Re: [PATCH v2 08/21] arm64: KVM: Implement debug save/restore

[Christoffer Dall <christoffer.dall@linaro.org>]

On Tue, Dec 01, 2015 at 03:01:16PM +0000, Marc Zyngier wrote:
> On 01/12/15 14:47, Christoffer Dall wrote:
> > On Tue, Dec 01, 2015 at 01:06:31PM +0000, Marc Zyngier wrote:
> >> On 01/12/15 12:56, Christoffer Dall wrote:
> >>> On Fri, Nov 27, 2015 at 06:50:02PM +0000, Marc Zyngier wrote:
> >>>> Implement the debug save restore as a direct translation of
> >>>> the assembly code version.
> >>>>
> >>>> Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
> >>>> ---
> >>>>  arch/arm64/kvm/hyp/Makefile   |   1 +
> >>>>  arch/arm64/kvm/hyp/debug-sr.c | 130 ++++++++++++++++++++++++++++++++++++++++++
> >>>>  arch/arm64/kvm/hyp/hyp.h      |   9 +++
> >>>>  3 files changed, 140 insertions(+)
> >>>>  create mode 100644 arch/arm64/kvm/hyp/debug-sr.c
> >>>>
> >>>> diff --git a/arch/arm64/kvm/hyp/Makefile b/arch/arm64/kvm/hyp/Makefile
> >>>> index ec94200..ec14cac 100644
> >>>> --- a/arch/arm64/kvm/hyp/Makefile
> >>>> +++ b/arch/arm64/kvm/hyp/Makefile
> >>>> @@ -6,3 +6,4 @@ obj-$(CONFIG_KVM_ARM_HOST) += vgic-v2-sr.o
> >>>>  obj-$(CONFIG_KVM_ARM_HOST) += vgic-v3-sr.o
> >>>>  obj-$(CONFIG_KVM_ARM_HOST) += timer-sr.o
> >>>>  obj-$(CONFIG_KVM_ARM_HOST) += sysreg-sr.o
> >>>> +obj-$(CONFIG_KVM_ARM_HOST) += debug-sr.o
> >>>> diff --git a/arch/arm64/kvm/hyp/debug-sr.c b/arch/arm64/kvm/hyp/debug-sr.c
> >>>> new file mode 100644
> >>>> index 0000000..a0b2b99
> >>>> --- /dev/null
> >>>> +++ b/arch/arm64/kvm/hyp/debug-sr.c
> >>>> @@ -0,0 +1,130 @@
> >>>> +/*
> >>>> + * Copyright (C) 2015 - ARM Ltd
> >>>> + * Author: Marc Zyngier <marc.zyngier@arm.com>
> >>>> + *
> >>>> + * This program is free software; you can redistribute it and/or modify
> >>>> + * it under the terms of the GNU General Public License version 2 as
> >>>> + * published by the Free Software Foundation.
> >>>> + *
> >>>> + * This program is distributed in the hope that it will be useful,
> >>>> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> >>>> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> >>>> + * GNU General Public License for more details.
> >>>> + *
> >>>> + * You should have received a copy of the GNU General Public License
> >>>> + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
> >>>> + */
> >>>> +
> >>>> +#include <linux/compiler.h>
> >>>> +#include <linux/kvm_host.h>
> >>>> +
> >>>> +#include <asm/kvm_mmu.h>
> >>>> +
> >>>> +#include "hyp.h"
> >>>> +
> >>>> +#define read_debug(r,n)		read_sysreg(r##n##_el1)
> >>>> +#define write_debug(v,r,n)	write_sysreg(v, r##n##_el1)
> >>>> +
> >>>> +#define save_debug(ptr,reg,nr)						\
> >>>> +	switch (nr) {							\
> >>>> +	case 15:	ptr[15] = read_debug(reg, 15);			\
> >>>> +	case 14:	ptr[14] = read_debug(reg, 14);			\
> >>>> +	case 13:	ptr[13] = read_debug(reg, 13);			\
> >>>> +	case 12:	ptr[12] = read_debug(reg, 12);			\
> >>>> +	case 11:	ptr[11] = read_debug(reg, 11);			\
> >>>> +	case 10:	ptr[10] = read_debug(reg, 10);			\
> >>>> +	case 9:		ptr[9] = read_debug(reg, 9);			\
> >>>> +	case 8:		ptr[8] = read_debug(reg, 8);			\
> >>>> +	case 7:		ptr[7] = read_debug(reg, 7);			\
> >>>> +	case 6:		ptr[6] = read_debug(reg, 6);			\
> >>>> +	case 5:		ptr[5] = read_debug(reg, 5);			\
> >>>> +	case 4:		ptr[4] = read_debug(reg, 4);			\
> >>>> +	case 3:		ptr[3] = read_debug(reg, 3);			\
> >>>> +	case 2:		ptr[2] = read_debug(reg, 2);			\
> >>>> +	case 1:		ptr[1] = read_debug(reg, 1);			\
> >>>> +	default:	ptr[0] = read_debug(reg, 0);			\
> >>>> +	}
> >>>> +
> >>>> +#define restore_debug(ptr,reg,nr)					\
> >>>> +	switch (nr) {							\
> >>>> +	case 15:	write_debug(ptr[15], reg, 15);			\
> >>>> +	case 14:	write_debug(ptr[14], reg, 14);			\
> >>>> +	case 13:	write_debug(ptr[13], reg, 13);			\
> >>>> +	case 12:	write_debug(ptr[12], reg, 12);			\
> >>>> +	case 11:	write_debug(ptr[11], reg, 11);			\
> >>>> +	case 10:	write_debug(ptr[10], reg, 10);			\
> >>>> +	case 9:		write_debug(ptr[9], reg, 9);			\
> >>>> +	case 8:		write_debug(ptr[8], reg, 8);			\
> >>>> +	case 7:		write_debug(ptr[7], reg, 7);			\
> >>>> +	case 6:		write_debug(ptr[6], reg, 6);			\
> >>>> +	case 5:		write_debug(ptr[5], reg, 5);			\
