Skip to contentSkip to conversation

net: Fixup cpu_mask usage

PATCH In review [email protected] 5 participants 21 messages Oct 14, 2022 – Oct 17, 2022

Subsystems (3)
lore
Share & tools
lore
#1 lore raw

PATCH v2 1/2 net: Fixup netif_attrmask_next_and warning

In reply to [email protected]

From: Guo Ren <[email protected]>

Don't pass nr_bits as arg1, cpu_max_bits_warn would cause warning now 854701ba4c39 ("net: fix cpu_max_bits_warn() usage in netif_attrmask_next{,_and}").

------------[ cut here ]------------
WARNING: CPU: 2 PID: 1 at include/linux/cpumask.h:110 __netif_set_xps_queue+0x14e/0x770
Modules linked in:
CPU: 2 PID: 1 Comm: swapper/0 Not tainted 6.0.0-rc4-00018-g854701ba4c39 #324
Hardware name: riscv-virtio,qemu (DT)
epc : __netif_set_xps_queue+0x14e/0x770
 ra : __netif_set_xps_queue+0x552/0x770

epc : ffffffff806fe448 ra : ffffffff806fe84c sp : ff600000023279d0
gp : ffffffff815fff88 tp : ff600000023a0000 t0 : ff6000000308ab40 t1 : 0000000000000003 t2 : 0000000000000000 s0 : ff60000002327a90 s1 : 0000000000000000 a0 : ff6000000308ab00 a1 : ff6000000308ab00 a2 : ff6000000308a8e8 a3 : 0000000000000004 a4 : 0000000000000000 a5 : 0000000000000000 a6 : 0000000000000000 a7 : 0000000000000000 s2 : 0000000000000000 s3 : 0000000000000000 s4 : ff60000002327aa0 s5 : ffffffff816031c8 s6 : 0000000000000000 s7 : 0000000000000001 s8 : 0000000000000000 s9 : 0000000000000004 s10: ff6000000308a8c0
s11: 0000000000000004 t3 : 0000000000000000 t4 : 0000000000000014 t5 : 0000000000000000 t6 : 0000000000000000
status: 0000000200000120 badaddr: 0000000000000000 cause: 0000000000000003

[<ffffffff805e5824>] virtnet_set_affinity+0x14a/0x1c0
[<ffffffff805e7b04>] virtnet_probe+0x7fc/0xee2
[<ffffffff8050e120>] virtio_dev_probe+0x164/0x2de
[<ffffffff8055b69e>] really_probe+0x82/0x224
[<ffffffff8055b89a>] __driver_probe_device+0x5a/0xaa
[<ffffffff8055b916>] driver_probe_device+0x2c/0xb8
[<ffffffff8055bf34>] __driver_attach+0x76/0x108
[<ffffffff805597c0>] bus_for_each_dev+0x4a/0x8e
[<ffffffff8055b072>] driver_attach+0x1a/0x28
[<ffffffff8055ab8c>] bus_add_driver+0x13c/0x1a6
[<ffffffff8055c722>] driver_register+0x4a/0xfc
[<ffffffff8050dc34>] register_virtio_driver+0x1c/0x2c
[<ffffffff80a2bae4>] virtio_net_driver_init+0x7a/0xb0
[<ffffffff80002840>] do_one_initcall+0x66/0x2e4
[<ffffffff80a01212>] kernel_init_freeable+0x28a/0x304
[<ffffffff808b21e2>] kernel_init+0x1e/0x110
[<ffffffff80003c46>] ret_from_exception+0x0/0x10
---[ end trace 0000000000000000 ]---
Fixes 80d19669ecd3 ("net: Refactor XPS for CPUs and Rx queues")
---
 net/core/dev.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
net/core/dev.c +2 -2
index fa53830d0683..9ec8b10ae329 100644
@@ -2589,8 +2589,8 @@ int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
 		copy = true;
 
 	/* allocate memory for queue storage */
-	for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
-	     j < nr_ids;) {
+	for (j = -1; j < nr_ids;
+	     j = netif_attrmask_next_and(j, online_mask, mask, nr_ids)) {
 		if (!new_dev_maps) {
 			new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
 			if (!new_dev_maps) {
Signature · 1 line
2.36.1
#0 lore raw

v2 cover net: Fixup cpu_mask usage

From: Guo Ren <[email protected]>

Yury Norov repaired cpumask_check() in [1], which eveals the problem in other code. The series fixup them one by one clearly.

