Driver core changes for 6.0-rc1
The following changes since commit f2906aa863381afb0015a9eb7fefad885d4e5a56:
Linux 5.19-rc1 (2022-06-05 17:18:54 -0700)
are available in the Git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core.git tags/driver-core-6.0-rc1
for you to fetch changes up to 273aaa24369cb8d0f246bb16f7122b91a1ef5188:
docs: embargoed-hardware-issues: fix invalid AMD contact email (2022-07-29 16:10:04 +0200)
----------------------------------------------------------------
Driver core / kernfs changes for 6.0-rc1
Here is the set of driver core and kernfs changes for 6.0-rc1.
"biggest" thing in here is some scalability improvements for kernfs for large systems. Other than that, included in here are:
- arch topology and cache info changes that have been reviewed and discussed a lot.
- potential error path cleanup fixes
- deferred driver probe cleanups
- firmware loader cleanups and tweaks
- documentation updates
- other small things
All of these have been in the linux-next tree for a while with no reported problems.
----------------------------------------------------------------
Andy Shevchenko (2):
driver core: Introduce device_find_any_child() helper
spi: Use device_find_any_child() instead of custom approach
Dave Airlie (1):
docs: driver-api: firmware: add driver firmware guidelines. (v3)
Duoming Zhou (2):
devcoredump: remove the useless gfp_t parameter in dev_coredumpv and dev_coredumpm
mwifiex: fix sleep in atomic context bugs caused by dev_coredumpv
Fabio M. De Francesco (1):
firmware_loader: Replace kmap() with kmap_local_page()
Florian Fainelli (1):
MAINTAINERS: Change mentions of mpm to olivia
Greg Kroah-Hartman (4):
Revert "mwifiex: fix sleep in atomic context bugs caused by dev_coredumpv"
Revert "devcoredump: remove the useless gfp_t parameter in dev_coredumpv and dev_coredumpm"
Merge tag 'arch-cache-topo-5.20' of git://git.kernel.org/pub/scm/linux/kernel/git/sudeep.holla/linux into driver-core-next
docs: embargoed-hardware-issues: fix invalid AMD contact email
Imran Khan (5):
kernfs: make ->attr.open RCU protected.
kernfs: Change kernfs_notify_list to llist.
kernfs: Introduce interface to access global kernfs_open_file_mutex.
kernfs: Replace global kernfs_open_file_mutex with hashed mutexes. Revert "kernfs: Change kernfs_notify_list to llist."
Ionela Voinescu (1):
arch_topology: Limit span of cpu_clustergroup_mask()
Lee Jones (2):
docs: ABI: sysfs-class-pwm: Update Lee Jones' email address
docs: ABI: sysfs-devices-soc: Update Lee Jones' email address
Liang He (1):
firmware: Hold a reference for of_find_compatible_node()
Lin Feng (1):
kernfs/file.c: remove redundant error return counter assignment
Mauro Carvalho Chehab (1):
ABI: testing/sysfs-devices-system-cpu: remove duplicated core_id
Nick Desaulniers (1):
Documentation/process: Add embargoed HW contact for LLVM
Phil Auld (1):
drivers/base: fix userspace break from using bin_attributes for cpumap and cpulist
Randy Dunlap (1):
kobject: fix Kconfig.debug "its" grammar
Saravana Kannan (11):
PM: domains: Delete usage of driver_deferred_probe_check_state()
pinctrl: devicetree: Delete usage of driver_deferred_probe_check_state()
net: mdio: Delete usage of driver_deferred_probe_check_state() driver core: Add wait_for_init_devices_probe helper function
net: ipconfig: Relax fw_devlink if we need to mount a network rootfs Revert "driver core: Set default deferred_probe_timeout back to 0." driver core: Set fw_devlink.strict=1 by default
iommu/of: Delete usage of driver_deferred_probe_check_state() driver core: Delete driver_deferred_probe_check_state() driver core: fw_devlink: Allow firmware to mark devices as best effort
of: base: Avoid console probe delay when fw_devlink.strict=1
Slark Xiao (2):
kernfs: Fix typo 'the the' in comment
sysfs docs: ABI: Fix typo in comment
Sudeep Holla (23):
ACPI: PPTT: Use table offset as fw_token instead of virtual address
cacheinfo: Use of_cpu_device_node_get instead cpu_dev->of_node
cacheinfo: Add helper to access any cache index for a given CPU
cacheinfo: Move cache_leaves_are_shared out of CONFIG_OF
cacheinfo: Add support to check if last level cache(LLC) is valid or shared
cacheinfo: Allow early detection and population of cache attributes
cacheinfo: Use cache identifiers to check if the caches are shared if available
cacheinfo: Align checks in cache_shared_cpu_map_{setup,remove} for readability
arch_topology: Add support to parse and detect cache attributes
