Re: [Basics] spinlocks & copy_to_user

[Linus Torvalds]

In article <[e-mail]>, ejc  <[e-mail]> wrote:
>Summary:
>
>What is the Right Way(tm) for a driver (in read(), write(), or ioctl(),
>etc.) to copy data, possibly modified at interrupt time, to/from a user?

There are three:
 (a) use a queue, where interrupts add to the end of the queue, but
     leave the part you're copying from unmodified.
 (b) copy to a temporary buffer with spinlocks, then copy from that temp
     buffer to user space.
 (c) don't do it. 

There are _very_ few reasons to do what you suggest, almost no driver
I've ever seen depends on anything like that. You may just have a design
that is fundamentally flawed. So (c) is not just being facetious, I'm
serious.

		Linus