bug in dma_if_pcie_rd when max read request size is set to 4096 bytes

[EricCSun]

Hi Alex,
please consider this situation:

1. max read request size is set to 4096 bytes.
2. request a dma read from any address that is 4K aligned (say 0x1000), dma length is 4096 bytes.

Normally only one Mrd TLP will be sent, but dma_if_pcie_rd generates two TLPs, one is a Mrd from address 0x1000 and another is a Mrd from address 0x2000, both TLP's dword count is 1024 (4096 bytes). The later Mrd will corrupt the data.

I think the cause is in dma_if_pcie_rd.v, line 675:

req_last_tlp = (((req_pcie_addr_reg & 12'hfff) + (req_op_count_reg & 12'hfff)) & 12'hfff) == 0 && req_op_count_reg >> 12 == 0;

In above situation,"(((req_pcie_addr_reg & 12'hfff) + (req_op_count_reg & 12'hfff)) & 12'hfff) == 0" is true, req_op_count_reg is 4096, hence “req_op_count_reg >> 12” equals 1,"req_op_count_reg >> 12 == 0" is false, req_last_tlp is not assigned to 1, an unexpected tlp will be generated.

Is it OK to delete "&& req_op_count_reg >> 12 == 0" in line 675?
req_op_count_reg must be not greater than max read request size to reach line 675 (see line 663) and the maximum value of max read request size is 4096. "req_op_count_reg >> 12 == 0" is false only when req_op_count_reg equals 4096.
In the case that req_op_count_reg equals 4096, there is a chance that it is the last TLP (above situation), and "(((req_pcie_addr_reg & 12'hfff) + (req_op_count_reg & 12'hfff)) & 12'hfff) == 0" will determine wheter it is the last TLP.
In other cases that req_op_count_reg is less than 4096, "req_op_count_reg >> 12 == 0" is always false, so there is no need to evaluate it.

Please tell me if I was wrong.