arch_topology: Use the last level cache information from the cacheinfo
arm64: topology: Remove redundant setting of llc_id in CPU topology
arch_topology: Drop LLC identifier stash from the CPU topology
arch_topology: Set thread sibling cpumask only within the cluster
arch_topology: Check for non-negative value rather than -1 for IDs validity
arch_topology: Avoid parsing through all the CPUs once a outlier CPU is found
arch_topology: Don't set cluster identifier as physical package identifier
arch_topology: Set cluster identifier in each core/thread from /cpu-map
arch_topology: Add support for parsing sockets in /cpu-map
arch_topology: Warn that topology for nested clusters is not supported
ACPI: Remove the unused find_acpi_cpu_cache_topology()
cacheinfo: Use atomic allocation for percpu cache attributes
ACPI: PPTT: Leave the table mapped for the runtime usage
arch_topology: Fix cache attributes detection in the CPU hotplug path
Yangxi Xiang (1):
devtmpfs: fix the dangling pointer of global devtmpfsd thread
Yushan Zhou (1):
kernfs: fix potential NULL dereference in __kernfs_remove
Zhang Wensheng (1):
driver core: fix potential deadlock in __driver_attach
Documentation/ABI/stable/sysfs-module | 2 +- Documentation/ABI/testing/sysfs-class-pwm | 2 +- Documentation/ABI/testing/sysfs-class-rtrs-client | 2 +- Documentation/ABI/testing/sysfs-class-rtrs-server | 2 +- .../ABI/testing/sysfs-devices-platform-ACPI-TAD | 2 +- Documentation/ABI/testing/sysfs-devices-power | 2 +- Documentation/ABI/testing/sysfs-devices-soc | 14 +- Documentation/ABI/testing/sysfs-devices-system-cpu | 7 +- Documentation/driver-api/firmware/core.rst | 1 + .../firmware/firmware-usage-guidelines.rst | 44 +++++ .../process/embargoed-hardware-issues.rst | 5 +- .../zh_CN/process/embargoed-hardware-issues.rst | 2 +- .../zh_TW/process/embargoed-hardware-issues.rst | 2 +- MAINTAINERS | 4 +- arch/arm64/kernel/topology.c | 14 -- drivers/acpi/pptt.c | 142 +++++--------- drivers/base/arch_topology.c | 100 +++++++--- drivers/base/base.h | 1 + drivers/base/cacheinfo.c | 145 +++++++++------ drivers/base/core.c | 123 ++++++++++++- drivers/base/dd.c | 59 +++--- drivers/base/devtmpfs.c | 1 + drivers/base/firmware_loader/main.c | 4 +- drivers/base/firmware_loader/sysfs.c | 10 +- drivers/base/node.c | 4 +- drivers/base/power/domain.c | 2 +- drivers/base/topology.c | 32 ++-- drivers/iommu/of_iommu.c | 2 +- drivers/net/mdio/fwnode_mdio.c | 4 +- drivers/of/base.c | 2 + drivers/pinctrl/devicetree.c | 2 +- drivers/spi/spi.c | 9 +- fs/kernfs/dir.c | 7 +- fs/kernfs/file.c | 205 ++++++++++++++------- fs/kernfs/kernfs-internal.h | 4 + fs/kernfs/mount.c | 19 ++ include/linux/acpi.h | 5 - include/linux/arch_topology.h | 1 - include/linux/cacheinfo.h | 3 + include/linux/cpumask.h | 18 ++ include/linux/device.h | 2 + include/linux/device/driver.h | 2 +- include/linux/firmware/trusted_foundations.h | 8 +- include/linux/fwnode.h | 4 + include/linux/kernfs.h | 59 +++++- lib/Kconfig.debug | 2 +- net/ipv4/ipconfig.c | 6 + 47 files changed, 718 insertions(+), 374 deletions(-) create mode 100644 Documentation/driver-api/firmware/firmware-usage-guidelines.rst
The pull request you sent on Wed, 3 Aug 2022 16:16:16 +0200:
The pull request you sent on Wed, 3 Aug 2022 16:16:16 +0200:
Quoted 1 line
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core.gittags/driver-core-6.0-rc1
has been merged into torvalds/linux.git:
https://git.kernel.org/torvalds/c/cfeafd94668910334a77c9437a18212baf9f5610
Thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/prtracker.html
Hi,
Quoted 4 lines · Greg KH
Saravana Kannan (11):
PM: domains: Delete usage ofdriver_deferred_probe_check_state()
pinctrl: devicetree: Delete usage ofdriver_deferred_probe_check_state()
net: mdio: Delete usage ofdriver_deferred_probe_check_state()
Quoted 8 lines
driver core: Add
wait_for_init_devices_probehelper function
net: ipconfig: Relaxfw_devlinkif we need to mount a network rootfs Revert "driver core: Set defaultdeferred_probe_timeoutback to 0." driver core: Set fw_devlink.strict=1 by default
iommu/of: Delete usage ofdriver_deferred_probe_check_state()driver core: Deletedriver_deferred_probe_check_state()driver core:fw_devlink: Allow firmware to mark devices as best effort
of: base: Avoid console probe delay when fw_devlink.strict=1
The last patch in this list regresses my HoneyComb LX2K (ironically the machine I do maintainer work on). It stops PCIe from probing, but without a single message indicating why.