[1] https://lore.kernel.org/all/[email protected]/

Changes:
v2:
- Fixup fixes tag in commit log with Yury Norov advice.

v1:
https://lore.kernel.org/all/[email protected]/
https://lore.kernel.org/all/[email protected]/

Guo Ren (2):
net: Fixup netif_attrmask_next_and warning
net: Fixup virtnet_set_affinity() cause cpumask warning

 drivers/net/virtio_net.c | 2 ++
 net/core/dev.c           | 4 ++--
 2 files changed, 4 insertions(+), 2 deletions(-)
Signature · 1 line
2.36.1
#15 lore raw

PATCH v2 2/2 net: Fixup virtnet_set_affinity() cause cpumask warning

From: Guo Ren <[email protected]>

Don't pass nr_bits-1 as arg1 for cpumask_next_wrap, which would cause warning now 78e5a3399421 ("cpumask: fix checking valid cpu range").

------------[ cut here ]------------
WARNING: CPU: 0 PID: 1 at include/linux/cpumask.h:110 cpumask_next_wrap+0x5c/0x80
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 6.0.0-11659-ge7e38f6cce55-dirty #328
Hardware name: riscv-virtio,qemu (DT)
epc : cpumask_next_wrap+0x5c/0x80
 ra : virtnet_set_affinity+0x1ba/0x1fc

epc : ffffffff808992ca ra : ffffffff805d84ca sp : ff60000002327a50
gp : ffffffff81602390 tp : ff600000023a0000 t0 : 5f74656e74726976 t1 : 0000000000000000 t2 : 735f74656e747269 s0 : ff60000002327a90 s1 : 0000000000000003 a0 : 0000000000000003 a1 : ffffffff816051c0 a2 : 0000000000000004 a3 : 0000000000000000 a4 : 0000000000000000 a5 : 0000000000000004 a6 : 0000000000000000 a7 : 0000000000000000 s2 : 0000000000000000 s3 : ffffffff816051c0 s4 : ffffffff8160224c s5 : 0000000000000004 s6 : 0000000000000004 s7 : 0000000000000000 s8 : 0000000000000003 s9 : ffffffff810aa398 s10: ffffffff80e97d20
s11: 0000000000000004 t3 : ffffffff819acc97 t4 : ffffffff819acc97 t5 : ffffffff819acc98 t6 : ff60000002327878
status: 0000000200000120 badaddr: 0000000000000000 cause: 0000000000000003

[<ffffffff805d84ca>] virtnet_set_affinity+0x1ba/0x1fc
[<ffffffff805da7ac>] virtnet_probe+0x832/0xf1e
[<ffffffff804fe61c>] virtio_dev_probe+0x164/0x2de
[<ffffffff8054c4c4>] really_probe+0x82/0x224
[<ffffffff8054c6c0>] __driver_probe_device+0x5a/0xaa
[<ffffffff8054c73c>] driver_probe_device+0x2c/0xb8
[<ffffffff8054cd66>] __driver_attach+0x76/0x108
[<ffffffff8054a482>] bus_for_each_dev+0x52/0x9a
[<ffffffff8054be8c>] driver_attach+0x1a/0x28
[<ffffffff8054b996>] bus_add_driver+0x154/0x1c2
[<ffffffff8054d592>] driver_register+0x52/0x108
[<ffffffff804fe120>] register_virtio_driver+0x1c/0x2c
[<ffffffff80a29142>] virtio_net_driver_init+0x7a/0xb0
[<ffffffff80002854>] do_one_initcall+0x66/0x2e4
[<ffffffff80a01222>] kernel_init_freeable+0x28a/0x304
[<ffffffff808cb1be>] kernel_init+0x1e/0x110
[<ffffffff80003c4e>] ret_from_exception+0x0/0x10
---[ end trace 0000000000000000 ]---
Fixes 2ca653d607ce ("virtio_net: Stripe queue affinities across cores.")
---
 drivers/net/virtio_net.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/net/virtio_net.c +2 -0
index 7106932c6f88..e4b56523b2b5 100644
@@ -2300,6 +2300,8 @@ static void virtnet_set_affinity(struct virtnet_info *vi)
 
 		for (j = 0; j < group_size; j++) {
 			cpumask_set_cpu(cpu, mask);
+			if (cpu == (nr_cpu_ids - 1))
+				break;
 			cpu = cpumask_next_wrap(cpu, cpu_online_mask,
 						nr_cpu_ids, false);
 		}
Signature · 1 line
2.36.1
maintainer
#2 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

In reply to [email protected]

net/core/dev.c int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, · line 2593

Quoted 4 lines · On Thu, 13 Oct 2022 23:04:58 -0400 [email protected] wrote:
net/core/dev.c +2 -2
-	for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
-	     j < nr_ids;) {
+	for (j = -1; j < nr_ids;
+	     j = netif_attrmask_next_and(j, online_mask, mask, nr_ids)) {

This does not look equivalent, have you tested it?

nr_ids is unsigned, doesn't it mean we'll never enter the loop?

