From: Sumit Saxena <sumit.saxena@broadcom.com>
To: Thomas Gleixner <tglx@linutronix.de>
Cc: Liyang Dou <douliyangs@gmail.com>,
LKML <linux-kernel@vger.kernel.org>,
linux-pci@vger.kernel.org,
Kashyap Desai <kashyap.desai@broadcom.com>,
Shivasharan Srikanteshwara
<shivasharan.srikanteshwara@broadcom.com>,
Ming Lei <ming.lei@redhat.com>, Christoph Hellwig <hch@lst.de>,
Bjorn Helgaas <bhelgaas@google.com>,
douliyang1@huawei.com
Subject: Re: [PATCH 0/3] irq/core: Fix and expand the irq affinity descriptor
Date: Fri, 28 Dec 2018 15:24:04 +0530 [thread overview]
Message-ID: <CAL2rwxrVt2o=HE5Q-reFDp8N3i0psmbN6TbhsWuL-zha_QUeSg@mail.gmail.com> (raw)
In-Reply-To: <CAL2rwxq8kKMtPY6SvWzMCUUfYGSJyjjSFCKu4zvLpRyWvQtWNw@mail.gmail.com>
On Wed, Dec 19, 2018 at 6:25 PM Sumit Saxena <sumit.saxena@broadcom.com> wrote:
>
> On Wed, Dec 19, 2018 at 4:23 PM Thomas Gleixner <tglx@linutronix.de> wrote:
> >
> > On Tue, 4 Dec 2018, Dou Liyang wrote:
> >
> > > Now, Spreading the interrupt affinity info by a cpumask pointer is not
> > > enough, meets a problem[1] and hard to expand in the future.
> > >
> > > Fix it by:
> > >
> > > +-----------------------------------+
> > > | |
> > > | struct cpumask *affinity |
> > > | |
> > > +-----------------------------------+
> > > |
> > > +------------------v-------------------+
> > > | |
> > > | struct irq_affinity_desc { |
> > > | struct cpumask mask; |
> > > | unsigned int is_managed : 1; |
> > > | }; |
> > > | |
> > > +--------------------------------------+
> > >
> >
> > So, I've applied that lot for 4.21 (or whatever number it will be). That's
> > only the first step for solving Kashyap's problem.
> >
> > IIRC, then Kashap wanted to get initial interrupt spreading for these extra
> > magic interrupts as well, but not have them marked managed.
> >
> > That's trivial to do now with the two queued changes in that area:
> >
> > - The rework above
> >
> > - The support for interrupt sets from Jens
> >
> > Just adding a small bitfield to struct irq_affinity which allows to tell
> > the core that a particular interrupt set is not managed does the trick.
> >
> > Untested patch below.
> >
> > Kashyap, is that what you were looking for and if so, does it work?
> Thomas,
> We could not test these patches as they did net get applied to latest
> linux-block tree cleanly.
>
> Our requirement is: 1. extra interrupts should be un-managed and 2.
> should be spread to CPUs of local NUMA node.
> If interrupts are un-managed but not spread as per our requirement,
> then still driver/userspace apps can manage by spreading
> them as required by calling API- irq_set_affinity_hint().
>
> Thanks,
> Sumit
I tested this patchset with some minor rework to apply it on latest
linux block tree(4.20-rc7).
It worked as our expectation. For "pre_vectors" IRQs(extra set of
interrupts), "is_managed" flag is set to 0
and later driver can affine these "pre_vectors" to CPUs of local NUMA
node through API- irq_set_affinity_hint().
Regular set of interrupts(not pre_vectors/post_vectors) are managed,
"is_managed" set to 1.
Below are some data from my test setup-
# numactl --hardware
available: 2 nodes (0-1)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 36 37 38 39
40 41 42 43 44 45 46 47 48 49 50 51 52 53
node 0 size: 31822 MB
node 0 free: 30241 MB
node 1 cpus: 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 54
55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71
node 1 size: 32248 MB
node 1 free: 31960 MB
node distances:
node 0 1
0: 10 21
1: 21 10
MegaRAID Controller(PCI device 86:00.0) is attached to node 1.
