Re: [dpdk-dev] [External] RE: [PATCH v2] vhost: avoid iotlb mempool allocation while IOMMU disabled

Hi,

-----Original Message-----
From: Wan Junjie <redacted>
Sent: Thursday, May 13, 2021 6:52 PM
To: Xia, Chenbo <redacted>
Cc: David Marchand <redacted>; Maxime Coquelin
[off-list ref]; dev@dpdk.org; Peng He
[off-list ref]; Zhihong Wang [off-list ref];
Thomas Monjalon [off-list ref]
Subject: Re: [External] RE: [dpdk-dev] [PATCH v2] vhost: avoid iotlb mempool
allocation while IOMMU disabled

On Thu, May 13, 2021 at 6:24 PM Xia, Chenbo [off-list ref] wrote:

quoted

quoted

-----Original Message-----
From: David Marchand <redacted>
Sent: Thursday, May 13, 2021 4:39 PM
To: Xia, Chenbo <redacted>; Maxime Coquelin
[off-list ref]
Cc: Wan Junjie <redacted>; dev@dpdk.org; Peng He
[off-list ref]; Zhihong Wang [off-list ref]
Subject: Re: [dpdk-dev] [PATCH v2] vhost: avoid iotlb mempool allocation

while

quoted

quoted

IOMMU disabled

On Wed, Apr 7, 2021 at 9:10 AM Xia, Chenbo [off-list ref] wrote:

quoted

quoted

-----Original Message-----
From: Wan Junjie <redacted>
Sent: Tuesday, February 2, 2021 4:15 PM
To: Xia, Chenbo <redacted>; Maxime Coquelin
[off-list ref]
Cc: dev@dpdk.org; Wan Junjie <redacted>; Peng He
[off-list ref]; Zhihong Wang

[off-list ref]

quoted

quoted

quoted

quoted

Subject: [PATCH v2] vhost: avoid iotlb mempool allocation while IOMMU

disabled

quoted

quoted

If vhost device's IOMMU feature is disabled, iotlb mempool allocation
is unnecessary.

Reported-by: Peng He <redacted>
Signed-off-by: Wan Junjie <redacted>
Reviewed-by: Zhihong Wang <redacted>
---
2.11.0

Patch applied to next-virtio/main with conflict resolved, Thanks

This patch triggered a regression reported by Red Hat QE.
https://bugs.dpdk.org/show_bug.cgi?id=703

quoted

quoted

Reverting the patch is enough to fix the regression.

I'll let you guys decide on the next step for 21.05.


--
David Marchand

I cannot reproduce this issue on my machine.

@Wan Junjie Could you reproduce this? I prefer to revert the patch if things
are not very clear as it is only a minor optimization.

Thanks,
Chenbo

Hi Chenbo,

I am not available for this now. Please revert it.
Will dig it later.

Thanks,
Junjie

I know the root cause, it is because when multi-queue is on, qemu only enable the
first queue pair and does not 'VHOST_USER_SET_VRING_NUM' for other queue pairs, which leads
to disabled queue pairs' iotlb_pool not init-ed. Then comes the iotlb message that operate
on all queue pairs iotlb pool.

A fix is coming.

Thanks,
Chenbo