Can we instead revert 854701ba4c and take the larger rework Yury has posted a week ago into net-next?

maintainer
#3 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

Quoted 2 lines · On Thu, 13 Oct 2022 20:35:44 -0700 Jakub Kicinski wrote:

Can we instead revert 854701ba4c and take the larger rework Yury has posted a week ago into net-next?

Oh, it was reposted today:

https://lore.kernel.org/all/[email protected]/

But we need a revert of 854701ba4c as well to cover the issue back up for 6.1, AFAIU.

#4 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

Quoted 10 lines · Jakub Kicinski
Quoted 2 lines · On Thu, 13 Oct 2022 20:35:44 -0700 Jakub Kicinski wrote:

Can we instead revert 854701ba4c and take the larger rework Yury has posted a week ago into net-next?

Oh, it was reposted today:

https://lore.kernel.org/all/[email protected]/

But we need a revert of 854701ba4c as well to cover the issue back up for 6.1, AFAIU.

The patch 854701ba4c is technically correct. I fixed most of warnings in advance, but nobody can foresee everything, right? I expected some noise, and now we have just a few things to fix. This is what for -rc releases exist, didn't they?

I suggest to keep the patch, because this is the only way to make cpumask_check()-related issues visible to people. If things will go as they go now, I expect that -rc3 will be clean from cpumask_check() warnings.

Thanks,
Yury

author
#10 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

In reply to Jakub Kicinski
Quoted 10 lines · Jakub Kicinski

net/core/dev.c int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, · line 2593

Quoted 4 lines · On Thu, 13 Oct 2022 23:04:58 -0400 [email protected] wrote:
net/core/dev.c +2 -2
-     for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
-          j < nr_ids;) {
+     for (j = -1; j < nr_ids;
+          j = netif_attrmask_next_and(j, online_mask, mask, nr_ids)) {

This does not look equivalent, have you tested it?

nr_ids is unsigned, doesn't it mean we'll never enter the loop?

Yes, you are right. Any unsigned int would break the result. (gdb) p (int)-1 < (int)2
$1 = 1
(gdb) p (int)-1 < (unsigned int)2
$2 = 0
(gdb) p (unsigned int)-1 < (int)2
$4 = 0

So it should be:
- for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
- j < nr_ids;) {
+ for (j = -1; j < (int)nr_ids;
+ j = netif_attrmask_next_and(j, online_mask, mask, nr_ids)) {

Right? Of cause, nr_ids couldn't be 0xffffffff (-1).

Quoted 3 lines

Can we instead revert 854701ba4c and take the larger rework Yury has posted a week ago into net-next?

Signature · 2 lines
Best Regards
 Guo Ren
author
#5 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

In reply to Yury Norov
Quoted 10 lines · Yury Norov
Quoted 8 lines · Jakub Kicinski
Quoted 2 lines · On Thu, 13 Oct 2022 20:35:44 -0700 Jakub Kicinski wrote:

Can we instead revert 854701ba4c and take the larger rework Yury has posted a week ago into net-next?

Oh, it was reposted today:

https://lore.kernel.org/all/[email protected]/

Quoted 7 lines
Quoted 2 lines

But we need a revert of 854701ba4c as well to cover the issue back up for 6.1, AFAIU.

The patch 854701ba4c is technically correct. I fixed most of warnings in advance, but nobody can foresee everything, right? I expected some noise, and now we have just a few things to fix. This is what for -rc releases exist, didn't they?

Your job is great, I just want to help with some fixes. Fixes them in -rc would be a good point.

Quoted 8 lines

I suggest to keep the patch, because this is the only way to make cpumask_check()-related issues visible to people. If things will go as they go now, I expect that -rc3 will be clean from cpumask_check() warnings.

Thanks,
Yury

Signature · 2 lines
Best Regards
 Guo Ren
#16 lore raw

Re: [PATCH V2 2/2] net: Fixup virtnet_set_affinity() cause cpumask warning

In reply to [email protected]
Quoted 9 lines · On Thu, Oct 13, 2022 at 11:04:59PM -0400, [email protected] wrote:

From: Guo Ren <[email protected]>

Don't pass nr_bits-1 as arg1 for cpumask_next_wrap, which would cause warning now 78e5a3399421 ("cpumask: fix checking valid cpu range").

------------[ cut here ]------------
WARNING: CPU: 0 PID: 1 at include/linux/cpumask.h:110 cpumask_next_wrap+0x5c/0x80
Modules linked in:

Submitting Patches document suggests to cut this huge warning to only what makes sense to see in the report.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#13 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

In reply to [email protected]
Quoted 9 lines · On Thu, Oct 13, 2022 at 11:04:58PM -0400, [email protected] wrote:

From: Guo Ren <[email protected]>

Don't pass nr_bits as arg1, cpu_max_bits_warn would cause warning now 854701ba4c39 ("net: fix cpu_max_bits_warn() usage in netif_attrmask_next{,_and}").