> >>>> +	case 4:		write_debug(ptr[4], reg, 4);			\
> >>>> +	case 3:		write_debug(ptr[3], reg, 3);			\
> >>>> +	case 2:		write_debug(ptr[2], reg, 2);			\
> >>>> +	case 1:		write_debug(ptr[1], reg, 1);			\
> >>>> +	default:	write_debug(ptr[0], reg, 0);			\
> >>>> +	}
> >>>> +
> >>>> +void __hyp_text __debug_save_state(struct kvm_vcpu *vcpu,
> >>>> +				   struct kvm_guest_debug_arch *dbg,
> >>>> +				   struct kvm_cpu_context *ctxt)
> >>>> +{
> >>>> +	if (vcpu->arch.debug_flags & KVM_ARM64_DEBUG_DIRTY) {
> >>>> +		u64 aa64dfr0 = read_sysreg(id_aa64dfr0_el1);
> >>>> +		int brps, wrps;
> >>>> +
> >>>> +		brps = (aa64dfr0 >> 12) & 0xf;
> >>>> +		wrps = (aa64dfr0 >> 20) & 0xf;
> >>>> +
> >>>> +		save_debug(dbg->dbg_bcr, dbgbcr, brps);
> >>>> +		save_debug(dbg->dbg_bvr, dbgbvr, brps);
> >>>> +		save_debug(dbg->dbg_wcr, dbgwcr, wrps);
> >>>> +		save_debug(dbg->dbg_wvr, dbgwvr, wrps);
> >>>> +
> >>>> +		ctxt->sys_regs[MDCCINT_EL1] = read_sysreg(mdccint_el1);
> >>>> +	}
> >>>> +}
> >>>> +
> >>>> +void __hyp_text __debug_restore_state(struct kvm_vcpu *vcpu,
> >>>> +				      struct kvm_guest_debug_arch *dbg,
> >>>> +				      struct kvm_cpu_context *ctxt)
> >>>> +{
> >>>> +	if (vcpu->arch.debug_flags & KVM_ARM64_DEBUG_DIRTY) {
> >>>> +		u64 aa64dfr0 = read_sysreg(id_aa64dfr0_el1);
> >>>> +		int brps, wrps;
> >>>> +
> >>>> +		brps = (aa64dfr0 >> 12) & 0xf;
> >>>> +		wrps = (aa64dfr0 >> 20) & 0xf;
> >>>> +
> >>>> +		restore_debug(dbg->dbg_bcr, dbgbcr, brps);
> >>>> +		restore_debug(dbg->dbg_bvr, dbgbvr, brps);
> >>>> +		restore_debug(dbg->dbg_wcr, dbgwcr, wrps);
> >>>> +		restore_debug(dbg->dbg_wvr, dbgwvr, wrps);
> >>>> +
> >>>> +		write_sysreg(ctxt->sys_regs[MDCCINT_EL1], mdccint_el1);
> >>>> +	}
> >>>> +}
> >>>> +
> >>>> +void __hyp_text __debug_cond_save_host_state(struct kvm_vcpu *vcpu)
> >>>> +{
> >>>> +	if ((vcpu->arch.ctxt.sys_regs[MDSCR_EL1] & DBG_MDSCR_KDE) ||
> >>>> +	    (vcpu->arch.ctxt.sys_regs[MDSCR_EL1] & DBG_MDSCR_KDE))
> >>>> +		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
> >>>> +
> >>>> +	__debug_save_state(vcpu, &vcpu->arch.host_debug_state,
> >>>> +			   kern_hyp_va(vcpu->arch.host_cpu_context));
> >>>
> >>> doesn't the assmebly code jump across saving this state neither bits are
> >>> set where this always saves the state?
> >>
> >> It doesn't. The save/restore functions are guarded by tests on
> >> KVM_ARM64_DEBUG_DIRTY, just like we have skip_debug_state on all actions
> >> involving the save/restore in the assembly version.
> > 
> > I think the confusing part is that the save function unconditionally
> > calls __debug_save_state where the restore function only calls it when
> > the dirty flag is set.  Plus I suck at reading assembly apparently.
> 
> So the way I initially wrote it, I had the same 'if' statement as in the
> restore function, making them fairly symmetric. But it quickly became
> obvious that this double-if was a bit pointless.
> 
> And actually, I wonder if I shouldn't drop it from the restore function,
> because it only save us a spurious clear of the dirty bit.
> 

I would just move the __debug_restore_state call above the conditional,
then they look more symmetric.  Does that work?

> >>> in any case, I feel some context is lost when this is moved away from
> >>> assembly and understanding this patch would be easier if the semantics
> >>> of these two _cond functions were documented.
> >>
> >> I can migrate the existing comments if you think that helps.
> >>
> > It just wasn't not quite clear to me exactly when
> > __debug_cond_save_host_state is called for example - is this going to be
> > called unconditionally on every entry - that's how I understand it now
> > anyway.
> 
> On every entry, yes. I'm trying to have the guest_run function as simple
> as possible, with the various subsystems making their 'own' decisions.
> 
> Not optimal (you get to branch for nothing), but clearer. At least for
> me, but I may be the odd duck out here. Any idea to make the flow look
> clearer?
> 

For me, if you always make the call unconditionally on both paths and
then change the implementations to do 

	if (!(vcpu->arch.debug_flags & KVM_ARM64_DEBUG_DIRTY))
		return;

then I think it's clear enough.

Thanks,
-Christoffer