memcpy_toio transfers data in 4 byte chunks, but I need to transfer data in one lump.

I am writing a PCIe device driver that requires me to write data to the address referenced by a Base Address Register (BAR). This particular BAR is setup to be in a memory configuration (as opposed to I/O). I've checked the flags (pci_resource_flags) and confirmed that the address space is I/O Resource Memory. I need to transfer data ranging from a couple bytes to 1/3 of a kilo byte, to this space. My code does the following:

1) calls pci_iomap(), specifying the size requested by using pci_resource_len() //for setup
2) calls memcpy_tio() //for actual transfer

The data is going to and being recieved by the proper location, but the problem is that the data is being split into 4 byte packets (instead of one lump). I used a Lecroy PCIe analyzer, which showed that the data I am transferring is being split into TLPs with 4 byte payloads (this confirms the behavior). It was mentioned to me that I/O, in general, is transfered in 4 byte packets. Is there some way to have it send all the data I want to transfer in one TLP (one lump)?

Computer Info:
1) SUSE Linux Enterprise Server 10 SP1 (kernel version 2.6.16)
2) x86_64 (8 core) processor
3) Dell Precision 490

Any suggestions of how to fix this or places to look to to research this would be more than appreciated. Thank you for taking the time to read my post.

In a 64 bit processor 4 bytes = 1 word. At the lowest level the computer can only move data one word at a time, AFAIK.

Thanks for responding. Yes, 4bytes for a transfer rate makes sense to a space at a time. I am more concerned about the TLPs (Transaction Layer Packets). I believe the bus_driver is responsible for creating those TLPs, which are sent off to the Root Complex for the PCIe system. At some point, which I'm not aware of how it does it, the bus_driver decides how much data should be put into a single TLP. That TLP can have numerous different sizes, beyond 4 bytes. I need to be able to create a TLP of size 1/3 KB. I believe that this will require me to re-write the bus_driver and recompile my kernel.