# find /sys -name *numa_node* | grep "86:00" | xargs cat
1
IRQ-CPU affinity of extra 16 interrupts for PCI device 86:00.0:
irq 149, cpu list 18-35,54-71
irq 150, cpu list 18-35,54-71
irq 151, cpu list 18-35,54-71
irq 152, cpu list 18-35,54-71
irq 153, cpu list 18-35,54-71
irq 154, cpu list 18-35,54-71
irq 155, cpu list 18-35,54-71
irq 156, cpu list 18-35,54-71
irq 157, cpu list 18-35,54-71
irq 158, cpu list 18-35,54-71
irq 159, cpu list 18-35,54-71
irq 160, cpu list 18-35,54-71
irq 161, cpu list 18-35,54-71
irq 162, cpu list 18-35,54-71
irq 163, cpu list 18-35,54-71
irq 164, cpu list 18-35,54-71
---
# cat /sys/kernel/debug/irq/irqs/164 | grep is_managed
is_managed: 0
Tested-by: Sumit Saxena <sumit.saxena@broadcom.com>
> >
> > Thanks,
> >
> > tglx
> >
> > 8<-----------------
> >
> > Subject: genirq/affinity: Add support for non-managed affinity sets
> > From: Thomas Gleixner <tglx@linutronix.de>
> > Date: Tue, 18 Dec 2018 16:46:47 +0100
> >
> > Some drivers need an extra set of interrupts which are not marked managed,
> > but should get initial interrupt spreading.
> >
> > Add a bitmap to struct irq_affinity which allows the driver to mark a
> > particular set of interrupts as non managed. Check the bitmap during
> > spreading and use the result to mark the interrupts in the sets
> > accordingly.
> >
> > The unmanaged interrupts get initial spreading, but user space can change
> > their affinity later on.
> >
> > Usage example:
> >
> > struct irq_affinity affd = { .pre_vectors = 2 };
> > int sets[2];
> >
> > /* Fill in sets[] */
> >
> > affd.nr_sets = 2;
> > affd.sets = &sets;
> > affd.unmanaged_sets = 0x02;
> >
> > ......
> >
> > So both sets are properly spread out, but the second set is not marked
> > managed.
> >
> > Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> > ---
> > include/linux/interrupt.h | 10 ++++++----
> > kernel/irq/affinity.c | 24 ++++++++++++++----------
> > 2 files changed, 20 insertions(+), 14 deletions(-)
> >
> > --- a/kernel/irq/affinity.c
> > +++ b/kernel/irq/affinity.c
> > @@ -99,7 +99,8 @@ static int __irq_build_affinity_masks(co
> > cpumask_var_t *node_to_cpumask,
> > const struct cpumask *cpu_mask,
> > struct cpumask *nmsk,
> > - struct irq_affinity_desc *masks)
> > + struct irq_affinity_desc *masks,
> > + bool managed)
> > {
> > int n, nodes, cpus_per_vec, extra_vecs, done = 0;
> > int last_affv = firstvec + numvecs;
> > @@ -154,6 +155,7 @@ static int __irq_build_affinity_masks(co
> > }
> > irq_spread_init_one(&masks[curvec].mask, nmsk,
> > cpus_per_vec);
> > + masks[curvec].is_managed = managed;
> > }
> >
> > done += v;
> > @@ -176,7 +178,8 @@ static int __irq_build_affinity_masks(co
> > static int irq_build_affinity_masks(const struct irq_affinity *affd,
> > int startvec, int numvecs, int firstvec,
> > cpumask_var_t *node_to_cpumask,
> > - struct irq_affinity_desc *masks)
> > + struct irq_affinity_desc *masks,
> > + bool managed)
> > {
> > int curvec = startvec, nr_present, nr_others;
> > int ret = -ENOMEM;
> > @@ -196,7 +199,8 @@ static int irq_build_affinity_masks(cons
> > /* Spread on present CPUs starting from affd->pre_vectors */
> > nr_present = __irq_build_affinity_masks(affd, curvec, numvecs,
> > firstvec, node_to_cpumask,
> > - cpu_present_mask, nmsk, masks);
> > + cpu_present_mask, nmsk, masks,
> > + managed);
> >
> > /*
> > * Spread on non present CPUs starting from the next vector to be
> > @@ -211,7 +215,7 @@ static int irq_build_affinity_masks(cons
