On Wed, 2008-12-17 at 07:31 +0800, Herbert Xu wrote: > On Mon, Dec 15, 2008 at 11:38:01PM +1100, Herbert Xu wrote: > > On Mon, Dec 15, 2008 at 04:21:06PM +1100, Herbert Xu wrote: > > > > > > > a. Do not touch SSE state in soft_irq > > > > b. Disable/restore soft_irq in kernel_fpu_begin/kernel_fpu_end > > > > c. Use a per-CPU data structure to save kernel FPU state during > > > > soft_irq. > > > Here's another option > > > > d. When we're in interrupt context, schedule a task to perform > > the encryption asynchronously. > > We can also hybridise b. and d.: > > e. When we're in interrupt context, if TS is clear, then we defer > the operation to a thread. Otherwise if user-space has touched > the FPU we save the state, if not then we simply clear TS. In > either case we perform the operation immediately and then reset > TS if user-space didn't touch the FPU. > > This is based on the fact that it should be fairly rare for us > to interrupt a kernel FPU/SSE operation. The common case would > be interrupting a user process or a kernel context which is not > engaging in any kernel FPU operations. Yes. This is a better solution with much better performance. How about hybridise b. and a.: f. if TS is clear, then use x86_64 implementation. Otherwise if user-space has touched the FPU, we save the state, if not then simply clear TS. I think that could be simpler to be implemented. Best Regards, Huang Ying
Attachment:
signature.asc
Description: This is a digitally signed message part
