On 2020-09-23, Jeremie Courreges-Anglas wrote:
> ok?
Seems fine.
> Note: I inlined the apmd(8)->apm(8) perfpolicy conversion for now, which
> brings a question. I find it weird that there is a special "high"
> perfpolicy (effectively similar to perfpolicy=manual + setperf=100) but
> no "low"
Theo de Raadt wrote:
> kettenis, mortimer, and I have had long discussions about the
> uvm_map_inentry() checks in MD trap.c being excessively broad. We can
> this check-function less often.
>
> I don't want to explain the full picture about how SP and PC checks
> in syscall() and trap()
Theo de Raadt wrote:
> Theo de Raadt wrote:
>
> > kettenis, mortimer, and I have had long discussions about the
> > uvm_map_inentry() checks in MD trap.c being excessively broad. We can
> > this check-function less often.
> >
> > I don't want to explain the full picture about how SP and PC
Theo de Raadt wrote:
> kettenis, mortimer, and I have had long discussions about the
> uvm_map_inentry() checks in MD trap.c being excessively broad. We can
> this check-function less often.
>
> I don't want to explain the full picture about how SP and PC checks
> in syscall() and trap()
kettenis, mortimer, and I have had long discussions about the
uvm_map_inentry() checks in MD trap.c being excessively broad. We can
this check-function less often.
I don't want to explain the full picture about how SP and PC checks
in syscall() and trap() hinder ROP, ROP stack pivoting, gadgets
On Wed, Sep 23 2020, Jeremie Courreges-Anglas wrote:
> Prompted by a report from Miod: setting hw.setperf works only if the
> kernel doesn't have a usable cpu_setperf implementation. The current
> apmd(8) code warns if setting hw.perfpolicy fails, but then handles
> back bogus values to apm(8)
Prompted by a report from Miod: setting hw.setperf works only if the
kernel doesn't have a usable cpu_setperf implementation. The current
apmd(8) code warns if setting hw.perfpolicy fails, but then handles
back bogus values to apm(8) clients.
The easy fix is to just query the kernel about the