RE: [PATCH 4/6] Drivers: hv: vmbus: Introduce vmbus_request_addr_match()

[Michael Kelley (LINUX)]

From: Andrea Parri <parri.andrea@xxxxxxxxx> Sent: Friday, April 8, 2022 9:47 AM
> 
> > > @@ -1300,25 +1294,60 @@ u64 vmbus_request_addr(struct vmbus_channel
> > > *channel, u64 trans_id)
> > >  	if (!trans_id)
> > >  		return VMBUS_RQST_ERROR;
> > >
> > > -	spin_lock_irqsave(&rqstor->req_lock, flags);
> > > -
> > >  	/* Data corresponding to trans_id is stored at trans_id - 1 */
> > >  	trans_id--;
> > >
> > >  	/* Invalid trans_id */
> > > -	if (trans_id >= rqstor->size || !test_bit(trans_id, rqstor->req_bitmap)) {
> > > -		spin_unlock_irqrestore(&rqstor->req_lock, flags);
> > > +	if (trans_id >= rqstor->size || !test_bit(trans_id, rqstor->req_bitmap))
> > >  		return VMBUS_RQST_ERROR;
> > > -	}
> > >
> > >  	req_addr = rqstor->req_arr[trans_id];
> > > -	rqstor->req_arr[trans_id] = rqstor->next_request_id;
> > > -	rqstor->next_request_id = trans_id;
> > > +	if (rqst_addr == VMBUS_RQST_ADDR_ANY || req_addr == rqst_addr) {
> > > +		rqstor->req_arr[trans_id] = rqstor->next_request_id;
> > > +		rqstor->next_request_id = trans_id;
> > >
> > > -	/* The already held spin lock provides atomicity */
> > > -	bitmap_clear(rqstor->req_bitmap, trans_id, 1);
> > > +		/* The already held spin lock provides atomicity */
> > > +		bitmap_clear(rqstor->req_bitmap, trans_id, 1);
> > > +	}
> >
> > In the case where a specific match is required, and trans_id is
> > valid but the addr's do not match, it looks like this function will
> > return the addr that didn't match, without removing the entry.
> 
> Yes, that is consistent with the description on vmbus_request_addr_match():
> 
>   Returns the memory address stored at @trans_id, or VMBUS_RQST_ERROR if
>   @trans_id is not contained in the requestor.
> 
> 
> > Shouldn't it return VMBUS_RQST_ERROR in that case?
> 
> Can certainly be done, although I'm not sure to follow your concerns.  Can
> you elaborate?
> 

Having the function return "success" when it failed to match is unexpected
for me.  There's only one invocation where we care about matching
(in hv_compose_msi_msg).  In that invocation the purpose for matching is to
not remove the wrong entry, and the return value is ignored.  So I think
it all works correctly.

Just thinking out loud, maybe vmbus_request_addr_match() should be
renamed to vmbus_request_addr_remove(), and not have a return value?
That would be a bit more consistent with the actual purpose.

Michael