Re: Spectre defence for armv7

2018-01-14 Thread Mark Kettenis
> Date: Sat, 13 Jan 2018 23:08:55 + > From: Dimitris Papastamos > > > + case CPU_ID_CORTEX_A15: > > + case CPU_ID_CORTEX_A57: > > + case CPU_ID_CORTEX_A72: > > + /* > > +* Vulnerable; BPIALL is "not effective" so must use > > +* ICIALLU

Re: Spectre defence for armv7

2018-01-13 Thread Dimitris Papastamos
> + case CPU_ID_CORTEX_A15: > + case CPU_ID_CORTEX_A57: > + case CPU_ID_CORTEX_A72: > + /* > + * Vulnerable; BPIALL is "not effective" so must use > + * ICIALLU and hope the firmware set the magic bit in > + * the ACTLR that actually

Spectre defence for armv7

2018-01-13 Thread Mark Kettenis
The diff below improves our resiliency against "variant 2". Like on x86 the defence is based on flushing the branch predictor cache at the appropriate points. It turns out we are already in pretty good shape as we are already flushing on context switches. I believe we're forced to do that