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:   Mon, 29 May 2023 09:48:48 +0100
Message-ID: <86cz2jcykv.wl-maz@kernel.org>
From:   Marc Zyngier <maz@kernel.org>
To:     "Liao, Chang" <liaochang1@huawei.com>
Cc:     Shanker Donthineni <sdonthineni@nvidia.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
        Michael Walle <michael@walle.cc>,
        <linux-kernel@vger.kernel.org>, Vikram Sethi <vsethi@nvidia.com>,
        Jason Sequeira <jsequeira@nvidia.com>
Subject: Re: [PATCH v5 1/3] genirq: Use hlist for managing resend handlers
In-Reply-To: <6dc6642a-1e7c-f111-1fa2-be54826ecef6@huawei.com>
References: <20230519134902.1495562-1-sdonthineni@nvidia.com>
        <20230519134902.1495562-2-sdonthineni@nvidia.com>
        <6dc6642a-1e7c-f111-1fa2-be54826ecef6@huawei.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
 (aarch64-unknown-linux-gnu) MULE/6.0 (HANACHIRUSATO)
MIME-Version: 1.0 (generated by SEMI-EPG 1.14.7 - "Harue")
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Mon, 29 May 2023 08:57:07 +0100,
"Liao, Chang" <liaochang1@huawei.com> wrote:
>=20
> Hi, Shanker
>=20
> =E5=9C=A8 2023/5/19 21:49, Shanker Donthineni =E5=86=99=E9=81=93:
> > The current implementation utilizes a bitmap for managing IRQ resend
> > handlers, which is allocated based on the SPARSE_IRQ/NR_IRQS macros.
> > However, this method may not efficiently utilize memory during runtime,
> > particularly when IRQ_BITMAP_BITS is large.
> >=20
> > Address this issue by using the hlist to manage IRQ resend handlers
> > instead of relying on a static bitmap memory allocation. Additionally,
> > a new function, clear_irq_resend(), is introduced and called from
> > irq_shutdown to ensure a graceful teardown of the interrupt.
> >=20
> > Signed-off-by: Shanker Donthineni <sdonthineni@nvidia.com>
> > ---
> >  include/linux/irqdesc.h |  3 +++
> >  kernel/irq/chip.c       |  1 +
> >  kernel/irq/internals.h  |  2 ++
> >  kernel/irq/irqdesc.c    |  2 ++
> >  kernel/irq/resend.c     | 47 ++++++++++++++++++++++++++---------------
> >  5 files changed, 38 insertions(+), 17 deletions(-)
> >=20
> > diff --git a/include/linux/irqdesc.h b/include/linux/irqdesc.h
> > index 844a8e30e6de..d9451d456a73 100644
> > --- a/include/linux/irqdesc.h
> > +++ b/include/linux/irqdesc.h
> > @@ -102,6 +102,9 @@ struct irq_desc {
> >  	int			parent_irq;
> >  	struct module		*owner;
> >  	const char		*name;
> > +#ifdef CONFIG_HARDIRQS_SW_RESEND
> > +	struct hlist_node	resend_node;
> > +#endif
> >  } ____cacheline_internodealigned_in_smp;
>=20
> Although there is no documented rule that limits the change of the KABI
> struct irq_desc, it is still better to keep the irq_desc definition stabl=
e.

On what grounds? There is no such thing as a stable in-kernel ABI,
specially for things as internal as irq_desc.