The reason seems to be that the iommu-maps property doesn't get patched up by my (older) u-boot, and thus isn't a valid reference. System works fine without IOMMU, which is how I've ran it for a couple of years.
It's also extremely hard to diagnose out of the box because there are *no error messages*. And there were no warnings leading up to this strict enforcement.
This "feature" seems to have been done backwards. The checks should have been running (and not skipped due to the "optional" flag), but also not causing errors, just warnings. That would have given users a chance to know that this is something that needs to be fixed.
And when you flip the switch, at least report what failed so that people don't need to spend a whole night bisecting kernels, please.
Greg, mind reverting just the last one? If I hit this, I presume others would too.
-Olof
Quoted 36 lines · Olof Johansson
Hi,
Quoted 4 lines · Greg KH
Saravana Kannan (11):
PM: domains: Delete usage ofdriver_deferred_probe_check_state()
pinctrl: devicetree: Delete usage ofdriver_deferred_probe_check_state()
net: mdio: Delete usage ofdriver_deferred_probe_check_state()Quoted 8 lines
driver core: Add
wait_for_init_devices_probehelper function
net: ipconfig: Relaxfw_devlinkif we need to mount a network rootfs Revert "driver core: Set defaultdeferred_probe_timeoutback to 0." driver core: Set fw_devlink.strict=1 by default
iommu/of: Delete usage ofdriver_deferred_probe_check_state()driver core: Deletedriver_deferred_probe_check_state()driver core:fw_devlink: Allow firmware to mark devices as best effort
of: base: Avoid console probe delay when fw_devlink.strict=1The last patch in this list regresses my HoneyComb LX2K (ironically the machine I do maintainer work on). It stops PCIe from probing, but without a single message indicating why.
The reason seems to be that the iommu-maps property doesn't get patched up by my (older) u-boot, and thus isn't a valid reference. System works fine without IOMMU, which is how I've ran it for a couple of years.
It's also extremely hard to diagnose out of the box because there are *no error messages*. And there were no warnings leading up to this strict enforcement.
This "feature" seems to have been done backwards. The checks should have been running (and not skipped due to the "optional" flag), but also not causing errors, just warnings. That would have given users a chance to know that this is something that needs to be fixed.
Quoted 5 lines
And when you flip the switch, at least report what failed so that people don't need to spend a whole night bisecting kernels, please.
Greg, mind reverting just the last one? If I hit this, I presume others would too.
Apologies, wrong patch pointed out. The culprit is "driver core: Set fw_devlink.strict=1 by default", 71066545b48e42.
-Olof
Quoted 37 lines · Olof Johansson
Quoted 36 lines · Olof Johansson
Hi,
Quoted 4 lines · Greg KH
Saravana Kannan (11):
PM: domains: Delete usage ofdriver_deferred_probe_check_state()
pinctrl: devicetree: Delete usage ofdriver_deferred_probe_check_state()
net: mdio: Delete usage ofdriver_deferred_probe_check_state()Quoted 8 lines
driver core: Add
wait_for_init_devices_probehelper function
net: ipconfig: Relaxfw_devlinkif we need to mount a network rootfs Revert "driver core: Set defaultdeferred_probe_timeoutback to 0." driver core: Set fw_devlink.strict=1 by default
iommu/of: Delete usage ofdriver_deferred_probe_check_state()driver core: Deletedriver_deferred_probe_check_state()driver core:fw_devlink: Allow firmware to mark devices as best effort
of: base: Avoid console probe delay when fw_devlink.strict=1The last patch in this list regresses my HoneyComb LX2K (ironically the machine I do maintainer work on). It stops PCIe from probing, but without a single message indicating why.
The reason seems to be that the iommu-maps property doesn't get patched up by my (older) u-boot, and thus isn't a valid reference. System works fine without IOMMU, which is how I've ran it for a couple of years.
It's also extremely hard to diagnose out of the box because there are *no error messages*. And there were no warnings leading up to this strict enforcement.
This "feature" seems to have been done backwards. The checks should have been running (and not skipped due to the "optional" flag), but also not causing errors, just warnings. That would have given users a chance to know that this is something that needs to be fixed.
Quoted 8 lines
Quoted 5 lines
And when you flip the switch, at least report what failed so that people don't need to spend a whole night bisecting kernels, please.