------------[ cut here ]------------
WARNING: CPU: 2 PID: 1 at include/linux/cpumask.h:110 __netif_set_xps_queue+0x14e/0x770
Modules linked in:

Submitting Patches documentation suggests to cut this to only what makes sense for the report.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
author
#14 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

Quoted 8 lines · Andy Shevchenko
Quoted 6 lines · On Thu, Oct 13, 2022 at 11:04:58PM -0400, [email protected] wrote:

From: Guo Ren <[email protected]>

Don't pass nr_bits as arg1, cpu_max_bits_warn would cause warning now 854701ba4c39 ("net: fix cpu_max_bits_warn() usage in netif_attrmask_next{,_and}").

Quoted 6 lines
Quoted 3 lines
------------[ cut here ]------------
WARNING: CPU: 2 PID: 1 at include/linux/cpumask.h:110 __netif_set_xps_queue+0x14e/0x770
Modules linked in:

Submitting Patches documentation suggests to cut this to only what makes sense for the report.

Right, thx for mentioning.

Quoted 6 lines

--
With Best Regards,
Andy Shevchenko

Signature · 2 lines
Best Regards
 Guo Ren
author
#17 lore raw

Re: [PATCH V2 2/2] net: Fixup virtnet_set_affinity() cause cpumask warning

In reply to Andy Shevchenko
Quoted 8 lines · Andy Shevchenko
Quoted 5 lines · On Thu, Oct 13, 2022 at 11:04:59PM -0400, [email protected] wrote:

From: Guo Ren <[email protected]>

Don't pass nr_bits-1 as arg1 for cpumask_next_wrap, which would cause warning now 78e5a3399421 ("cpumask: fix checking valid cpu range").

Quoted 6 lines
Quoted 3 lines
------------[ cut here ]------------
WARNING: CPU: 0 PID: 1 at include/linux/cpumask.h:110 cpumask_next_wrap+0x5c/0x80
Modules linked in:

Submitting Patches document suggests to cut this huge warning to only what makes sense to see in the report.

Okay

Quoted 6 lines

--
With Best Regards,
Andy Shevchenko

Signature · 2 lines
Best Regards
 Guo Ren
maintainer
#11 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

In reply to Guo Ren
Quoted 12 lines · On Fri, 14 Oct 2022 14:38:56 +0800 Guo Ren wrote:
Quoted 3 lines

This does not look equivalent, have you tested it?

nr_ids is unsigned, doesn't it mean we'll never enter the loop?

Yes, you are right. Any unsigned int would break the result. (gdb) p (int)-1 < (int)2
$1 = 1
(gdb) p (int)-1 < (unsigned int)2
$2 = 0
(gdb) p (unsigned int)-1 < (int)2
$4 = 0

Quoted 7 lines

So it should be:
- for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
- j < nr_ids;) {
+ for (j = -1; j < (int)nr_ids;
+ j = netif_attrmask_next_and(j, online_mask, mask, nr_ids)) {

Right? Of cause, nr_ids couldn't be 0xffffffff (-1).

No. You can't enter the loop with -1 as the iterator either. Let's move on.

maintainer
#6 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

In reply to Yury Norov
Quoted 10 lines · On Thu, 13 Oct 2022 21:42:41 -0700 Yury Norov wrote:
Quoted 6 lines

Oh, it was reposted today:

https://lore.kernel.org/all/[email protected]/

But we need a revert of 854701ba4c as well to cover the issue back up for 6.1, AFAIU.

The patch 854701ba4c is technically correct. I fixed most of warnings in advance, but nobody can foresee everything, right? I expected some noise, and now we have just a few things to fix.

I got 6 warnings booting my machine after pulling back from Linus (which included your patches in net for the first time). And that's not including the XPS and the virtio warning.

Quoted 6 lines

This is what for -rc releases exist, didn't they?

I suggest to keep the patch, because this is the only way to make cpumask_check()-related issues visible to people. If things will go as they go now, I expect that -rc3 will be clean from cpumask_check() warnings.

This sounds too close to saying that "it's okay for -rc1 to be broken". Why were your changes not in linux-next for a month before the merge window? :(

We will not be merging a refactoring series into net to silence an arguably over-eager warning. We need a minimal fix, Guo Ren's patches seem to miss the mark so I reckon the best use of everyone's time is to just drop the exposing patch and retry in -next 🤷

#7 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

Quoted 19 lines · Jakub Kicinski
Quoted 10 lines · On Thu, 13 Oct 2022 21:42:41 -0700 Yury Norov wrote:
Quoted 6 lines

Oh, it was reposted today:

https://lore.kernel.org/all/[email protected]/

But we need a revert of 854701ba4c as well to cover the issue back up for 6.1, AFAIU.

The patch 854701ba4c is technically correct. I fixed most of warnings in advance, but nobody can foresee everything, right? I expected some noise, and now we have just a few things to fix.

I got 6 warnings booting my machine after pulling back from Linus (which included your patches in net for the first time). And that's not including the XPS and the virtio warning.

Quoted 2 lines

This is what for -rc releases exist, didn't they?

Quoted 8 lines
Quoted 4 lines

I suggest to keep the patch, because this is the only way to make cpumask_check()-related issues visible to people. If things will go as they go now, I expect that -rc3 will be clean from cpumask_check() warnings.

This sounds too close to saying that "it's okay for -rc1 to be broken". Why were your changes not in linux-next for a month before the merge window? :(

They spent about a month in -next. Nobody cared.

Quoted 4 lines

We will not be merging a refactoring series into net to silence an arguably over-eager warning. We need a minimal fix, Guo Ren's patches seem to miss the mark so I reckon the best use of everyone's time is to just drop the exposing patch and retry in -next 🤷

If you prefer treating symptoms rather than the disease - I have nothing to add.

maintainer
#8 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

Quoted 7 lines · On Fri, 14 Oct 2022 09:16:01 -0700 Yury Norov wrote:
Quoted 4 lines

We will not be merging a refactoring series into net to silence an arguably over-eager warning. We need a minimal fix, Guo Ren's patches seem to miss the mark so I reckon the best use of everyone's time is to just drop the exposing patch and retry in -next 🤷

If you prefer treating symptoms rather than the disease - I have nothing to add.

I don't, but we may consider different things to be "the disease". Please do not insinuate that I don't care about fixing bugs.

What I can grok from the history and your commit messages is that you want to catch people who pass what you consider invalid inputs to the helpers, but nothing will crash/OOB access here, because the helper double checks that the input is < nr_bits.

So it's a nice cleanup and refactoring, sure, but not an urgent fix that needs to go to Linus ASAP.

If that's not what you're fixing please explain, I believe I already asked you to clarify before. And the commit message aren't exactly informative either.

author
#12 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

In reply to Jakub Kicinski
Quoted 20 lines · Jakub Kicinski
Quoted 18 lines · On Fri, 14 Oct 2022 14:38:56 +0800 Guo Ren wrote:
Quoted 3 lines

This does not look equivalent, have you tested it?

nr_ids is unsigned, doesn't it mean we'll never enter the loop?

Yes, you are right. Any unsigned int would break the result. (gdb) p (int)-1 < (int)2
$1 = 1
(gdb) p (int)-1 < (unsigned int)2
$2 = 0
(gdb) p (unsigned int)-1 < (int)2
$4 = 0

So it should be:
- for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
- j < nr_ids;) {
+ for (j = -1; j < (int)nr_ids;
+ j = netif_attrmask_next_and(j, online_mask, mask, nr_ids)) {

Quoted 4 lines
Quoted 1 line

Right? Of cause, nr_ids couldn't be 0xffffffff (-1).

No. You can't enter the loop with -1 as the iterator either. Let's move on.

Oops, how about the below:

     for (j = netif_attrmask_next_and(-1, online_mask, mask, nr_ids);
j < (int)nr_ids;
          j = netif_attrmask_next_and(j, online_mask, mask, nr_ids)) {
Signature · 2 lines
Best Regards
 Guo Ren
author
#9 lore raw

Re: [PATCH V2 1/2] net: Fixup netif_attrmask_next_and warning

In reply to Jakub Kicinski
Quoted 9 lines · Jakub Kicinski
Quoted 7 lines · On Thu, 13 Oct 2022 21:42:41 -0700 Yury Norov wrote:
Quoted 6 lines

Oh, it was reposted today:

https://lore.kernel.org/all/[email protected]/

But we need a revert of 854701ba4c as well to cover the issue back up for 6.1, AFAIU.

Quoted 7 lines
Quoted 3 lines

The patch 854701ba4c is technically correct. I fixed most of warnings in advance, but nobody can foresee everything, right? I expected some noise, and now we have just a few things to fix.

I got 6 warnings booting my machine after pulling back from Linus (which included your patches in net for the first time). And that's not including the XPS and the virtio warning.

Oh, that's a wide effect than we thought.

Quoted 16 lines
Quoted 7 lines

This is what for -rc releases exist, didn't they?

I suggest to keep the patch, because this is the only way to make cpumask_check()-related issues visible to people. If things will go as they go now, I expect that -rc3 will be clean from cpumask_check() warnings.

This sounds too close to saying that "it's okay for -rc1 to be broken". Why were your changes not in linux-next for a month before the merge window? :(

We will not be merging a refactoring series into net to silence an arguably over-eager warning. We need a minimal fix, Guo Ren's patches seem to miss the mark so I reckon the best use of everyone's time is to just drop the exposing patch and retry in -next 🤷

Signature · 2 lines
Best Regards
 Guo Ren
reviewer
#18 lore raw

Re: [PATCH V2 2/2] net: Fixup virtnet_set_affinity() cause cpumask warning

In reply to [email protected]
Quoted 56 lines · On Thu, 13 Oct 2022 23:04:59 -0400, [email protected] wrote:

From: Guo Ren <[email protected]>

Don't pass nr_bits-1 as arg1 for cpumask_next_wrap, which would cause warning now 78e5a3399421 ("cpumask: fix checking valid cpu range").

------------[ cut here ]------------
WARNING: CPU: 0 PID: 1 at include/linux/cpumask.h:110 cpumask_next_wrap+0x5c/0x80
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 6.0.0-11659-ge7e38f6cce55-dirty #328
Hardware name: riscv-virtio,qemu (DT)
epc : cpumask_next_wrap+0x5c/0x80
 ra : virtnet_set_affinity+0x1ba/0x1fc

epc : ffffffff808992ca ra : ffffffff805d84ca sp : ff60000002327a50
gp : ffffffff81602390 tp : ff600000023a0000 t0 : 5f74656e74726976 t1 : 0000000000000000 t2 : 735f74656e747269 s0 : ff60000002327a90 s1 : 0000000000000003 a0 : 0000000000000003 a1 : ffffffff816051c0 a2 : 0000000000000004 a3 : 0000000000000000 a4 : 0000000000000000 a5 : 0000000000000004 a6 : 0000000000000000 a7 : 0000000000000000 s2 : 0000000000000000 s3 : ffffffff816051c0 s4 : ffffffff8160224c s5 : 0000000000000004 s6 : 0000000000000004 s7 : 0000000000000000 s8 : 0000000000000003 s9 : ffffffff810aa398 s10: ffffffff80e97d20
s11: 0000000000000004 t3 : ffffffff819acc97 t4 : ffffffff819acc97 t5 : ffffffff819acc98 t6 : ff60000002327878
status: 0000000200000120 badaddr: 0000000000000000 cause: 0000000000000003

[<ffffffff805d84ca>] virtnet_set_affinity+0x1ba/0x1fc
[<ffffffff805da7ac>] virtnet_probe+0x832/0xf1e
[<ffffffff804fe61c>] virtio_dev_probe+0x164/0x2de
[<ffffffff8054c4c4>] really_probe+0x82/0x224
[<ffffffff8054c6c0>] __driver_probe_device+0x5a/0xaa
[<ffffffff8054c73c>] driver_probe_device+0x2c/0xb8
[<ffffffff8054cd66>] __driver_attach+0x76/0x108
[<ffffffff8054a482>] bus_for_each_dev+0x52/0x9a
[<ffffffff8054be8c>] driver_attach+0x1a/0x28
[<ffffffff8054b996>] bus_add_driver+0x154/0x1c2
[<ffffffff8054d592>] driver_register+0x52/0x108
[<ffffffff804fe120>] register_virtio_driver+0x1c/0x2c
[<ffffffff80a29142>] virtio_net_driver_init+0x7a/0xb0
[<ffffffff80002854>] do_one_initcall+0x66/0x2e4
[<ffffffff80a01222>] kernel_init_freeable+0x28a/0x304
[<ffffffff808cb1be>] kernel_init+0x1e/0x110
[<ffffffff80003c4e>] ret_from_exception+0x0/0x10
---[ end trace 0000000000000000 ]---
Fixes 2ca653d607ce ("virtio_net: Stripe queue affinities across cores.")
---
 drivers/net/virtio_net.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/net/virtio_net.c +0 -0
index 7106932c6f88..e4b56523b2b5 100644

drivers/net/virtio_net.c static void virtnet_set_affinity(struct virtnet_info *vi) · line 2304

Quoted 5 lines
drivers/net/virtio_net.c +2 -0
@@ -2300,6 +2300,8 @@ static void virtnet_set_affinity(struct virtnet_info *vi)
 
 		for (j = 0; j < group_size; j++) {
 			cpumask_set_cpu(cpu, mask);
+			if (cpu == (nr_cpu_ids - 1))
+				break;

The problem seems to be a problem with cpumask_next_wrap(), I'm not particularly sure.

But I think there is something wrong with your modification, which will cause subsequent queues to be bound to (nr_cpu_ids - 1).

Thanks.

Quoted 6 lines
drivers/net/virtio_net.c +0 -1
 			cpu = cpumask_next_wrap(cpu, cpu_online_mask,
 						nr_cpu_ids, false);
 		}
--

2.36.1

author
#19 lore raw

Re: [PATCH V2 2/2] net: Fixup virtnet_set_affinity() cause cpumask warning

Quoted 67 lines · Xuan Zhuo
Quoted 56 lines · On Thu, 13 Oct 2022 23:04:59 -0400, [email protected] wrote:

From: Guo Ren <[email protected]>

Don't pass nr_bits-1 as arg1 for cpumask_next_wrap, which would cause warning now 78e5a3399421 ("cpumask: fix checking valid cpu range").

------------[ cut here ]------------
WARNING: CPU: 0 PID: 1 at include/linux/cpumask.h:110 cpumask_next_wrap+0x5c/0x80
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 6.0.0-11659-ge7e38f6cce55-dirty #328
Hardware name: riscv-virtio,qemu (DT)
epc : cpumask_next_wrap+0x5c/0x80
 ra : virtnet_set_affinity+0x1ba/0x1fc

epc : ffffffff808992ca ra : ffffffff805d84ca sp : ff60000002327a50
gp : ffffffff81602390 tp : ff600000023a0000 t0 : 5f74656e74726976 t1 : 0000000000000000 t2 : 735f74656e747269 s0 : ff60000002327a90 s1 : 0000000000000003 a0 : 0000000000000003 a1 : ffffffff816051c0 a2 : 0000000000000004 a3 : 0000000000000000 a4 : 0000000000000000 a5 : 0000000000000004 a6 : 0000000000000000 a7 : 0000000000000000 s2 : 0000000000000000 s3 : ffffffff816051c0 s4 : ffffffff8160224c s5 : 0000000000000004 s6 : 0000000000000004 s7 : 0000000000000000 s8 : 0000000000000003 s9 : ffffffff810aa398 s10: ffffffff80e97d20
s11: 0000000000000004 t3 : ffffffff819acc97 t4 : ffffffff819acc97 t5 : ffffffff819acc98 t6 : ff60000002327878
status: 0000000200000120 badaddr: 0000000000000000 cause: 0000000000000003

[<ffffffff805d84ca>] virtnet_set_affinity+0x1ba/0x1fc
[<ffffffff805da7ac>] virtnet_probe+0x832/0xf1e
[<ffffffff804fe61c>] virtio_dev_probe+0x164/0x2de
[<ffffffff8054c4c4>] really_probe+0x82/0x224
[<ffffffff8054c6c0>] __driver_probe_device+0x5a/0xaa
[<ffffffff8054c73c>] driver_probe_device+0x2c/0xb8
[<ffffffff8054cd66>] __driver_attach+0x76/0x108
[<ffffffff8054a482>] bus_for_each_dev+0x52/0x9a
[<ffffffff8054be8c>] driver_attach+0x1a/0x28
[<ffffffff8054b996>] bus_add_driver+0x154/0x1c2
[<ffffffff8054d592>] driver_register+0x52/0x108
[<ffffffff804fe120>] register_virtio_driver+0x1c/0x2c
[<ffffffff80a29142>] virtio_net_driver_init+0x7a/0xb0
[<ffffffff80002854>] do_one_initcall+0x66/0x2e4
[<ffffffff80a01222>] kernel_init_freeable+0x28a/0x304
[<ffffffff808cb1be>] kernel_init+0x1e/0x110
[<ffffffff80003c4e>] ret_from_exception+0x0/0x10
---[ end trace 0000000000000000 ]---
Fixes 2ca653d607ce ("virtio_net: Stripe queue affinities across cores.")
---
 drivers/net/virtio_net.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/net/virtio_net.c +0 -0
index 7106932c6f88..e4b56523b2b5 100644

drivers/net/virtio_net.c static void virtnet_set_affinity(struct virtnet_info *vi) · line 2304

Quoted 5 lines
drivers/net/virtio_net.c +2 -0
@@ -2300,6 +2300,8 @@ static void virtnet_set_affinity(struct virtnet_info *vi)
 
              for (j = 0; j < group_size; j++) {
                      cpumask_set_cpu(cpu, mask);
+                     if (cpu == (nr_cpu_ids - 1))
+                             break;

The problem seems to be a problem with cpumask_next_wrap(), I'm not particularly sure.

Quoted 2 lines

But I think there is something wrong with your modification, which will cause subsequent queues to be bound to (nr_cpu_ids - 1).

Yes, it would lose cpu[nr_cpu_ids - 1]. We've moved to reverting the patch to fix problem:
https://lore.kernel.org/all/[email protected]/

Quoted 10 lines

Thanks.

