Hello,

I have develop a linux driver and it works fine.

But I have some doubts it. For example I have a PCI card with FPGA and it have two CAN ports. Anyway, I can write and read pci registers.

My problem is that I do not what happen is there are two threads with RT priority. For example thread one with priority -90 and thread two with priority -85. In this conditions, thread one can interrupt thread two. Then, my questions is, if thread two try to read 4 registers from PCI and thread one interrupt it to write other 4 register in pci card, what is the result???

I am interesting in all threads can completed all read or all write register without interrupt. I have read information about atomic and spinlock but I not sure if it is necessary.

I am very sorry if I have not explained correctly.

Best reagards.

Spinlocks et al are absolutely necessary. You must design your entire driver so that it uses these mechanisms correctly and appropriately. This is the solution to your problem.

Consider this: your driver code will be executed thousands of times a second. If the odds are "one in a million" that this-or-that timing problem might occur, then the odds are near-100% that the system will crash-and-burn in just a few minutes (at best).

Remember also: there may be multiple CPUs involved. Therefore, the number of timing-possibilities cannot be predicted. Your logic therefore must use kernel locking mechanisms in such a way that race-conditions cannot occur.