From: "Roger Pau Monné" <roger.pau@citrix.com>
To: Jan Beulich <jbeulich@suse.com>
Cc: Andrew Cooper <andrew.cooper3@citrix.com>, Wei Liu <wl@xen.org>,
<xen-devel@lists.xenproject.org>
Subject: Re: [PATCH v3 06/11] x86/hvm: allowing registering EOI callbacks for GSIs
Date: Wed, 7 Apr 2021 19:08:06 +0200 [thread overview]
Message-ID: <YG3m9nW6xPeL7MPr@Air-de-Roger> (raw)
In-Reply-To: <a57f0f75-341d-e6e1-823c-2083184a8f08@suse.com>
On Wed, Apr 07, 2021 at 05:51:14PM +0200, Jan Beulich wrote:
> On 31.03.2021 12:32, Roger Pau Monne wrote:
> > --- a/xen/arch/x86/hvm/irq.c
> > +++ b/xen/arch/x86/hvm/irq.c
> > +void hvm_gsi_execute_callbacks(unsigned int gsi)
> > +{
> > + struct hvm_irq *hvm_irq = hvm_domain_irq(current->domain);
> > + struct hvm_gsi_eoi_callback *cb;
> > +
> > + read_lock(&hvm_irq->gsi_callbacks_lock);
> > + list_for_each_entry ( cb, &hvm_irq->gsi_callbacks[gsi], list )
> > + cb->callback(gsi, cb->data);
> > + read_unlock(&hvm_irq->gsi_callbacks_lock);
> > +}
>
> Just as an observation (for now at least) - holding the lock here
> means the callbacks cannot re-register themselves.
Well, re-registering would be weird, as the callback is not
unregistered after execution. What is likely more relevant is that the
callback cannot unregister itself. I haven't found a need for this so
far, so I think it's fine.
> > +bool hvm_gsi_has_callbacks(const struct domain *d, unsigned int gsi)
> > +{
> > + struct hvm_irq *hvm_irq = hvm_domain_irq(d);
> > + bool has_callbacks;
> > +
> > + read_lock(&hvm_irq->gsi_callbacks_lock);
> > + has_callbacks = !list_empty(&hvm_irq->gsi_callbacks[gsi]);
> > + read_unlock(&hvm_irq->gsi_callbacks_lock);
> > +
> > + return has_callbacks;
> > +}
>
> What use is this function? Its result is stale by the time the
> caller can look at it, as you've dropped the lock.
Right, that function is only used to decide whether the vIOAPIC needs
to register an EOI callback when injecting a vector to the vlapic. The
workflow is to first register a callback with the vIOAPIC and
afterwards inject an interrupt which will trigger the callback
logic.
Playing with the callback registration while interrupts can be
injected will likely result in a malfunction of the device that relies
on those callbacks, but that's to be expected anyway when playing such
games.
That said multiple users sharing a vIOAPIC pin should be fine as long
as they follow the logic above: always register a callback before
attempting to inject an interrupt.
> > @@ -421,13 +423,25 @@ static void eoi_callback(unsigned int vector, void *data)
> > if ( is_iommu_enabled(d) )
> > {
> > spin_unlock(&d->arch.hvm.irq_lock);
> > - hvm_dpci_eoi(vioapic->base_gsi + pin);
> > + hvm_dpci_eoi(gsi);
> > spin_lock(&d->arch.hvm.irq_lock);
> > }
> >
> > + /*
> > + * Callbacks don't expect to be executed with any lock held, so
> > + * drop the lock that protects the vIO-APIC fields from changing.
> > + *
> > + * Note that the redirection entry itself cannot go away, so upon
> > + * retaking the lock we only need to avoid making assumptions on
> > + * redirection entry field values (ie: recheck the IRR field).
> > + */
> > + spin_unlock(&d->arch.hvm.irq_lock);
> > + hvm_gsi_execute_callbacks(gsi);
> > + spin_lock(&d->arch.hvm.irq_lock);
>
> The two pairs of unlock / re-lock want folding, I think - there's
> no point causing extra contention on the lock here.
The chunk above will go away on the next patch - there's no need to
fold it as it makes the following patch less clear.
> > @@ -443,7 +457,8 @@ static void ioapic_inj_irq(
> > struct vlapic *target,
> > uint8_t vector,
> > uint8_t trig_mode,
> > - uint8_t delivery_mode)
> > + uint8_t delivery_mode,
> > + bool callback)
> > {
> > HVM_DBG_LOG(DBG_LEVEL_IOAPIC, "irq %d trig %d deliv %d",
> > vector, trig_mode, delivery_mode);
> > @@ -452,7 +467,7 @@ static void ioapic_inj_irq(
> > (delivery_mode == dest_LowestPrio));
> >
> > vlapic_set_irq_callback(target, vector, trig_mode,
> > - trig_mode ? eoi_callback : NULL, NULL);
> > + callback ? eoi_callback : NULL, NULL);
>
> I think you'd better use trig_mode || callback here and ...