> > cpumask_andnot(npresmsk, cpu_possible_mask, cpu_present_mask);
> > nr_others = __irq_build_affinity_masks(affd, curvec, numvecs,
> > firstvec, node_to_cpumask,
> > - npresmsk, nmsk, masks);
> > + npresmsk, nmsk, masks, managed);
> > put_online_cpus();
> >
> > if (nr_present < numvecs)
> > @@ -268,10 +272,11 @@ irq_create_affinity_masks(int nvecs, con
> >
> > for (i = 0, usedvecs = 0; i < nr_sets; i++) {
> > int this_vecs = affd->sets ? affd->sets[i] : affvecs;
> > + bool managed = !test_bit(i, &affd->unmanaged_sets);
> > int ret;
> >
> > - ret = irq_build_affinity_masks(affd, curvec, this_vecs,
> > - curvec, node_to_cpumask, masks);
> > + ret = irq_build_affinity_masks(affd, curvec, this_vecs, curvec,
> > + node_to_cpumask, masks, managed);
> > if (ret) {
> > kfree(masks);
> > masks = NULL;
> > @@ -289,10 +294,6 @@ irq_create_affinity_masks(int nvecs, con
> > for (; curvec < nvecs; curvec++)
> > cpumask_copy(&masks[curvec].mask, irq_default_affinity);
> >
> > - /* Mark the managed interrupts */
> > - for (i = affd->pre_vectors; i < nvecs - affd->post_vectors; i++)
> > - masks[i].is_managed = 1;
> > -
> > outnodemsk:
> > free_node_to_cpumask(node_to_cpumask);
> > return masks;
> > @@ -316,6 +317,9 @@ int irq_calc_affinity_vectors(int minvec
> > if (affd->nr_sets) {
> > int i;
> >
> > + if (WARN_ON_ONCE(affd->nr_sets > BITS_PER_LONG))
> > + return 0;
> > +
> > for (i = 0, set_vecs = 0; i < affd->nr_sets; i++)
> > set_vecs += affd->sets[i];
> > } else {
> > --- a/include/linux/interrupt.h
> > +++ b/include/linux/interrupt.h
> > @@ -249,12 +249,14 @@ struct irq_affinity_notify {
> > * the MSI(-X) vector space
> > * @nr_sets: Length of passed in *sets array
> > * @sets: Number of affinitized sets
> > + * @unmanaged_sets: Bitmap to mark members of @sets as unmanaged
> > */
> > struct irq_affinity {
> > - int pre_vectors;
> > - int post_vectors;
> > - int nr_sets;
> > - int *sets;
> > + int pre_vectors;
> > + int post_vectors;
> > + int nr_sets;
> > + int *sets;
> > + unsigned long unmanaged_sets;
> > };
> >
> > /**
prev parent reply other threads:[~2018-12-28 9:54 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-12-04 15:51 [PATCH 0/3] irq/core: Fix and expand the irq affinity descriptor Dou Liyang
2018-12-04 15:51 ` [PATCH 1/3] genirq/core: Add a new interrupt " Dou Liyang
2018-12-19 10:37 ` [tip:irq/core] genirq/core: Introduce struct irq_affinity_desc tip-bot for Dou Liyang
2018-12-04 15:51 ` [PATCH 2/3] irq/affinity: Add is_managed into " Dou Liyang
2018-12-18 15:26 ` Thomas Gleixner
2018-12-19 10:38 ` [tip:irq/core] genirq/affinity: Add is_managed to " tip-bot for Dou Liyang
2018-12-04 15:51 ` [PATCH 3/3] irq/affinity: Fix a possible breakage Dou Liyang
2018-12-05 8:28 ` Thomas Gleixner
2018-12-11 16:27 ` Dou Liyang
2018-12-19 10:53 ` [PATCH 0/3] irq/core: Fix and expand the irq affinity descriptor Thomas Gleixner
2018-12-19 12:55 ` Sumit Saxena
2018-12-28 9:54 ` Sumit Saxena [this message]
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='CAL2rwxrVt2o=HE5Q-reFDp8N3i0psmbN6TbhsWuL-zha_QUeSg@mail.gmail.com' \
--to=sumit.saxena@broadcom.com \
--cc=bhelgaas@google.com \
--cc=douliyang1@huawei.com \
--cc=douliyangs@gmail.com \
--cc=hch@lst.de \
--cc=kashyap.desai@broadcom.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-pci@vger.kernel.org \
--cc=ming.lei@redhat.com \
--cc=shivasharan.srikanteshwara@broadcom.com \
--cc=tglx@linutronix.de \
/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).