Greg, mind reverting just the last one? If I hit this, I presume others would too.
Apologies, wrong patch pointed out. The culprit is "driver core: Set fw_devlink.strict=1 by default",
71066545b48e42.
Is this still an issue in -rc5? A number of patches in the above series was just reverted and hopefully should have resolved the issue you are seeing.
thanks,
greg k-h
_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Quoted 45 lines · Greg KH
Quoted 43 lines · Olof Johansson
Quoted 41 lines · Olof Johansson
Hi,
Quoted 4 lines · Greg KH
Saravana Kannan (11):
PM: domains: Delete usage ofdriver_deferred_probe_check_state()
pinctrl: devicetree: Delete usage ofdriver_deferred_probe_check_state()
net: mdio: Delete usage ofdriver_deferred_probe_check_state()Quoted 8 lines
driver core: Add
wait_for_init_devices_probehelper function
net: ipconfig: Relaxfw_devlinkif we need to mount a network rootfs Revert "driver core: Set defaultdeferred_probe_timeoutback to 0." driver core: Set fw_devlink.strict=1 by default
iommu/of: Delete usage ofdriver_deferred_probe_check_state()driver core: Deletedriver_deferred_probe_check_state()driver core:fw_devlink: Allow firmware to mark devices as best effort
of: base: Avoid console probe delay when fw_devlink.strict=1The last patch in this list regresses my HoneyComb LX2K (ironically the machine I do maintainer work on). It stops PCIe from probing, but without a single message indicating why.
The reason seems to be that the iommu-maps property doesn't get patched up by my (older) u-boot, and thus isn't a valid reference. System works fine without IOMMU, which is how I've ran it for a couple of years.
It's also extremely hard to diagnose out of the box because there are *no error messages*. And there were no warnings leading up to this strict enforcement.
This "feature" seems to have been done backwards. The checks should have been running (and not skipped due to the "optional" flag), but also not causing errors, just warnings. That would have given users a chance to know that this is something that needs to be fixed.
And when you flip the switch, at least report what failed so that people don't need to spend a whole night bisecting kernels, please.
Greg, mind reverting just the last one? If I hit this, I presume others would too.
Quoted 6 lines
Quoted 2 lines
Apologies, wrong patch pointed out. The culprit is "driver core: Set fw_devlink.strict=1 by default",
71066545b48e42.Is this still an issue in -rc5? A number of patches in the above series was just reverted and hopefully should have resolved the issue you are seeing.
Unfortunately, I discovered this regression with -rc5 in the first place, so it's still there.
-Olof
_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Quoted 49 lines · Olof Johansson
Quoted 48 lines · Greg KH
Quoted 45 lines · Olof Johansson
Quoted 36 lines · Olof Johansson
Hi,
Quoted 4 lines · Greg KH
Saravana Kannan (11):
PM: domains: Delete usage ofdriver_deferred_probe_check_state()
pinctrl: devicetree: Delete usage ofdriver_deferred_probe_check_state()
net: mdio: Delete usage ofdriver_deferred_probe_check_state()Quoted 8 lines
driver core: Add
wait_for_init_devices_probehelper function
net: ipconfig: Relaxfw_devlinkif we need to mount a network rootfs Revert "driver core: Set defaultdeferred_probe_timeoutback to 0." driver core: Set fw_devlink.strict=1 by default
iommu/of: Delete usage ofdriver_deferred_probe_check_state()driver core: Deletedriver_deferred_probe_check_state()driver core:fw_devlink: Allow firmware to mark devices as best effort
of: base: Avoid console probe delay when fw_devlink.strict=1The last patch in this list regresses my HoneyComb LX2K (ironically the machine I do maintainer work on). It stops PCIe from probing, but without a single message indicating why.
The reason seems to be that the iommu-maps property doesn't get patched up by my (older) u-boot, and thus isn't a valid reference. System works fine without IOMMU, which is how I've ran it for a couple of years.
It's also extremely hard to diagnose out of the box because there are *no error messages*. And there were no warnings leading up to this strict enforcement.
This "feature" seems to have been done backwards. The checks should have been running (and not skipped due to the "optional" flag), but also not causing errors, just warnings. That would have given users a chance to know that this is something that needs to be fixed.
Quoted 5 lines
And when you flip the switch, at least report what failed so that people don't need to spend a whole night bisecting kernels, please.
Greg, mind reverting just the last one? If I hit this, I presume others would too.
Apologies, wrong patch pointed out. The culprit is "driver core: Set fw_devlink.strict=1 by default",
71066545b48e42.
Quoted 6 lines
Quoted 3 lines
Is this still an issue in -rc5? A number of patches in the above series was just reverted and hopefully should have resolved the issue you are seeing.