>
> > @@ -466,6 +481,7 @@ static void vioapic_deliver(struct hvm_vioapic *vioapic, unsigned int pin)
> > struct vlapic *target;
> > struct vcpu *v;
> > unsigned int irq = vioapic->base_gsi + pin;
> > + bool callback = trig_mode || hvm_gsi_has_callbacks(d, irq);
> >
> > ASSERT(spin_is_locked(&d->arch.hvm.irq_lock));
> >
> > @@ -492,7 +508,8 @@ static void vioapic_deliver(struct hvm_vioapic *vioapic, unsigned int pin)
> > target = vlapic_lowest_prio(d, NULL, 0, dest, dest_mode);
> > if ( target != NULL )
> > {
> > - ioapic_inj_irq(vioapic, target, vector, trig_mode, delivery_mode);
> > + ioapic_inj_irq(vioapic, target, vector, trig_mode, delivery_mode,
> > + callback);
>
> ... invoke hvm_gsi_has_callbacks() right here and ...
>
> > @@ -507,7 +524,7 @@ static void vioapic_deliver(struct hvm_vioapic *vioapic, unsigned int pin)
> > for_each_vcpu ( d, v )
> > if ( vlapic_match_dest(vcpu_vlapic(v), NULL, 0, dest, dest_mode) )
> > ioapic_inj_irq(vioapic, vcpu_vlapic(v), vector, trig_mode,
> > - delivery_mode);
> > + delivery_mode, callback);
>
> ... here, avoiding to call the function when you don't need the
> result.
I think there's a slim chance of not needing to use the callback local
variable, and hence didn't consider limiting it. I can do, but I'm
unsure this will bring any real benefit while making the code more
complex IMO.
Thanks, Roger.
next prev parent reply other threads:[~2021-04-07 17:08 UTC|newest]
Thread overview: 51+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-03-31 10:32 [PATCH v3 00/11] x86/intr: introduce EOI callbacks and fix vPT Roger Pau Monne
2021-03-31 10:32 ` [PATCH v3 01/11] x86/hvm: drop vcpu parameter from vlapic EOI callbacks Roger Pau Monne
2021-03-31 16:02 ` Jan Beulich
2021-03-31 16:24 ` Andrew Cooper
2021-04-01 9:12 ` Roger Pau Monné
2021-04-01 11:06 ` Jan Beulich
2021-04-07 7:41 ` Roger Pau Monné
2021-04-07 8:19 ` Jan Beulich
2021-03-31 10:32 ` [PATCH v3 02/11] x86/hvm: drop domain parameter from vioapic/vpic " Roger Pau Monne
2021-03-31 16:04 ` Jan Beulich
2021-04-01 9:15 ` Roger Pau Monné
2021-04-01 9:28 ` Jan Beulich
2021-03-31 10:32 ` [PATCH v3 03/11] x86/vlapic: introduce an EOI callback mechanism Roger Pau Monne
2021-03-31 11:47 ` Andrew Cooper
2021-03-31 12:50 ` Roger Pau Monné
2021-04-07 14:55 ` Jan Beulich
2021-04-07 16:27 ` Roger Pau Monné
2021-04-08 6:20 ` Jan Beulich
2021-04-08 9:12 ` Roger Pau Monné
2021-04-08 10:49 ` Jan Beulich
2021-04-08 10:56 ` Roger Pau Monné
2021-03-31 10:32 ` [PATCH v3 04/11] x86/vmsi: use the newly introduced EOI callbacks Roger Pau Monne
2021-04-07 14:59 ` Jan Beulich
2021-03-31 10:32 ` [PATCH v3 05/11] x86/vioapic: switch to use the EOI callback mechanism Roger Pau Monne
2021-04-07 15:19 ` Jan Beulich
2021-04-07 16:46 ` Roger Pau Monné
2021-04-08 6:27 ` Jan Beulich
2021-04-08 8:59 ` Roger Pau Monné
2021-04-08 10:52 ` Jan Beulich
2021-03-31 10:32 ` [PATCH v3 06/11] x86/hvm: allowing registering EOI callbacks for GSIs Roger Pau Monne
2021-04-07 15:51 ` Jan Beulich
2021-04-07 17:08 ` Roger Pau Monné [this message]
2021-04-08 6:34 ` Jan Beulich
2021-04-15 16:04 ` Roger Pau Monné
2021-04-16 7:29 ` Jan Beulich
2021-04-19 8:31 ` Roger Pau Monné
2021-04-08 12:52 ` Roger Pau Monné
2021-04-08 14:31 ` Jan Beulich
2021-04-08 15:06 ` Roger Pau Monné
2021-03-31 10:32 ` [PATCH v3 07/11] x86/dpci: move code Roger Pau Monne
2021-03-31 10:33 ` [PATCH v3 08/11] x86/dpci: switch to use a GSI EOI callback Roger Pau Monne
2021-04-08 14:49 ` Jan Beulich
2021-04-08 15:23 ` Roger Pau Monné
2021-03-31 10:33 ` [PATCH v3 09/11] x86/vpt: switch interrupt injection model Roger Pau Monne
2021-04-14 10:28 ` Jan Beulich
2021-04-14 13:37 ` Roger Pau Monné
2021-04-14 14:05 ` Jan Beulich
2021-04-14 14:20 ` Roger Pau Monné
2021-03-31 10:33 ` [PATCH v3 10/11] x86/vpt: remove vPT timers per-vCPU lists Roger Pau Monne
2021-04-14 10:38 ` Jan Beulich
2021-03-31 10:33 ` [PATCH v3 11/11] x86/vpt: introduce a per-vPT lock Roger Pau Monne
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=YG3m9nW6xPeL7MPr@Air-de-Roger \
--to=roger.pau@citrix.com \
--cc=andrew.cooper3@citrix.com \
--cc=jbeulich@suse.com \
--cc=wl@xen.org \
--cc=xen-devel@lists.xenproject.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).