Thomas Falcon [off-list ref] writes:

On 11/24/20 11:43 PM, Michael Ellerman wrote:

quoted

Thomas Falcon [off-list ref] writes:

quoted

Ensure that received Subordinate Command-Response Queue (SCRQ)
entries are properly read in order by the driver. These queues
are used in the ibmvnic device to process RX buffer and TX completion
descriptors. dma_rmb barriers have been added after checking for a
pending descriptor to ensure the correct descriptor entry is checked
and after reading the SCRQ descriptor to ensure the entire
descriptor is read before processing.

Fixes: 032c5e828 ("Driver for IBM System i/p VNIC protocol")
Signed-off-by: Thomas Falcon <redacted>
---
  drivers/net/ethernet/ibm/ibmvnic.c | 8 ++++++++
  1 file changed, 8 insertions(+)

diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c
index 2aa40b2..489ed5e 100644
--- a/drivers/net/ethernet/ibm/ibmvnic.c
+++ b/drivers/net/ethernet/ibm/ibmvnic.c

@@ -2403,6 +2403,8 @@ static int ibmvnic_poll(struct napi_struct *napi, int budget)
  
  		if (!pending_scrq(adapter, adapter->rx_scrq[scrq_num]))
  			break;
+		/* ensure that we do not prematurely exit the polling loop */
+		dma_rmb();

I'd be happier if these comments were more specific about which read(s)
they are ordering vs which other read(s).

I'm sure it's obvious to you, but it may not be to a future author,
and/or after the code has been refactored over time.

Thank you for reviewing! I will submit a v2 soon with clearer comments 
on the reads being ordered here.

Thanks.

cheers