Am 11/09/2014 16:54, schrieb Paweł Dziepak:
I am going to negotiate with lvm and see if I can get one of my disks back to do a proper bare metal test.
Good luck with that :-)
If we disable SSE in interrupt handlers we will have to basically disable SSE in almost all kernel code anyway. For example: interrupt -> page fault handler -> vfs io map -> any file system module + scheduler + timers. That "target" attribute won't be enough if the vfs does an indirect call to a module, GCC wouldn't be able to ensure that the module doesn't use SSE.
You're right; I was only thinking of driver interrupts which usually have very low demands. But if that is ever going to be a problem, we could still use different interrupt paths for those with and those without the need for SSE (probably hard coded to driver vs. page faults etc. -- or maybe via a flag on interrupt registration).
Bye, Axel.