Unfortunately, I discovered this regression with -rc5 in the first place, so it's still there.
Ick, ok, Saravana, any thoughts? I know you're at the conference this week with me, maybe you can give Olof a hint as to what to look for here?
thanks,
greg k-h
Hi,
Quoted 55 lines · Greg KH
Quoted 53 lines · Olof Johansson
Quoted 51 lines · Greg KH
Quoted 37 lines · Olof Johansson
Quoted 36 lines · Olof Johansson
Hi,
Quoted 4 lines · Greg KH
Saravana Kannan (11):
PM: domains: Delete usage ofdriver_deferred_probe_check_state()
pinctrl: devicetree: Delete usage ofdriver_deferred_probe_check_state()
net: mdio: Delete usage ofdriver_deferred_probe_check_state()Quoted 8 lines
driver core: Add
wait_for_init_devices_probehelper function
net: ipconfig: Relaxfw_devlinkif we need to mount a network rootfs Revert "driver core: Set defaultdeferred_probe_timeoutback to 0." driver core: Set fw_devlink.strict=1 by default
iommu/of: Delete usage ofdriver_deferred_probe_check_state()driver core: Deletedriver_deferred_probe_check_state()driver core:fw_devlink: Allow firmware to mark devices as best effort
of: base: Avoid console probe delay when fw_devlink.strict=1The last patch in this list regresses my HoneyComb LX2K (ironically the machine I do maintainer work on). It stops PCIe from probing, but without a single message indicating why.
The reason seems to be that the iommu-maps property doesn't get patched up by my (older) u-boot, and thus isn't a valid reference. System works fine without IOMMU, which is how I've ran it for a couple of years.
It's also extremely hard to diagnose out of the box because there are *no error messages*. And there were no warnings leading up to this strict enforcement.
This "feature" seems to have been done backwards. The checks should have been running (and not skipped due to the "optional" flag), but also not causing errors, just warnings. That would have given users a chance to know that this is something that needs to be fixed.
Quoted 8 lines
Quoted 5 lines
And when you flip the switch, at least report what failed so that people don't need to spend a whole night bisecting kernels, please.
Greg, mind reverting just the last one? If I hit this, I presume others would too.
Apologies, wrong patch pointed out. The culprit is "driver core: Set fw_devlink.strict=1 by default",
71066545b48e42.Is this still an issue in -rc5? A number of patches in the above series was just reverted and hopefully should have resolved the issue you are seeing.
Quoted 6 lines
Quoted 2 lines
Unfortunately, I discovered this regression with -rc5 in the first place, so it's still there.
Ick, ok, Saravana, any thoughts? I know you're at the conference this week with me, maybe you can give Olof a hint as to what to look for here?
I'm not sure what you want me to look for. The patch turns on enforcement of DT contents that never used to be enforced, so now my computer no longer boots. And it does it in a way that makes it impossible for someone not rebuilding kernels to debug to figure out what happened.
The patch needs to be reverted.
-Olof
Quoted 66 lines · Olof Johansson
Hi,
Quoted 55 lines · Greg KH
Quoted 53 lines · Olof Johansson
Quoted 51 lines · Greg KH
Quoted 37 lines · Olof Johansson
Quoted 36 lines · Olof Johansson
Hi,
Quoted 4 lines · Greg KH
Saravana Kannan (11):
PM: domains: Delete usage ofdriver_deferred_probe_check_state()
pinctrl: devicetree: Delete usage ofdriver_deferred_probe_check_state()
net: mdio: Delete usage ofdriver_deferred_probe_check_state()Quoted 8 lines
driver core: Add
wait_for_init_devices_probehelper function
net: ipconfig: Relaxfw_devlinkif we need to mount a network rootfs Revert "driver core: Set defaultdeferred_probe_timeoutback to 0." driver core: Set fw_devlink.strict=1 by default
iommu/of: Delete usage ofdriver_deferred_probe_check_state()driver core: Deletedriver_deferred_probe_check_state()driver core:fw_devlink: Allow firmware to mark devices as best effort
of: base: Avoid console probe delay when fw_devlink.strict=1The last patch in this list regresses my HoneyComb LX2K (ironically the machine I do maintainer work on). It stops PCIe from probing, but without a single message indicating why.
The reason seems to be that the iommu-maps property doesn't get patched up by my (older) u-boot, and thus isn't a valid reference. System works fine without IOMMU, which is how I've ran it for a couple of years.
It's also extremely hard to diagnose out of the box because there are *no error messages*. And there were no warnings leading up to this strict enforcement.
This "feature" seems to have been done backwards. The checks should have been running (and not skipped due to the "optional" flag), but also not causing errors, just warnings. That would have given users a chance to know that this is something that needs to be fixed.
