Nowadays I have this line in my source code:

iowrite32(dout,BASE_ADDR_0_START);

so, I only can write 32 bits data at time. I would like to know if there are ways to send more than 32 bits data, to improve the performance. As I read the PCIe has the TLP packet which can contain a maximum of 1024 bytes, how can I introduce 1024 at the same time in my device driver?

Thanks =)

The kernel's iowrite32() call writes data to an I/O address that has been mapped to kernel memory space. Most of your device's registers are almost certainly only 32 bits wide, and an iowrite32 is appropriate in those cases where a 32-bit register is the target.

The PCIe bus will collect I/O writes from the OS and write them to your device in bursts, rather than one at a time (unless a PCI posting read is executed, which forces all cached data to be written).

You can't improve performance by changing iowrite32 to anything else.

Thanks for your answer.

So, the only way to improve the performance is implementing a DMA. Am I right?

Regards.

Well, there may be other improvements that can be made in the driver, but DMA would help.

I've already read the Linux device drivers, but I couldn't get started in implementing my DMA. Is there others sources that can help me?

Regards.