Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Laurent Vivier <lvivier@redhat.com>
To:     linux-kernel@vger.kernel.org
Cc:     linux-block@vger.kernel.org, Paul Mackerras <paulus@samba.org>,
        linux-pci@vger.kernel.org, Thomas Gleixner <tglx@linutronix.de>,
        Marc Zyngier <maz@kernel.org>, linuxppc-dev@lists.ozlabs.org,
        Benjamin Herrenschmidt <benh@kernel.crashing.org>,
        "Michael S . Tsirkin" <mst@redhat.com>, Greg Kurz <groug@kaod.org>,
        Christoph Hellwig <hch@lst.de>,
        Michael Ellerman <mpe@ellerman.id.au>,
        Laurent Vivier <lvivier@redhat.com>
Subject: [PATCH v2 0/2] powerpc/pseries: fix MSI/X IRQ affinity on pseries
Date:   Wed, 25 Nov 2020 12:16:55 +0100
Message-Id: <20201125111657.1141295-1-lvivier@redhat.com>
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

With virtio, in multiqueue case, each queue IRQ is normally=0D
bound to a different CPU using the affinity mask.=0D
=0D
This works fine on x86_64 but totally ignored on pseries.=0D
=0D
This is not obvious at first look because irqbalance is doing=0D
some balancing to improve that.=0D
=0D
It appears that the "managed" flag set in the MSI entry=0D
is never copied to the system IRQ entry.=0D
=0D
This series passes the affinity mask from rtas_setup_msi_irqs()=0D
to irq_domain_alloc_descs() by adding an affinity parameter to=0D
irq_create_mapping().=0D
=0D
The first patch adds the parameter (no functional change), the=0D
second patch passes the actual affinity mask to irq_create_mapping()=0D
in rtas_setup_msi_irqs().=0D
=0D
For instance, with 32 CPUs VM and 32 queues virtio-scsi interface:=0D
=0D
... -smp 32 -device virtio-scsi-pci,id=3Dvirtio_scsi_pci0,num_queues=3D32=0D
=0D
for IRQ in $(grep virtio2-request /proc/interrupts |cut -d: -f1); do=0D
    for file in /proc/irq/$IRQ/ ; do=0D
        echo -n "IRQ: $(basename $file) CPU: " ; cat $file/smp_affinity_lis=
t=0D
    done=0D
done=0D
=0D
Without the patch (and without irqbalanced)=0D
=0D
IRQ: 268 CPU: 0-31=0D
IRQ: 269 CPU: 0-31=0D
IRQ: 270 CPU: 0-31=0D
IRQ: 271 CPU: 0-31=0D
IRQ: 272 CPU: 0-31=0D
IRQ: 273 CPU: 0-31=0D
IRQ: 274 CPU: 0-31=0D
IRQ: 275 CPU: 0-31=0D
IRQ: 276 CPU: 0-31=0D
IRQ: 277 CPU: 0-31=0D
IRQ: 278 CPU: 0-31=0D
IRQ: 279 CPU: 0-31=0D
IRQ: 280 CPU: 0-31=0D
IRQ: 281 CPU: 0-31=0D
IRQ: 282 CPU: 0-31=0D
IRQ: 283 CPU: 0-31=0D
IRQ: 284 CPU: 0-31=0D
IRQ: 285 CPU: 0-31=0D
IRQ: 286 CPU: 0-31=0D
IRQ: 287 CPU: 0-31=0D
IRQ: 288 CPU: 0-31=0D
IRQ: 289 CPU: 0-31=0D
IRQ: 290 CPU: 0-31=0D
IRQ: 291 CPU: 0-31=0D
IRQ: 292 CPU: 0-31=0D
IRQ: 293 CPU: 0-31=0D
IRQ: 294 CPU: 0-31=0D
IRQ: 295 CPU: 0-31=0D
IRQ: 296 CPU: 0-31=0D
IRQ: 297 CPU: 0-31=0D
IRQ: 298 CPU: 0-31=0D
IRQ: 299 CPU: 0-31=0D
=0D
With the patch:=0D
=0D
IRQ: 265 CPU: 0=0D
IRQ: 266 CPU: 1=0D
IRQ: 267 CPU: 2=0D
IRQ: 268 CPU: 3=0D
IRQ: 269 CPU: 4=0D
IRQ: 270 CPU: 5=0D
IRQ: 271 CPU: 6=0D
IRQ: 272 CPU: 7=0D
IRQ: 273 CPU: 8=0D
IRQ: 274 CPU: 9=0D
IRQ: 275 CPU: 10=0D
IRQ: 276 CPU: 11=0D
IRQ: 277 CPU: 12=0D
IRQ: 278 CPU: 13=0D
IRQ: 279 CPU: 14=0D
IRQ: 280 CPU: 15=0D
IRQ: 281 CPU: 16=0D
IRQ: 282 CPU: 17=0D
IRQ: 283 CPU: 18=0D
IRQ: 284 CPU: 19=0D
IRQ: 285 CPU: 20=0D
IRQ: 286 CPU: 21=0D
IRQ: 287 CPU: 22=0D
IRQ: 288 CPU: 23=0D
IRQ: 289 CPU: 24=0D
IRQ: 290 CPU: 25=0D
IRQ: 291 CPU: 26=0D
IRQ: 292 CPU: 27=0D
IRQ: 293 CPU: 28=0D
IRQ: 294 CPU: 29=0D
IRQ: 295 CPU: 30=0D
IRQ: 299 CPU: 31=0D
=0D
This matches what we have on an x86_64 system.=0D
=0D
v2: add a wrapper around original irq_create_mapping() with the=0D
    affinity parameter. Update comments=0D
=0D
Laurent Vivier (2):=0D
  genirq: add an irq_create_mapping_affinity() function=0D
  powerpc/pseries: pass MSI affinity to irq_create_mapping()=0D
=0D
 arch/powerpc/platforms/pseries/msi.c |  3 ++-=0D
 include/linux/irqdomain.h            | 12 ++++++++++--=0D
 kernel/irq/irqdomain.c               | 13 ++++++++-----=0D
 3 files changed, 20 insertions(+), 8 deletions(-)=0D
=0D
-- =0D
2.28.0=0D
=0D