Quoted 8 lines
Quoted 5 lines
And when you flip the switch, at least report what failed so that people don't need to spend a whole night bisecting kernels, please.
Greg, mind reverting just the last one? If I hit this, I presume others would too.
Apologies, wrong patch pointed out. The culprit is "driver core: Set fw_devlink.strict=1 by default",
71066545b48e42.Is this still an issue in -rc5? A number of patches in the above series was just reverted and hopefully should have resolved the issue you are seeing.
Quoted 6 lines
Quoted 2 lines
Unfortunately, I discovered this regression with -rc5 in the first place, so it's still there.
Ick, ok, Saravana, any thoughts? I know you're at the conference this week with me, maybe you can give Olof a hint as to what to look for here?
Quoted 5 lines
I'm not sure what you want me to look for. The patch turns on enforcement of DT contents that never used to be enforced, so now my computer no longer boots. And it does it in a way that makes it impossible for someone not rebuilding kernels to debug to figure out what happened.
Hi Olof,
Sorry for the trouble. It doesn't print any error messages because there are cases where it's block the probe where it wouldn't be an error. If I printed it every time fw_devlink blocked a probe, it'd be a ton of messages.
Btw, when I enabled fw_devlink.strict=1, it was AFTER making changes that'll stop indefinitely blocking probes. So what you are seeing shouldn't be happening. After about 10 seconds (configurable), it should stop blocking the probes.
If you actually want to see the reasons the probe is being blocked, you can enable the existing dbg messages in drivers/base/core.c.
Would you mind pointing me to the dts (not dtsi) file that corresponds to this board please? And which specific PCI device is being blocked from probing? I'll can try to debug it further. Also, can you try to see why it doesn't get unblocked when driver_deferred_probe_timeout expires? Or why that's not helping here?
In the meantime, if you want this patch reverted, I'm not opposed to that. But if you can use fw_devlinks.strict=0 in your commandline for now and give me time to debug, that'd be nicer.
Thanks,
Saravana
Quoted 81 lines · Saravana Kannan
Quoted 66 lines · Olof Johansson
Hi,
Quoted 55 lines · Greg KH
Quoted 53 lines · Olof Johansson
Quoted 51 lines · Greg KH
Quoted 37 lines · Olof Johansson
Quoted 36 lines · Olof Johansson
Hi,
Quoted 4 lines · Greg KH
Saravana Kannan (11):
PM: domains: Delete usage ofdriver_deferred_probe_check_state()
pinctrl: devicetree: Delete usage ofdriver_deferred_probe_check_state()
net: mdio: Delete usage ofdriver_deferred_probe_check_state()Quoted 8 lines
driver core: Add
wait_for_init_devices_probehelper function
net: ipconfig: Relaxfw_devlinkif we need to mount a network rootfs Revert "driver core: Set defaultdeferred_probe_timeoutback to 0." driver core: Set fw_devlink.strict=1 by default
iommu/of: Delete usage ofdriver_deferred_probe_check_state()driver core: Deletedriver_deferred_probe_check_state()driver core:fw_devlink: Allow firmware to mark devices as best effort
of: base: Avoid console probe delay when fw_devlink.strict=1The last patch in this list regresses my HoneyComb LX2K (ironically the machine I do maintainer work on). It stops PCIe from probing, but without a single message indicating why.
The reason seems to be that the iommu-maps property doesn't get patched up by my (older) u-boot, and thus isn't a valid reference. System works fine without IOMMU, which is how I've ran it for a couple of years.
It's also extremely hard to diagnose out of the box because there are *no error messages*. And there were no warnings leading up to this strict enforcement.
This "feature" seems to have been done backwards. The checks should have been running (and not skipped due to the "optional" flag), but also not causing errors, just warnings. That would have given users a chance to know that this is something that needs to be fixed.
Quoted 8 lines
Quoted 5 lines
And when you flip the switch, at least report what failed so that people don't need to spend a whole night bisecting kernels, please.
Greg, mind reverting just the last one? If I hit this, I presume others would too.
Apologies, wrong patch pointed out. The culprit is "driver core: Set fw_devlink.strict=1 by default",
71066545b48e42.Is this still an issue in -rc5? A number of patches in the above series was just reverted and hopefully should have resolved the issue you are seeing.
Quoted 6 lines
Quoted 2 lines
Unfortunately, I discovered this regression with -rc5 in the first place, so it's still there.
Ick, ok, Saravana, any thoughts? I know you're at the conference this week with me, maybe you can give Olof a hint as to what to look for here?
Quoted 5 lines
I'm not sure what you want me to look for. The patch turns on enforcement of DT contents that never used to be enforced, so now my computer no longer boots. And it does it in a way that makes it impossible for someone not rebuilding kernels to debug to figure out what happened.
