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;
From:   KY Srinivasan <kys@microsoft.com>
To:     Thomas Gleixner <tglx@linutronix.de>
CC:     "x86@kernel.org" <x86@kernel.org>,
        "gregkh@linuxfoundation.org" <gregkh@linuxfoundation.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "devel@linuxdriverproject.org" <devel@linuxdriverproject.org>,
        "olaf@aepfle.de" <olaf@aepfle.de>,
        "apw@canonical.com" <apw@canonical.com>,
        "jasowang@redhat.com" <jasowang@redhat.com>,
        "hpa@zytor.com" <hpa@zytor.com>,
        Stephen Hemminger <sthemmin@microsoft.com>,
        "Michael Kelley (EOSG)" <Michael.H.Kelley@microsoft.com>,
        "vkuznets@redhat.com" <vkuznets@redhat.com>
Subject: RE: [PATCH 2/5] X86: Hyper-V: Enable IPI enlightenments
Thread-Topic: [PATCH 2/5] X86: Hyper-V: Enable IPI enlightenments
Thread-Index: AQHT3MEu4uYWJdgpmkipM8S1SvacPaQTnMUAgACBqNA=
Date:   Fri, 27 Apr 2018 06:11:17 +0000
Message-ID: <MW2PR2101MB11131D0D8C0B985B68308C25A08D0@MW2PR2101MB1113.namprd21.prod.outlook.com>
References: <20180425181110.8683-1-kys@linuxonhyperv.com>
 <20180425181250.8740-1-kys@linuxonhyperv.com>
 <20180425181250.8740-2-kys@linuxonhyperv.com>
 <alpine.DEB.2.21.1804262350040.1596@nanos.tec.linutronix.de>
In-Reply-To: <alpine.DEB.2.21.1804262350040.1596@nanos.tec.linutronix.de>
Accept-Language: en-US
Content-Language: en-US
received-spf: None (protection.outlook.com: microsoft.com does not designate
 permitted sender hosts)
spamdiagnosticoutput: 1:99
spamdiagnosticmetadata: NSPM
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
X-MS-Exchange-CrossTenant-Network-Message-Id: 60d109e8-9715-470c-e8f9-08d5ac05b050
X-MS-Exchange-CrossTenant-originalarrivaltime: 27 Apr 2018 06:11:17.3568
 (UTC)
X-MS-Exchange-CrossTenant-fromentityheader: Hosted
X-MS-Exchange-CrossTenant-id: 72f988bf-86f1-41af-91ab-2d7cd011db47
X-MS-Exchange-Transport-CrossTenantHeadersStamped: MW2PR2101MB1065
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org


> -----Original Message-----
> From: Thomas Gleixner <tglx@linutronix.de>
> Sent: Thursday, April 26, 2018 3:09 PM
> To: KY Srinivasan <kys@microsoft.com>
> Cc: x86@kernel.org; gregkh@linuxfoundation.org; linux-
> kernel@vger.kernel.org; devel@linuxdriverproject.org; olaf@aepfle.de;
> apw@canonical.com; jasowang@redhat.com; hpa@zytor.com; Stephen
> Hemminger <sthemmin@microsoft.com>; Michael Kelley (EOSG)
> <Michael.H.Kelley@microsoft.com>; vkuznets@redhat.com
> Subject: Re: [PATCH 2/5] X86: Hyper-V: Enable IPI enlightenments
>=20
> On Wed, 25 Apr 2018, kys@linuxonhyperv.com wrote:
> > +static int __send_ipi_mask(const struct cpumask *mask, int vector)
> > +{
> > +	int cur_cpu, vcpu;
> > +	struct ipi_arg_non_ex **arg;
> > +	struct ipi_arg_non_ex *ipi_arg;
> > +	int ret =3D 1;
>=20
> So this indicates whether __send_ipi_mask() can send to @mask or not. So
> please make it a bool and let it return false when it does not work, true
> otherwise. If you had used -Exxxx then it would have been more obvious,
> but
> this is really a boolean decision.

