On Wed, Jun 24, 2015 at 09:11:24PM +0100, Timur Tabi wrote:
> On 06/22/2015 08:12 AM, Will Deacon wrote:
> > I still think we should be disabling userspace access to the DCC if the
> > kernel is using it as its console.
>
> I still need help with this. I know you said a year ago that
> MDSCR_EL1.TDCC needs to be set to disable userspace access. Where and
> how should I do this? I can do this:
Well, it's up to you to figure out the details, but I'd start by adding
some static inlines to the arch-specific header files for enabling/disabling
userspace access.
>From there, I think I'd get the architecture init code to reset the thing
to "disabled" (so it's disabled regardless of whether we build the hvc_dcc
driver) and then if you wanted to go all-out, we could have a sysfs entry
provided by the driver to toggle it on and off.
> static int __init hvc_dcc_console_init(void)
> {
> #ifdef CONFIG_ARM64
> u32 val;
>
> asm("msr mdscr_el1, %0 "
> "orr %0, %0, #4096 " /* TDCC */
> "msr %0, mdscr_el1 "
> : "=r" (val));
> #endif
>
> But this seems clunky.
Yeah, that's super ugly.
> I am concerned about KVM, though. There appears to be code in KVM in
> hyp.s and sys_regs.c that touches and/or emulates MDSCR_EL1.
>
> On a side note, it does not appear that ARM32 blocks userspace DCC. I
> don't see where DBGDSCR.UDCCdis is set.
That's a bug imo.
Will
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