Hi Olof,
Sorry for the trouble. It doesn't print any error messages because there are cases where it's block the probe where it wouldn't be an error. If I printed it every time
fw_devlinkblocked a probe, it'd be a ton of messages.
Quoted 4 lines
Btw, when I enabled fw_devlink.strict=1, it was AFTER making changes that'll stop indefinitely blocking probes. So what you are seeing shouldn't be happening. After about 10 seconds (configurable), it should stop blocking the probes.
"Shouldn't be happening" is a pretty bold statement. It's not actually stuck on timeout in my case, and doesn't recover.
Instead, what seems to be happening is that the PCIe driver, which registers as a platform_driver here:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/pci/controller/mobiveil/pcie-layerscape-gen4.c#n255
ends up registering, and the driver core now refuses to try to probe the device matches, since they no longer have their suppliers fulfilled (the smmu suppliers would not be tracked since they are optional here:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/of/property.c#n1449
So what happens is that the driver registration succeeds, but there have been no devices matched to it. So when it returns to the platform core, it thinks there are no devices bound to this driver, so it should be unregistered:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/base/platform.c#n951
That explains why the pcie core doesn't retry and just disappears, and stops retrying.
This is what it looks like with CONFIG_DEBUG_DRIVER and CONFIG_DEBUG_DEVRES:
[ 5.178538] bus: 'platform': add driver layerscape-pcie-gen4 [ 5.184301] bus: 'platform': __driver_probe_device: matched device 3600000.pcie with driver layerscape-pcie-gen4 [ 5.194498] platform 3600000.pcie: error -EPROBE_DEFER: supplier 5000000.iommu not ready [ 5.202607] platform 3600000.pcie: Added to deferred list [ 5.208024] bus: 'platform': __driver_probe_device: matched device 3800000.pcie with driver layerscape-pcie-gen4 [ 5.218227] platform 3800000.pcie: error -EPROBE_DEFER: supplier 5000000.iommu not ready [ 5.226333] platform 3800000.pcie: Added to deferred list [ 5.231814] bus: 'platform': remove driver layerscape-pcie-gen4 [ 5.237761] driver: 'layerscape-pcie-gen4': driver_release
Note that the platform driver registration sets flags to disable async probing, supposedly so it can assume that any matching devices would be found by the time registration returns:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/base/platform.c#n917
:
/*
* We have to run our probes synchronously because we check if
* we find any devices to bind to and exit with error if there
* are any.
*/drv->driver.probe_type = PROBE_FORCE_SYNCHRONOUS;
/*
* Prevent driver from requesting probe deferral to avoid further
* futile probe attempts.
*/drv->prevent_deferred_probe = true;
Bottom line: How was this code tested? This seems far from mature, this doesn't seem like that of an obscure condition to occur and it could create minefields for others down the road if it's fragile.
-Olof
Quoted 16 lines
If you actually want to see the reasons the probe is being blocked, you can enable the existing dbg messages in
drivers/base/core.c.Would you mind pointing me to the dts (not dtsi) file that corresponds to this board please? And which specific PCI device is being blocked from probing? I'll can try to debug it further. Also, can you try to see why it doesn't get unblocked when
driver_deferred_probe_timeoutexpires? Or why that's not helping here?In the meantime, if you want this patch reverted, I'm not opposed to that. But if you can use fw_devlinks.strict=0 in your commandline for now and give me time to debug, that'd be nicer.
Thanks,
Saravana
Quoted 148 lines · Olof Johansson
Quoted 81 lines · Saravana Kannan
Quoted 66 lines · Olof Johansson
Hi,
Quoted 55 lines · Greg KH
Quoted 53 lines · Olof Johansson
Quoted 51 lines · Greg KH
Quoted 37 lines · Olof Johansson
Quoted 36 lines · Olof Johansson
Hi,
Quoted 4 lines · Greg KH
Saravana Kannan (11):
PM: domains: Delete usage ofdriver_deferred_probe_check_state()
pinctrl: devicetree: Delete usage ofdriver_deferred_probe_check_state()
net: mdio: Delete usage ofdriver_deferred_probe_check_state()Quoted 8 lines
driver core: Add
wait_for_init_devices_probehelper function
net: ipconfig: Relaxfw_devlinkif we need to mount a network rootfs Revert "driver core: Set defaultdeferred_probe_timeoutback to 0." driver core: Set fw_devlink.strict=1 by default
iommu/of: Delete usage ofdriver_deferred_probe_check_state()driver core: Deletedriver_deferred_probe_check_state()driver core:fw_devlink: Allow firmware to mark devices as best effort
of: base: Avoid console probe delay when fw_devlink.strict=1The last patch in this list regresses my HoneyComb LX2K (ironically the machine I do maintainer work on). It stops PCIe from probing, but without a single message indicating why.
