-----Original Message-----
From: Michael Kelley (LINUX) <redacted>
Sent: Thursday, December 9, 2021 2:54 PM
To: Haiyang Zhang <haiyangz@microsoft.com>; Tianyu Lan <redacted>; KY
Srinivasan [off-list ref]; Stephen Hemminger [off-list ref];
wei.liu@kernel.org; Dexuan Cui [off-list ref]; tglx@linutronix.de; mingo@redhat.com;
bp@alien8.de; dave.hansen@linux.intel.com; x86@kernel.org; hpa@zytor.com;
davem@davemloft.net; kuba@kernel.org; jejb@linux.ibm.com; martin.petersen@oracle.com;
arnd@arndb.de; hch@infradead.org; m.szyprowski@samsung.com; robin.murphy@arm.com; Tianyu
Lan [off-list ref]; thomas.lendacky@amd.com
Cc: iommu@lists.linux-foundation.org; linux-arch@vger.kernel.org; linux-
hyperv@vger.kernel.org; linux-kernel@vger.kernel.org; linux-scsi@vger.kernel.org;
netdev@vger.kernel.org; vkuznets [off-list ref]; brijesh.singh@amd.com;
konrad.wilk@oracle.com; hch@lst.de; joro@8bytes.org; parri.andrea@gmail.com;
dave.hansen@intel.com
Subject: RE: [PATCH V6 5/5] net: netvsc: Add Isolation VM support for netvsc driver

From: Haiyang Zhang <haiyangz@microsoft.com> Sent: Wednesday, December 8, 2021 12:14 PM

quoted

quoted

From: Tianyu Lan <redacted>
Sent: Tuesday, December 7, 2021 2:56 AM

[snip]

quoted

quoted

 static inline int netvsc_send_pkt(
 	struct hv_device *device,
 	struct hv_netvsc_packet *packet,

@@ -986,14 +1105,24 @@ static inline int netvsc_send_pkt(

 	trace_nvsp_send_pkt(ndev, out_channel, rpkt);

+	packet->dma_range = NULL;
 	if (packet->page_buf_cnt) {
 		if (packet->cp_partial)
 			pb += packet->rmsg_pgcnt;

+		ret = netvsc_dma_map(ndev_ctx->device_ctx, packet, pb);
+		if (ret) {
+			ret = -EAGAIN;
+			goto exit;
+		}

Returning EAGAIN will let the upper network layer busy retry,
which may make things worse.
I suggest to return ENOSPC here like another place in this
function, which will just drop the packet, and let the network
protocol/app layer decide how to recover.

Thanks,
- Haiyang

I made the original suggestion to return -EAGAIN here.   A
DMA mapping failure should occur only if swiotlb bounce
buffer space is unavailable, which is a transient condition.
The existing code already stops the queue and returns
-EAGAIN when the ring buffer is full, which is also a transient
condition.  My sense is that the two conditions should be
handled the same way.  Or is there a reason why a ring
buffer full condition should stop the queue and retry, while
a mapping failure should drop the packet?

The netvsc_dma_map() fails in these two places. The dma_map_single() 
is doing the maping with swiotlb bounce buffer, correct? And it will 
become successful after the previous packets are unmapped?

+	packet->dma_range = kcalloc(page_count,
+				    sizeof(*packet->dma_range),
+				    GFP_KERNEL);

+		dma = dma_map_single(&hv_dev->device, src, len,
+				     DMA_TO_DEVICE);

I recalled your previous suggestion now, and agree with you that 
we can treat it the same way (return -EAGAIN) in this case. And 
the existing code will stop the queue temporarily.

Thanks,
- Haiyang