Another armv7 pmap diff

2016-08-20 Thread Mark Kettenis
The current pmap_enter() implementation always enters PROT_WRITE mappings as read-only and relies on pmap_fault_fixup() to fix this up. This is obviously not the intention, since we do go through the trouble of checking whether we have to do modified bit emulation. The diff below fixes this. ok?

Re: Another armv7 pmap diff

2016-08-07 Thread Marcus Glocker
On Sun, Aug 07, 2016 at 01:58:06PM +0200, Mark Kettenis wrote: > The ARMv7 ARM explicitly states that all data caches in the system are > effectively PIPT. This effectively means that there is no reason to > clean or invalidate data caches except when: > > 1. We need to synchronize the instructi

Re: Another armv7 pmap diff

2016-08-07 Thread Daniel Bolgheroni
On Sun, Aug 07, 2016 at 01:58:06PM +0200, Mark Kettenis wrote: > Further testing, especially on Cortex-A8, would be appreciated. Works for me. Tested on BeagleBone Black. Kernel compile times remains pretty much the same, though. -- db

Another armv7 pmap diff

2016-08-07 Thread Mark Kettenis
The ARMv7 ARM explicitly states that all data caches in the system are effectively PIPT. This effectively means that there is no reason to clean or invalidate data caches except when: 1. We need to synchronize the instruction cache with the data cache. 2. We change a page to be non-cachable. 3.