Altera IP Compiler for PCI Express - Manual (Page 326)

B–8 Chapter :
Descriptor/Data Interface
IP Compiler for PCI Express User Guide August 2014 Altera Corporation
Each virtual channel has a dedicated datapath and associated buffers and no ordering
relationships exist between virtual channels. While one virtual channel may be
temporarily blocked, data flow continues across other virtual channels without
impact. Within a virtual channel, reordering is mandatory only for non-posted
transactions to prevent deadlock. Reordering is not implemented in the following
cases:
■ Between traffic classes mapped in the same virtual channel
■ Between posted and completion transactions
■ Between transactions of the same type regardless of the relaxed-ordering bit of
the transaction layer packet
In Figure B–6 , the IP core receives a memory read request transaction of 4 DWORDS
that it cannot immediately accept. A second transaction (memory write transaction of
one DWORD) is waiting in the receive buffer. Bit 2 of rx_data[63:0] for the memory
write request is set to 1.
In clock cycle three, transmission of non-posted transactions is not permitted for as
long as rx_mask is asserted.
Flow control credits are updated only after a transaction layer packet has been
extracted from the receive buffer and both the descriptor phase and data phase (if
any) have ended. This update happens in clock cycles 8 and 12 in
Figure B–6 .
Figure B–6. RX Retried Transaction and Masked Non-Posted Transaction Waveform
clk
rx_req
rx_ack
rx_desc[135:128]
rx_desc[127:64]
rx_desc[63:0]
rx_abort
rx_retry
rx_mask
rx_dfr
rx_dv
rx_ws
rx_data[63:32]
rx_data[31:0]
rx_be[7:0]
valid valid valid
MEMRD 4 DW MEMWR 1DW MEMRD 4DW
valid valid valid
DW 0
0x00 0xF0 0x00
1 2 3 4 5 6 7 8 9 10 11 12
Descriptor
Signals
Data
Signals