You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
mem_err_i and mem_exokay_i are possibly multi-bit, if NumBanks is not equal to 1. Therefore, I think either a vector-reduction OR or AND should be placed in front of those signals when deciding on the appropriate AXI response types.
mem_err_i and mem_exokay_i, and, in the same fashion possibly additional side-channel/meta signals from the downstream memory are used "directly" here, while the other fields of axi_resp_o.b/r are taken from buffers "deep" within the module. Particularly, m2s_resp in the R response comes out of a FIFO that can be any number of elements deep. So does this not create misalignment issues between the fields in the responses, e.g., an error response from a different read than what the data in m2s_resp belong to?
For a multi-beat AXI write burst where only one or few beats in the middle of the burst cause the memory to respond with an error, won't this cause the error to get lost and not being propagated to the B response at the end of the beat?
For multi-beat read bursts, if the response FIFO in stream_to_mem actually fills and cannot operate only in pass-through, this should similarly lead to error responses being assigned to the wrong R-response, or lost.
The text was updated successfully, but these errors were encountered:
I have the following questions regarding the assignment of the
resp
field of the B and R response structs inaxi_to_detailed_mem
, linesaxi/src/axi_to_detailed_mem.sv
Line 502 in 634b743
axi/src/axi_to_detailed_mem.sv
Line 511 in 634b743
mem_err_i
andmem_exokay_i
are possibly multi-bit, ifNumBanks
is not equal to 1. Therefore, I think either a vector-reductionOR
orAND
should be placed in front of those signals when deciding on the appropriate AXI response types.mem_err_i
andmem_exokay_i
, and, in the same fashion possibly additional side-channel/meta signals from the downstream memory are used "directly" here, while the other fields ofaxi_resp_o.b/r
are taken from buffers "deep" within the module. Particularly,m2s_resp
in theR
response comes out of a FIFO that can be any number of elements deep. So does this not create misalignment issues between the fields in the responses, e.g., an error response from a different read than what the data inm2s_resp
belong to?For a multi-beat AXI write burst where only one or few beats in the middle of the burst cause the memory to respond with an error, won't this cause the error to get lost and not being propagated to the B response at the end of the beat?
For multi-beat read bursts, if the response FIFO in
stream_to_mem
actually fills and cannot operate only in pass-through, this should similarly lead to error responses being assigned to the wrong R-response, or lost.The text was updated successfully, but these errors were encountered: