Re: [patch V9 00/39] x86/entry: Rework leftovers (was part V)

[Peter Zijlstra]

On Fri, May 22, 2020 at 08:20:15AM +0100, Andrew Cooper wrote:
> Apologies for opening a related can of worms.
> 
> The new debug_enter() has propagated a pre-existing issue forward,
> ultimately caused by bad advice in the SDM.
> 
> Because the RTM status bit in DR6 has inverted polarity, writing DR6 to
> 0 causes RTM to appear asserted to any logic which cares, despite RTM
> debugging not being enabled.  The same is true in principle for what is
> handed to userspace via u_debugreg[DR_STATUS].
> 
> On the subject of DR6, the SDM now reads:
> 
> "Certain debug exceptions may clear bits 0-3. The remaining contents of
> the DR6 register are never cleared by the processor. To avoid confusion
> in identifying debug exceptions, debug handlers should clear the
> register (except bit 16, which they should set) before returning to the
> interrupted task."

*URGH*

> First of all, that should read "are never de-asserted by the processor"
> rather than "cleared", but the advice has still failed to learn from its
> first mistake.  The forward-compatible way to fix this is to set
> DR6_DEFAULT (0xffff0ff0) which also covers future inverted polarity bits.
> 
> As for what to do about userspace, that is harder.  One approach is to
> express everything in terms of positive polarity (i.e. pass on dr6 ^
> DR6_DEFAULT), so DR6_RTM only appears set when RTM debugging is
> enabled.  This approach is already taken with the VMCS PENDING_DBG
> field, so there is at least previous form.
> 
> I realise that "do nothing" might be acceptable at this point, given the
> lack of support for RTM debugging.

This! I'm thinking "do nothing" is, at this moment, the right thing to
do. If/when someone goes and tries to make RTM debugging work, they get
to figure out how to deal with this mess.