Re: [v6, 1/2] cxl: Add mechanism for delivering AFU driver specific events

[Philippe Bergheaud <felix@linux.vnet.ibm.com>]

> Should also check against the length of user-buffer (count) provided in the read
> call.Ideally this condition check should be moved to the read call where
> you have access to the count variable.
> 
> Right now libcxl is using a harcoded value of CXL_READ_MIN_SIZE to
> issue the read call and in kernel code we have a check to ensure that
> read buffer is atleast CXL_READ_MIN_SIZE in size.
> 
> But it might be a good idea to decouple driver from
> CXL_MAX_EVENT_DATA_SIZE. Ideally the maximum event size that we can
> support should be dependent on the amount user buffer we receive in the
> read call. That way future libcxl can support larger event_data without
> needing a change to the cxl.h
> 
[...]
>>+#define CXL_MAX_EVENT_DATA_SIZE 128
>>+
> 
> 
> Agree with Matt's earlier comments. 128 is very small and I would prefer
> for atleast a page size (4k/64K) limit.
> 

afu_read() enforces a minimum buffer size of CXL_READ_MIN_SIZE = 4K, as documented in Documentation/powerpc/cxl.txt. This information is missing from the man pages of the libcxl functions cxl_read_event/cxl_read_expected_event. I will fix these.

Regarding the maximum event size, as afu_read returns one event per call, and as there is no API to tell userland the maximum size of a cxl event, I think that we should simply use (and document) the same value (4K) as the maximum cxl event size.

Philippe