[PATCH 5.4.y] hv_utils: drain the timesync packets on onchannelcallback

[PATCH 5.4.y] hv_utils: drain the timesync packets on onchannelcallback

[Dexuan Cui]

From: Vineeth Pillai <viremana@linux.microsoft.com>

commit b46b4a8a57c377b72a98c7930a9f6969d2d4784e

There could be instances where a system stall prevents the timesync
packets to be consumed. And this might lead to more than one packet
pending in the ring buffer. Current code empties one packet per callback
and it might be a stale one. So drain all the packets from ring buffer
on each callback.

Signed-off-by: Vineeth Pillai <viremana@linux.microsoft.com>
Reviewed-by: Michael Kelley <mikelley@microsoft.com>
Link: https://lore.kernel.org/r/20200821152849.99517-1-viremana@linux.microsoft.com
Signed-off-by: Wei Liu <wei.liu@kernel.org>

The old code in the upstream commit uses HV_HYP_PAGE_SIZE, but
the old code in 5.4.y sitll uses PAGE_SIZE. Fixed this manually for 5.4.y.
Note: 5.4.y already has the define HV_HYP_PAGE_SIZE, so the new code in
in the upstream commit works for 5.4.y.

If there are multiple messages in the host-to-guest ringbuffer of the TimeSync
device, 5.4.y only handles 1 message, and later the host puts new messages
into the ringbuffer without signaling the guest because the ringbuffer is not
empty, causing a "hung" ringbuffer. Backported the mainline fix for this issue.

Signed-off-by: Dexuan Cui <decui@microsoft.com>
---
 drivers/hv/hv_util.c | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)

diff --git a/drivers/hv/hv_util.c b/drivers/hv/hv_util.c
index 1671f6f9ea80..48f9b1fbcbda 100644
--- a/drivers/hv/hv_util.c
+++ b/drivers/hv/hv_util.c
@@ -283,10 +283,23 @@ static void timesync_onchannelcallback(void *context)
 	struct ictimesync_ref_data *refdata;
 	u8 *time_txf_buf = util_timesynch.recv_buffer;
 
-	vmbus_recvpacket(channel, time_txf_buf,
-			 PAGE_SIZE, &recvlen, &requestid);
+	/*
+	 * Drain the ring buffer and use the last packet to update
+	 * host_ts
+	 */
+	while (1) {
+		int ret = vmbus_recvpacket(channel, time_txf_buf,
+					   HV_HYP_PAGE_SIZE, &recvlen,
+					   &requestid);
+		if (ret) {
+			pr_warn_once("TimeSync IC pkt recv failed (Err: %d)\n",
+				     ret);
+			break;
+		}
+
+		if (!recvlen)
+			break;
 
-	if (recvlen > 0) {
 		icmsghdrp = (struct icmsg_hdr *)&time_txf_buf[
 				sizeof(struct vmbuspipe_hdr)];
 
-- 
2.25.1

Re: [PATCH 5.4.y] hv_utils: drain the timesync packets on onchannelcallback

[Greg KH]

On Mon, Jun 17, 2024 at 04:25:07PM -0700, Dexuan Cui wrote:
> From: Vineeth Pillai <viremana@linux.microsoft.com>
> 
> commit b46b4a8a57c377b72a98c7930a9f6969d2d4784e
> 
> There could be instances where a system stall prevents the timesync
> packets to be consumed. And this might lead to more than one packet
> pending in the ring buffer. Current code empties one packet per callback
> and it might be a stale one. So drain all the packets from ring buffer
> on each callback.
> 
> Signed-off-by: Vineeth Pillai <viremana@linux.microsoft.com>
> Reviewed-by: Michael Kelley <mikelley@microsoft.com>
> Link: https://lore.kernel.org/r/20200821152849.99517-1-viremana@linux.microsoft.com
> Signed-off-by: Wei Liu <wei.liu@kernel.org>
> 
> The old code in the upstream commit uses HV_HYP_PAGE_SIZE, but
> the old code in 5.4.y sitll uses PAGE_SIZE. Fixed this manually for 5.4.y.
> Note: 5.4.y already has the define HV_HYP_PAGE_SIZE, so the new code in
> in the upstream commit works for 5.4.y.
> 
> If there are multiple messages in the host-to-guest ringbuffer of the TimeSync
> device, 5.4.y only handles 1 message, and later the host puts new messages
> into the ringbuffer without signaling the guest because the ringbuffer is not
> empty, causing a "hung" ringbuffer. Backported the mainline fix for this issue.
> 
> Signed-off-by: Dexuan Cui <decui@microsoft.com>

Both now queued up, thanks.

greg k-h