev->admin_tagset.timeout = ADMIN_TIMEOUT;
>  		dev->admin_tagset.numa_node = dev_to_node(dev->dev);
>  		dev->admin_tagset.cmd_size = sizeof(struct nvme_iod);
> -		dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED;
> +		dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED |
> +					  BLK_MQ_F_BLOCKING;

I think you want to only set the BLK_MQ_F_BLOCKING if the DMA is required
to be unencrypted. Unfortunately, force_dma_unencrypted() can't be called
from a module. Is there a DMA API that could be called to get that info?

Thanks,
Tom

>  		dev->admin_tagset.driver_data = dev;
>  
>  		if (blk_mq_alloc_tag_set(&dev->admin_tagset))
> @@ -2262,7 +2263,8 @@ static int nvme_dev_add(struct nvme_dev *dev)
>  		dev->tagset.queue_depth =
>  				min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
>  		dev->tagset.cmd_size = sizeof(struct nvme_iod);
> -		dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE;
> +		dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE |
> +				    BLK_MQ_F_BLOCKING;
>  		dev->tagset.driver_data = dev;
>  
>  		ret = blk_mq_alloc_tag_set(&dev->tagset);
> 
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu