[PATCH v3 0/5] dax: add PROBE_PREFER_ASYNCHRONOUS to all the dax drivers

[PATCH v3 0/5] dax: add PROBE_PREFER_ASYNCHRONOUS to all the dax drivers

[Michal Clapinski]

Comments in linux/device/driver.h say that the goal is to do async
probing on all devices. The current behavior unnecessarily slows down
the boot by synchronously probing dax devices, so let's change that.

For thousands of devices, this change saves >1s of boot time.

Michal Clapinski (5):
  dax: add PROBE_PREFER_ASYNCHRONOUS to the pmem driver
  dax: add PROBE_PREFER_ASYNCHRONOUS to the kmem driver
  dax: add PROBE_PREFER_ASYNCHRONOUS to the cxl driver
  dax: add PROBE_PREFER_ASYNCHRONOUS to the hmem drivers
  dax: add PROBE_PREFER_ASYNCHRONOUS to the dax device driver

 drivers/dax/cxl.c       | 1 +
 drivers/dax/device.c    | 3 +++
 drivers/dax/hmem/hmem.c | 2 ++
 drivers/dax/kmem.c      | 3 +++
 drivers/dax/pmem.c      | 1 +
 5 files changed, 10 insertions(+)

-- 
2.51.1.821.gb6fe4d2222-goog

[PATCH v3 1/5] dax: add PROBE_PREFER_ASYNCHRONOUS to the pmem driver

[Michal Clapinski]

Signed-off-by: Michal Clapinski <mclapinski@google.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/dax/pmem.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/dax/pmem.c b/drivers/dax/pmem.c
index bee93066a849..737654e8c5e8 100644
--- a/drivers/dax/pmem.c
+++ b/drivers/dax/pmem.c
@@ -77,6 +77,7 @@ static struct nd_device_driver dax_pmem_driver = {
 	.probe = dax_pmem_probe,
 	.drv = {
 		.name = "dax_pmem",
+		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
 	},
 	.type = ND_DRIVER_DAX_PMEM,
 };
-- 
2.51.1.821.gb6fe4d2222-goog

[PATCH v3 2/5] dax: add PROBE_PREFER_ASYNCHRONOUS to the kmem driver

[Michal Clapinski]

Signed-off-by: Michal Clapinski <mclapinski@google.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/dax/kmem.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/dax/kmem.c b/drivers/dax/kmem.c
index c036e4d0b610..4bfaab2cb728 100644
--- a/drivers/dax/kmem.c
+++ b/drivers/dax/kmem.c
@@ -274,6 +274,9 @@ static struct dax_device_driver device_dax_kmem_driver = {
 	.probe = dev_dax_kmem_probe,
 	.remove = dev_dax_kmem_remove,
 	.type = DAXDRV_KMEM_TYPE,
+	.drv = {
+		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
+	},
 };
 
 static int __init dax_kmem_init(void)
-- 
2.51.1.821.gb6fe4d2222-goog

[PATCH v3 3/5] dax: add PROBE_PREFER_ASYNCHRONOUS to the cxl driver

[Michal Clapinski]

Signed-off-by: Michal Clapinski <mclapinski@google.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/dax/cxl.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/dax/cxl.c b/drivers/dax/cxl.c
index 13cd94d32ff7..90734ddbd369 100644
--- a/drivers/dax/cxl.c
+++ b/drivers/dax/cxl.c
@@ -38,6 +38,7 @@ static struct cxl_driver cxl_dax_region_driver = {
 	.id = CXL_DEVICE_DAX_REGION,
 	.drv = {
 		.suppress_bind_attrs = true,
+		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
 	},
 };
 
-- 
2.51.1.821.gb6fe4d2222-goog

[PATCH v3 4/5] dax: add PROBE_PREFER_ASYNCHRONOUS to the hmem drivers

[Michal Clapinski]

Signed-off-by: Michal Clapinski <mclapinski@google.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/dax/hmem/hmem.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/dax/hmem/hmem.c b/drivers/dax/hmem/hmem.c
index c18451a37e4f..5a6d99d90f77 100644
--- a/drivers/dax/hmem/hmem.c
+++ b/drivers/dax/hmem/hmem.c
@@ -45,6 +45,7 @@ static struct platform_driver dax_hmem_driver = {
 	.probe = dax_hmem_probe,
 	.driver = {
 		.name = "hmem",
+		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
 	},
 };
 
@@ -131,6 +132,7 @@ static struct platform_driver dax_hmem_platform_driver = {
 	.probe = dax_hmem_platform_probe,
 	.driver = {
 		.name = "hmem_platform",
+		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
 	},
 };
 
-- 
2.51.1.821.gb6fe4d2222-goog

[PATCH v3 5/5] dax: add PROBE_PREFER_ASYNCHRONOUS to the dax device driver

[Michal Clapinski]

Signed-off-by: Michal Clapinski <mclapinski@google.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/dax/device.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/dax/device.c b/drivers/dax/device.c
index 2bb40a6060af..74f2381a7df6 100644
--- a/drivers/dax/device.c
+++ b/drivers/dax/device.c
@@ -470,6 +470,9 @@ static int dev_dax_probe(struct dev_dax *dev_dax)
 static struct dax_device_driver device_dax_driver = {
 	.probe = dev_dax_probe,
 	.type = DAXDRV_DEVICE_TYPE,
+	.drv = {
+		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
+	},
 };
 
 static int __init dax_init(void)
-- 
2.51.1.821.gb6fe4d2222-goog

Re: [PATCH v3 0/5] dax: add PROBE_PREFER_ASYNCHRONOUS to all the dax drivers

[dan.j.williams]

Michal Clapinski wrote:
> Comments in linux/device/driver.h say that the goal is to do async
> probing on all devices. The current behavior unnecessarily slows down
> the boot by synchronously probing dax devices, so let's change that.
> 
> For thousands of devices, this change saves >1s of boot time.
> 
> Michal Clapinski (5):
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the pmem driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the kmem driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the cxl driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the hmem drivers
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the dax device driver

Looks good, thanks for the split.

Re: [PATCH v3 0/5] dax: add PROBE_PREFER_ASYNCHRONOUS to all the dax drivers

[Dave Jiang]

On 10/24/25 2:05 PM, Michal Clapinski wrote:
> Comments in linux/device/driver.h say that the goal is to do async
> probing on all devices. The current behavior unnecessarily slows down
> the boot by synchronously probing dax devices, so let's change that.
> 
> For thousands of devices, this change saves >1s of boot time.
> 
> Michal Clapinski (5):
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the pmem driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the kmem driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the cxl driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the hmem drivers
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the dax device driver
> 
>  drivers/dax/cxl.c       | 1 +
>  drivers/dax/device.c    | 3 +++
>  drivers/dax/hmem/hmem.c | 2 ++
>  drivers/dax/kmem.c      | 3 +++
>  drivers/dax/pmem.c      | 1 +
>  5 files changed, 10 insertions(+)
> 

Reviewed-by: Dave Jiang <dave.jiang@intel.com>
for the series

Re: [PATCH v3 0/5] dax: add PROBE_PREFER_ASYNCHRONOUS to all the dax drivers

[Michał Cłapiński]

On Fri, Oct 24, 2025 at 11:05 PM Michal Clapinski <mclapinski@google.com> wrote:
>
> Comments in linux/device/driver.h say that the goal is to do async
> probing on all devices. The current behavior unnecessarily slows down
> the boot by synchronously probing dax devices, so let's change that.
>
> For thousands of devices, this change saves >1s of boot time.
>
> Michal Clapinski (5):
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the pmem driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the kmem driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the cxl driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the hmem drivers
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the dax device driver
>
>  drivers/dax/cxl.c       | 1 +
>  drivers/dax/device.c    | 3 +++
>  drivers/dax/hmem/hmem.c | 2 ++
>  drivers/dax/kmem.c      | 3 +++
>  drivers/dax/pmem.c      | 1 +
>  5 files changed, 10 insertions(+)
>
> --
> 2.51.1.821.gb6fe4d2222-goog
>

+Ira (you're not listed as a maintainer of drivers/dax/ in the MAINTAINERS file)

Is there anything else required from my side or can this be merged?

Re: [PATCH v3 5/5] dax: add PROBE_PREFER_ASYNCHRONOUS to the dax device driver

[Ira Weiny]

Michal Clapinski wrote:
> Signed-off-by: Michal Clapinski <mclapinski@google.com>
> Reviewed-by: Dan Williams <dan.j.williams@intel.com>

Sorry for the delay.  I picked up this series but I find that this breaks
the device-dax and daxctl-create.sh.

I was able to fix device-dax with a sleep, see below.

I'm not 100% sure what to do about this.

I don't want to sprinkle sleeps around the tests.  daxctl-create.sh also
randomly fail due to the races introduced.  So not sure exactly where to
sprinkle them without more work.

Could dropping just this patch and landing the others achieve most of what
you need?

Ira

diff --git a/test/device-dax.c b/test/device-dax.c
index 49c9bc8b1748..817c76b0a88b 100644
--- a/test/device-dax.c
+++ b/test/device-dax.c
@@ -246,6 +246,7 @@ static int __test_device_dax(unsigned long align, int loglevel,
                goto out;
        }

+sleep(1);
        sprintf(path, "/dev/%s", daxctl_dev_get_devname(dev));
        fd = open(path, O_RDONLY);
        if (fd < 0) {

Re: [PATCH v3 5/5] dax: add PROBE_PREFER_ASYNCHRONOUS to the dax device driver

[Michał Cłapiński]

On Tue, Nov 18, 2025 at 10:46 PM Ira Weiny <ira.weiny@intel.com> wrote:
>
> Michal Clapinski wrote:
> > Signed-off-by: Michal Clapinski <mclapinski@google.com>
> > Reviewed-by: Dan Williams <dan.j.williams@intel.com>
>
> Sorry for the delay.  I picked up this series but I find that this breaks
> the device-dax and daxctl-create.sh.
>
> I was able to fix device-dax with a sleep, see below.
>
> I'm not 100% sure what to do about this.
>
> I don't want to sprinkle sleeps around the tests.  daxctl-create.sh also
> randomly fail due to the races introduced.  So not sure exactly where to
> sprinkle them without more work.

I see 2 possible solutions here:
1. Modify the tests to just poll for the devices to appear.
2. Modify ndctl to poll for the devices to appear before returning.

What do you think about those?

> Could dropping just this patch and landing the others achieve most of what
> you need?

No, device-dax is the only one I actually care about.

Re: [PATCH v3 5/5] dax: add PROBE_PREFER_ASYNCHRONOUS to the dax device driver

[Ira Weiny]

Michał Cłapiński wrote:
> On Tue, Nov 18, 2025 at 10:46 PM Ira Weiny <ira.weiny@intel.com> wrote:
> >
> > Michal Clapinski wrote:
> > > Signed-off-by: Michal Clapinski <mclapinski@google.com>
> > > Reviewed-by: Dan Williams <dan.j.williams@intel.com>
> >
> > Sorry for the delay.  I picked up this series but I find that this breaks
> > the device-dax and daxctl-create.sh.
> >
> > I was able to fix device-dax with a sleep, see below.
> >
> > I'm not 100% sure what to do about this.
> >
> > I don't want to sprinkle sleeps around the tests.  daxctl-create.sh also
> > randomly fail due to the races introduced.  So not sure exactly where to
> > sprinkle them without more work.
> 
> I see 2 possible solutions here:
> 1. Modify the tests to just poll for the devices to appear.
> 2. Modify ndctl to poll for the devices to appear before returning.
> 
> What do you think about those?

Yes those would be correct ways fix to the tests.  But I do wonder if
there are any real codes which might break with this.

The real issue is the incompatibility of changing the behavior and having
an older ndctl start failing either the tests or code written to
libdaxctl/daxctl.

In this case I'm not seeing a good way around this other than trying to
make a good documentation update to indicate that the ndctl needs to be
updated for the tests at a minimum.  If we get reports of failures from
real users though it will need to be reverted.

> 
> > Could dropping just this patch and landing the others achieve most of what
> > you need?
> 
> No, device-dax is the only one I actually care about.

I was afraid you would say that...  :-/  ;-)

I'm going to drop this patch while I coordinate with Alison on updating
ndctl.  If we can get a fix next week to ndctl I can throw it back in for
the merge.  I think it has enough testing to work that way.  Otherwise it
will have to land in 6.20 after ndctl is fixed up.

Ira

Re: [PATCH v3 0/5] dax: add PROBE_PREFER_ASYNCHRONOUS to all the dax drivers

[dan.j.williams]

Michal Clapinski wrote:
> Comments in linux/device/driver.h say that the goal is to do async
> probing on all devices. The current behavior unnecessarily slows down
> the boot by synchronously probing dax devices, so let's change that.
> 
> For thousands of devices, this change saves >1s of boot time.
> 
> Michal Clapinski (5):
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the pmem driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the kmem driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the cxl driver
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the hmem drivers
>   dax: add PROBE_PREFER_ASYNCHRONOUS to the dax device driver

After seeing the trouble this causes with libdaxctl and failing to find
a quick fix I wonder if you should just go through route of eating the
potential regressions in your own environment.

I.e. instead of making it a problem that the kernel needs to debug for
all legacy users, how about you just boot with the command line option:

   driver_async_probe=device_dax

...or add the following to your mopdrobe configuration:

   options device_dax async_probe

I.e. do you really need to change this policy globally for everyone at
this point?

I do want to improve this, but I think it will take time for libdaxctl
to get ready for this flag day.

Re: [PATCH v3 0/5] dax: add PROBE_PREFER_ASYNCHRONOUS to all the dax drivers

[Michał Cłapiński]

On Fri, Nov 21, 2025 at 1:59 AM <dan.j.williams@intel.com> wrote:
>
> Michal Clapinski wrote:
> > Comments in linux/device/driver.h say that the goal is to do async
> > probing on all devices. The current behavior unnecessarily slows down
> > the boot by synchronously probing dax devices, so let's change that.
> >
> > For thousands of devices, this change saves >1s of boot time.
> >
> > Michal Clapinski (5):
> >   dax: add PROBE_PREFER_ASYNCHRONOUS to the pmem driver
> >   dax: add PROBE_PREFER_ASYNCHRONOUS to the kmem driver
> >   dax: add PROBE_PREFER_ASYNCHRONOUS to the cxl driver
> >   dax: add PROBE_PREFER_ASYNCHRONOUS to the hmem drivers
> >   dax: add PROBE_PREFER_ASYNCHRONOUS to the dax device driver
>
> After seeing the trouble this causes with libdaxctl and failing to find
> a quick fix I wonder if you should just go through route of eating the
> potential regressions in your own environment.
>
> I.e. instead of making it a problem that the kernel needs to debug for
> all legacy users, how about you just boot with the command line option:
>
>    driver_async_probe=device_dax
>
> ...or add the following to your mopdrobe configuration:
>
>    options device_dax async_probe
>
> I.e. do you really need to change this policy globally for everyone at
> this point?

I'll do that or I'll just cherry-pick this to our kernel. Thank you.

> I do want to improve this, but I think it will take time for libdaxctl
> to get ready for this flag day.