Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 17 Aug 2023 09:27:38 +0100
Message-ID: <87bkf6oyyt.wl-maz@kernel.org>
From:   Marc Zyngier <maz@kernel.org>
To:     Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com>
Cc:     linux-kernel@vger.kernel.org, kvmarm@lists.linux.dev,
        linux-arm-kernel@lists.infradead.org, eauger@redhat.com,
        miguel.luis@oracle.com, darren@os.amperecomputing.com,
        scott@os.amperecomputing.com,
        Christoffer Dall <Christoffer.Dall@arm.com>
Subject: Re: [PATCH 2/2] KVM: arm64: timers: Adjust CVAL of a ptimer across guest entry and exits
In-Reply-To: <20230817060314.535987-3-gankulkarni@os.amperecomputing.com>
References: <20230817060314.535987-1-gankulkarni@os.amperecomputing.com>
        <20230817060314.535987-3-gankulkarni@os.amperecomputing.com>
User-Agent: Wanderlust/2.15.9 (Almost Unreal) SEMI-EPG/1.14.7 (Harue)
 FLIM-LB/1.14.9 (=?UTF-8?B?R29qxY0=?=) APEL-LB/10.8 EasyPG/1.0.0 Emacs/28.2
 (x86_64-pc-linux-gnu) MULE/6.0 (HANACHIRUSATO)
MIME-Version: 1.0 (generated by SEMI-EPG 1.14.7 - "Harue")
Content-Type: text/plain; charset=US-ASCII
Precedence: bulk

[Fixing Christoffer's email address]

On Thu, 17 Aug 2023 07:03:14 +0100,
Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com> wrote:
> 
> As per FEAT_ECV, when HCR_EL2.{E2H, TGE} == {1, 1}, Enhanced Counter
> Virtualization functionality is disabled and CNTPOFF_EL2 value is treated
> as zero. On VHE host, E2H and TGE are set, hence it is required
> to adjust CVAL by incrementing it by CNTPOFF_EL2 after guest
> exit to avoid false physical timer interrupts and also
> decrement/restore CVAL before the guest entry.

No, this is wrong. Neither E2H nor TGE have any impact on writing to
CNTPOFF_EL2, nor does it have an impact on CNTP_CVAL_EL0. Just read
the pseudocode to convince yourself.

CNTPOFF_EL2 is applied at exactly two points: when SW is reading
CNTPCT_EL0 from EL1 while {E2H,TGE}=={1, 0} and when the HW is
comparing CNTPCT_EL0 with the CNTP_CVAL_EL0. In both cases the offset
is subtracted from the counter. And that's the point where the running
EL matters. Which means that CNTPOFF_EL2 behaves exactly like
CNTVOFF_EL2. No ifs, no buts.

The behaviour you are describing tends to indicate that your HW is
applying CNTPOFF_EL2 to CVAL, which would be a gold plated bug.

It doesn't mean that the KVM code is correct either. I'm seeing pretty
bad behaviours on a machine without CNTPOFF_EL2. But what you are
suggesting is IMO a misunderstanding of the architecture.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.