Quoted 6 lines
drivers/net/virtio_net.c +0 -1
                      cpu = cpumask_next_wrap(cpu, cpu_online_mask,
                                              nr_cpu_ids, false);
              }
--

2.36.1

Signature · 2 lines
Best Regards
 Guo Ren
reviewer
#20 lore raw

Re: [PATCH V2 2/2] net: Fixup virtnet_set_affinity() cause cpumask warning

Quoted 68 lines · Guo Ren
Quoted 67 lines · Xuan Zhuo
Quoted 56 lines · On Thu, 13 Oct 2022 23:04:59 -0400, [email protected] wrote:

From: Guo Ren <[email protected]>

Don't pass nr_bits-1 as arg1 for cpumask_next_wrap, which would cause warning now 78e5a3399421 ("cpumask: fix checking valid cpu range").

------------[ cut here ]------------
WARNING: CPU: 0 PID: 1 at include/linux/cpumask.h:110 cpumask_next_wrap+0x5c/0x80
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 6.0.0-11659-ge7e38f6cce55-dirty #328
Hardware name: riscv-virtio,qemu (DT)
epc : cpumask_next_wrap+0x5c/0x80
 ra : virtnet_set_affinity+0x1ba/0x1fc

epc : ffffffff808992ca ra : ffffffff805d84ca sp : ff60000002327a50
gp : ffffffff81602390 tp : ff600000023a0000 t0 : 5f74656e74726976 t1 : 0000000000000000 t2 : 735f74656e747269 s0 : ff60000002327a90 s1 : 0000000000000003 a0 : 0000000000000003 a1 : ffffffff816051c0 a2 : 0000000000000004 a3 : 0000000000000000 a4 : 0000000000000000 a5 : 0000000000000004 a6 : 0000000000000000 a7 : 0000000000000000 s2 : 0000000000000000 s3 : ffffffff816051c0 s4 : ffffffff8160224c s5 : 0000000000000004 s6 : 0000000000000004 s7 : 0000000000000000 s8 : 0000000000000003 s9 : ffffffff810aa398 s10: ffffffff80e97d20
s11: 0000000000000004 t3 : ffffffff819acc97 t4 : ffffffff819acc97 t5 : ffffffff819acc98 t6 : ff60000002327878
status: 0000000200000120 badaddr: 0000000000000000 cause: 0000000000000003

[<ffffffff805d84ca>] virtnet_set_affinity+0x1ba/0x1fc
[<ffffffff805da7ac>] virtnet_probe+0x832/0xf1e
[<ffffffff804fe61c>] virtio_dev_probe+0x164/0x2de
[<ffffffff8054c4c4>] really_probe+0x82/0x224
[<ffffffff8054c6c0>] __driver_probe_device+0x5a/0xaa
[<ffffffff8054c73c>] driver_probe_device+0x2c/0xb8
[<ffffffff8054cd66>] __driver_attach+0x76/0x108
[<ffffffff8054a482>] bus_for_each_dev+0x52/0x9a
[<ffffffff8054be8c>] driver_attach+0x1a/0x28
[<ffffffff8054b996>] bus_add_driver+0x154/0x1c2
[<ffffffff8054d592>] driver_register+0x52/0x108
[<ffffffff804fe120>] register_virtio_driver+0x1c/0x2c
[<ffffffff80a29142>] virtio_net_driver_init+0x7a/0xb0
[<ffffffff80002854>] do_one_initcall+0x66/0x2e4
[<ffffffff80a01222>] kernel_init_freeable+0x28a/0x304
[<ffffffff808cb1be>] kernel_init+0x1e/0x110
[<ffffffff80003c4e>] ret_from_exception+0x0/0x10
---[ end trace 0000000000000000 ]---
Fixes 2ca653d607ce ("virtio_net: Stripe queue affinities across cores.")
---
 drivers/net/virtio_net.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/net/virtio_net.c +0 -0
index 7106932c6f88..e4b56523b2b5 100644

drivers/net/virtio_net.c static void virtnet_set_affinity(struct virtnet_info *vi) · line 2304

Quoted 5 lines
drivers/net/virtio_net.c +2 -0
@@ -2300,6 +2300,8 @@ static void virtnet_set_affinity(struct virtnet_info *vi)
 
              for (j = 0; j < group_size; j++) {
                      cpumask_set_cpu(cpu, mask);
+                     if (cpu == (nr_cpu_ids - 1))
+                             break;

The problem seems to be a problem with cpumask_next_wrap(), I'm not particularly sure.

Quoted 5 lines
Quoted 2 lines

But I think there is something wrong with your modification, which will cause subsequent queues to be bound to (nr_cpu_ids - 1).

Yes, it would lose cpu[nr_cpu_ids - 1]. We've moved to reverting the patch to fix problem:
https://lore.kernel.org/all/[email protected]/

Great!!

Thanks.

Quoted 18 lines
Quoted 12 lines

Thanks.

Quoted 6 lines
drivers/net/virtio_net.c +0 -1
                      cpu = cpumask_next_wrap(cpu, cpu_online_mask,
                                              nr_cpu_ids, false);
              }
--

2.36.1

--
Best Regards
Guo Ren