Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Content-Type: text/plain;
        charset=us-ascii
Mime-Version: 1.0 (1.0)
Subject: Re: [patch 2/5] x86/kvm: Handle task_work on VMENTER/EXIT
From:   Andy Lutomirski <luto@amacapital.net>
In-Reply-To: <alpine.DEB.2.21.1908030015330.4029@nanos.tec.linutronix.de>
Date:   Fri, 2 Aug 2019 15:39:22 -0700
Cc:     Paolo Bonzini <pbonzini@redhat.com>,
        Sean Christopherson <sean.j.christopherson@intel.com>,
        Oleg Nesterov <oleg@redhat.com>,
        LKML <linux-kernel@vger.kernel.org>, x86@kernel.org,
        Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@kernel.org>,
        Sebastian Siewior <bigeasy@linutronix.de>,
        Anna-Maria Gleixner <anna-maria@linutronix.de>,
        Steven Rostedt <rostedt@goodmis.org>,
        Julia Cartwright <julia@ni.com>,
        Paul McKenney <paulmck@linux.vnet.ibm.com>,
        Frederic Weisbecker <fweisbec@gmail.com>, kvm@vger.kernel.org,
        Radim Krcmar <rkrcmar@redhat.com>,
        John Stultz <john.stultz@linaro.org>,
        Andy Lutomirski <luto@kernel.org>,
        "Paul E. McKenney" <paulmck@linux.ibm.com>
Content-Transfer-Encoding: quoted-printable
Message-Id: <89E42BCC-47A8-458B-B06A-D6A20D20512C@amacapital.net>
References: <20190801143250.370326052@linutronix.de> <20190801143657.887648487@linutronix.de> <20190801162451.GE31538@redhat.com> <alpine.DEB.2.21.1908012025100.1789@nanos.tec.linutronix.de> <20190801213550.GE6783@linux.intel.com> <alpine.DEB.2.21.1908012343530.1789@nanos.tec.linutronix.de> <alpine.DEB.2.21.1908012345000.1789@nanos.tec.linutronix.de> <c8294b01-62d1-95df-6ff6-213f945a434f@redhat.com> <alpine.DEB.2.21.1908030015330.4029@nanos.tec.linutronix.de>
To:     Thomas Gleixner <tglx@linutronix.de>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


> On Aug 2, 2019, at 3:22 PM, Thomas Gleixner <tglx@linutronix.de> wrote:
>=20
>> On Fri, 2 Aug 2019, Paolo Bonzini wrote:
>>> On 01/08/19 23:47, Thomas Gleixner wrote:
>>> Right you are about cond_resched() being called, but for SRCU this does n=
ot
>>> matter unless there is some way to do a synchronize operation on that SR=
CU
>>> entity. It might have some other performance side effect though.
>>=20
>> I would use srcu_read_unlock/lock around the call.
>>=20
>> However, I'm wondering if the API can be improved because basically we
>> have six functions for three checks of TIF flags.  Does it make sense to
>> have something like task_has_request_flags and task_do_requests (names
>> are horrible I know) that can be used like
>>=20
>>    if (task_has_request_flags()) {
>>        int err;
>>        ...srcu_read_unlock...
>>        // return -EINTR if signal_pending
>>        err =3D task_do_requests();
>>        if (err < 0)
>>            goto exit_no_srcu_read_unlock;
>>        ...srcu_read_lock...
>>    }
>>=20
>> taking care of all three cases with a single hook?  This is important
>> especially because all these checks are done by all KVM architectures in
>> slightly different ways, and a unified API would be a good reason to
>> make all architectures look the same.
>>=20
>> (Of course I could also define this unified API in virt/kvm/kvm_main.c,
>> so this is not blocking the series in any way!).
>=20
> You're not holding up something. Having a common function for this is
> definitely the right approach.
>=20
> As this is virt specific because it only checks for non arch specific bits=

> (TIF_NOTIFY_RESUME should be available for all KVM archs) and the TIF bits=

> are a subset of the available TIF bits because all others do not make any
> sense there, this really should be a common function for KVM so that all
> other archs which obviously lack a TIF_NOTIFY_RESUME check, can be fixed u=
p
> and consolidated. If we add another TIF check later then we only have to d=
o
> it in one place.
>=20
>=20

If we add a real API for this, can we make it, or a very similar API, work f=
or exit_to_usermode_loop() too?  Maybe:

bool usermode_work_pending();
bool guestmode_work_pending();

void do_usermode_work();
void do_guestmode_work();

The first two are called with IRQs off.  The latter two are called with IRQs=
 on.=