Agreed.
>=20
> > +	unsigned long flags;
> > +
> > +	if (cpumask_empty(mask))
> > +		return 0;
> > +
> > +	if (!hv_hypercall_pg)
> > +		return ret;
> > +
> > +	if ((vector < HV_IPI_LOW_VECTOR) || (vector >
> HV_IPI_HIGH_VECTOR))
> > +		return ret;
> > +
> > +	local_irq_save(flags);
> > +	arg =3D (struct ipi_arg_non_ex
> **)this_cpu_ptr(hyperv_pcpu_input_arg);
> > +
> > +	ipi_arg =3D *arg;
> > +	if (unlikely(!ipi_arg))
> > +		goto ipi_mask_done;
> > +
> > +
>=20
> Stray newline
>=20
> > +	ipi_arg->vector =3D vector;
> > +	ipi_arg->reserved =3D 0;
> > +	ipi_arg->cpu_mask =3D 0;
> > +
> > +	for_each_cpu(cur_cpu, mask) {
> > +		vcpu =3D hv_cpu_number_to_vp_number(cur_cpu);
> > +		if (vcpu >=3D 64)
> > +			goto ipi_mask_done;
>=20
> This is completely magic and deserves a comment.
>=20
> > +
> > +		__set_bit(vcpu, (unsigned long *)&ipi_arg->cpu_mask);
> > +	}
> > +
> > +	ret =3D hv_do_hypercall(HVCALL_SEND_IPI, ipi_arg, NULL);
> > +
> > +ipi_mask_done:
> > +	local_irq_restore(flags);
> > +	return ret;
> > +}
>=20
> ....
>=20
> >  static int hv_cpu_init(unsigned int cpu)
> >  {
> >  	u64 msr_vp_index;
> >  	struct hv_vp_assist_page **hvp =3D
> &hv_vp_assist_page[smp_processor_id()];
> > +	void **input_arg;
> > +
> > +	input_arg =3D (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
> > +	*input_arg =3D page_address(alloc_page(GFP_ATOMIC));
>=20
> This is called from the cpu hotplug thread and there is no need for an
> atomic allocation. Please use GFP_KERNEL.
>=20
> >  	hv_get_vp_index(msr_vp_index);
> >
> > @@ -217,6 +224,10 @@ static int hv_cpu_die(unsigned int cpu)
> >  {
> >  	struct hv_reenlightenment_control re_ctrl;
> >  	unsigned int new_cpu;
> > +	void **input_arg;
> > +
> > +	input_arg =3D (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
> > +	free_page((unsigned long)*input_arg);
>=20
> Hrm. Again this is called from the CPU hotplug thread when the cou is abo=
ut
> to go down. But you can be scheduled out after free() and before disablin=
g
> the assist thing below and the pointer persist. There is no guarantee tha=
t
> nothing sends an IPI anymore after this point.
>=20
> So you have two options here:
>=20
>   1) Disable interrupts, get the pointer, set the per cpu pointer to NULL=
,
>      reenable interruots and free the page
I will implement this approach.
>=20
>   2) Keep the page around and check for it in the CPU UP path and avoid t=
he
>      allocation when the CPU comes online again.
>=20
> >  	if (hv_vp_assist_page && hv_vp_assist_page[cpu])
> >  		wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
> > @@ -260,6 +271,12 @@ void __init hyperv_init(void)
> >  	if ((ms_hyperv.features & required_msrs) !=3D required_msrs)
> >  		return;
> >
> > +	/* Allocate the per-CPU state for the hypercall input arg */
> > +	hyperv_pcpu_input_arg =3D alloc_percpu(void  *);
> > +
> > +	if (hyperv_pcpu_input_arg =3D=3D NULL)
> > +		return;
>=20
> Huch. When that allocation fails, you return and ignore the rest of the
> function which has been there before. Weird decision.
I should have explained this. Failure of this allocation means that we woul=
d not have the
per-cpu hypercall input page which in turn would mean that we would not be =
able to invoke
any hypercalls.=20

Regards,

K. Y