> > =20
> >  #ifdef CONFIG_SPARSE_IRQ
> > diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
> > index 49e7bc871fec..2eac5532c3c8 100644
> > --- a/kernel/irq/chip.c
> > +++ b/kernel/irq/chip.c
> > @@ -306,6 +306,7 @@ static void __irq_disable(struct irq_desc *desc, bo=
ol mask);
> >  void irq_shutdown(struct irq_desc *desc)
> >  {
> >  	if (irqd_is_started(&desc->irq_data)) {
> > +		clear_irq_resend(desc);
> >  		desc->depth =3D 1;
> >  		if (desc->irq_data.chip->irq_shutdown) {
> >  			desc->irq_data.chip->irq_shutdown(&desc->irq_data);
> > diff --git a/kernel/irq/internals.h b/kernel/irq/internals.h
> > index 5fdc0b557579..51fc8c497c22 100644
> > --- a/kernel/irq/internals.h
> > +++ b/kernel/irq/internals.h
> > @@ -113,6 +113,8 @@ irqreturn_t handle_irq_event(struct irq_desc *desc);
> > =20
> >  /* Resending of interrupts :*/
> >  int check_irq_resend(struct irq_desc *desc, bool inject);
> > +void clear_irq_resend(struct irq_desc *desc);
> > +void irq_resend_init(struct irq_desc *desc);
> >  bool irq_wait_for_poll(struct irq_desc *desc);
> >  void __irq_wake_thread(struct irq_desc *desc, struct irqaction *action=
);
> > =20
> > diff --git a/kernel/irq/irqdesc.c b/kernel/irq/irqdesc.c
> > index 240e145e969f..b401b89b226a 100644
> > --- a/kernel/irq/irqdesc.c
> > +++ b/kernel/irq/irqdesc.c
> > @@ -415,6 +415,7 @@ static struct irq_desc *alloc_desc(int irq, int nod=
e, unsigned int flags,
> >  	desc_set_defaults(irq, desc, node, affinity, owner);
> >  	irqd_set(&desc->irq_data, flags);
> >  	kobject_init(&desc->kobj, &irq_kobj_type);
> > +	irq_resend_init(desc);
> > =20
> >  	return desc;
> > =20
> > @@ -581,6 +582,7 @@ int __init early_irq_init(void)
> >  		mutex_init(&desc[i].request_mutex);
> >  		init_waitqueue_head(&desc[i].wait_for_threads);
> >  		desc_set_defaults(i, &desc[i], node, NULL, NULL);
> > +		irq_resend_init(desc);
> >  	}
> >  	return arch_early_irq_init();
> >  }
> > diff --git a/kernel/irq/resend.c b/kernel/irq/resend.c
> > index 0c46e9fe3a89..edec335c0a7a 100644
> > --- a/kernel/irq/resend.c
> > +++ b/kernel/irq/resend.c
> > @@ -21,8 +21,9 @@
> > =20
> >  #ifdef CONFIG_HARDIRQS_SW_RESEND
> > =20
> > -/* Bitmap to handle software resend of interrupts: */
> > -static DECLARE_BITMAP(irqs_resend, IRQ_BITMAP_BITS);
> > +/* hlist_head to handle software resend of interrupts: */
> > +static HLIST_HEAD(irq_resend_list);
> > +static DEFINE_RAW_SPINLOCK(irq_resend_lock);
>=20
> What is the benefit of using hlist here? If you want to enjoy the
> low latency of querying elements by key, you must define a hlist table
> with a reasonable number of buckets. Otherwise, I don't think the time
> complexity of hlist is better than a regular double-linked list, right?

You do realise that the list is processed in order, one element after
the other, without ever querying any arbitrary element? Have you read
the code?

>
> > =20
> >  /*
> >   * Run software resends of IRQ's
> > @@ -30,18 +31,17 @@ static DECLARE_BITMAP(irqs_resend, IRQ_BITMAP_BITS);
> >  static void resend_irqs(struct tasklet_struct *unused)
> >  {
> >  	struct irq_desc *desc;
> > -	int irq;
> > -
> > -	while (!bitmap_empty(irqs_resend, nr_irqs)) {
> > -		irq =3D find_first_bit(irqs_resend, nr_irqs);
> > -		clear_bit(irq, irqs_resend);
> > -		desc =3D irq_to_desc(irq);
> > -		if (!desc)
> > -			continue;
> > -		local_irq_disable();
> > +
> > +	raw_spin_lock_irq(&irq_resend_lock);
> > +	while (!hlist_empty(&irq_resend_list)) {
> > +		desc =3D hlist_entry(irq_resend_list.first, struct irq_desc,
> > +				   resend_node);
> > +		hlist_del_init(&desc->resend_node);
> > +		raw_spin_unlock(&irq_resend_lock);
> >  		desc->handle_irq(desc);
> > -		local_irq_enable();
> > +		raw_spin_lock(&irq_resend_lock);
> >  	}
> > +	raw_spin_unlock_irq(&irq_resend_lock);
> >  }
> > =20
> >  /* Tasklet to handle resend: */
> > @@ -49,8 +49,6 @@ static DECLARE_TASKLET(resend_tasklet, resend_irqs);
> > =20
> >  static int irq_sw_resend(struct irq_desc *desc)
> >  {
> > -	unsigned int irq =3D irq_desc_get_irq(desc);
> > -
> >  	/*
> >  	 * Validate whether this interrupt can be safely injected from
> >  	 * non interrupt context
> > @@ -70,16 +68,31 @@ static int irq_sw_resend(struct irq_desc *desc)
> >  		 */
> >  		if (!desc->parent_irq)
> >  			return -EINVAL;
> > -		irq =3D desc->parent_irq;
>=20
> Why delete this code?

OK, now I know you haven't read this code at all :-(.

>=20
> >  	}
> > =20
> > -	/* Set it pending and activate the softirq: */
> > -	set_bit(irq, irqs_resend);
> > +	/* Add to resend_list and activate the softirq: */
> > +	raw_spin_lock(&irq_resend_lock);
> > +	hlist_add_head(&desc->resend_node, &irq_resend_list);
> > +	raw_spin_unlock(&irq_resend_lock);
>=20
> Do you conside a situation where irq_sw_resend() is running on two CPUs c=
oncurrently?
> If so, the same desc could be added into irq_resend_list twice by mistake.

Have you looked at the calling site (stress on singular), the locking
requirements, and the role IRQS_REPLAY plays when it comes to queuing
an interrupt for resend?

	M.

--=20
Without deviation from the norm, progress is not possible.