The reason seems to be that the iommu-maps property doesn't get patched up by my (older) u-boot, and thus isn't a valid reference. System works fine without IOMMU, which is how I've ran it for a couple of years.
It's also extremely hard to diagnose out of the box because there are *no error messages*. And there were no warnings leading up to this strict enforcement.
This "feature" seems to have been done backwards. The checks should have been running (and not skipped due to the "optional" flag), but also not causing errors, just warnings. That would have given users a chance to know that this is something that needs to be fixed.
Quoted 8 lines
Quoted 5 lines
And when you flip the switch, at least report what failed so that people don't need to spend a whole night bisecting kernels, please.
Greg, mind reverting just the last one? If I hit this, I presume others would too.
Apologies, wrong patch pointed out. The culprit is "driver core: Set fw_devlink.strict=1 by default",
71066545b48e42.Is this still an issue in -rc5? A number of patches in the above series was just reverted and hopefully should have resolved the issue you are seeing.
Quoted 6 lines
Quoted 2 lines
Unfortunately, I discovered this regression with -rc5 in the first place, so it's still there.
Ick, ok, Saravana, any thoughts? I know you're at the conference this week with me, maybe you can give Olof a hint as to what to look for here?
Quoted 5 lines
I'm not sure what you want me to look for. The patch turns on enforcement of DT contents that never used to be enforced, so now my computer no longer boots. And it does it in a way that makes it impossible for someone not rebuilding kernels to debug to figure out what happened.
Hi Olof,
Sorry for the trouble. It doesn't print any error messages because there are cases where it's block the probe where it wouldn't be an error. If I printed it every time
fw_devlinkblocked a probe, it'd be a ton of messages.Quoted 4 lines
Btw, when I enabled fw_devlink.strict=1, it was AFTER making changes that'll stop indefinitely blocking probes. So what you are seeing shouldn't be happening. After about 10 seconds (configurable), it should stop blocking the probes.
"Shouldn't be happening" is a pretty bold statement. It's not actually stuck on timeout in my case, and doesn't recover.
Instead, what seems to be happening is that the PCIe driver, which registers as a
platform_driverhere:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/pci/controller/mobiveil/pcie-layerscape-gen4.c#n255ends up registering, and the driver core now refuses to try to probe the device matches, since they no longer have their suppliers fulfilled (the smmu suppliers would not be tracked since they are optional here:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/of/property.c#n1449So what happens is that the driver registration succeeds, but there have been no devices matched to it. So when it returns to the platform core, it thinks there are no devices bound to this driver, so it should be unregistered:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/base/platform.c#n951
That explains why the pcie core doesn't retry and just disappears, and stops retrying.
This is what it looks like with
CONFIG_DEBUG_DRIVERandCONFIG_DEBUG_DEVRES:[ 5.178538] bus: 'platform': add driver layerscape-pcie-gen4 [ 5.184301] bus: 'platform': __driver_probe_device: matched device 3600000.pcie with driver layerscape-pcie-gen4 [ 5.194498] platform 3600000.pcie: error -EPROBE_DEFER: supplier 5000000.iommu not ready [ 5.202607] platform 3600000.pcie: Added to deferred list [ 5.208024] bus: 'platform': __driver_probe_device: matched device 3800000.pcie with driver layerscape-pcie-gen4 [ 5.218227] platform 3800000.pcie: error -EPROBE_DEFER: supplier 5000000.iommu not ready [ 5.226333] platform 3800000.pcie: Added to deferred list [ 5.231814] bus: 'platform': remove driver layerscape-pcie-gen4 [ 5.237761] driver: 'layerscape-pcie-gen4': driver_releaseNote that the platform driver registration sets flags to disable async probing, supposedly so it can assume that any matching devices would be found by the time registration returns:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/base/platform.c#n917
:/*
* We have to run our probes synchronously because we check if
* we find any devices to bind to and exit with error if there
* are any.
*/drv->driver.probe_type=PROBE_FORCE_SYNCHRONOUS;/*
* Prevent driver from requesting probe deferral to avoid further
* futile probe attempts.
*/drv->prevent_deferred_probe= true;
Quoted 3 lines
Bottom line: How was this code tested? This seems far from mature, this doesn't seem like that of an obscure condition to occur and it could create minefields for others down the road if it's fragile.
I've reverted it for now, let's get this worked out for later releases.
thanks,
greg k-h
Quoted 150 lines · Greg KH
Quoted 148 lines · Olof Johansson
Quoted 81 lines · Saravana Kannan
Quoted 66 lines · Olof Johansson
Hi,
Quoted 55 lines · Greg KH
Quoted 53 lines · Olof Johansson
Quoted 51 lines · Greg KH
Quoted 37 lines · Olof Johansson
Quoted 36 lines · Olof Johansson
Hi,
Quoted 4 lines · Greg KH
Saravana Kannan (11):
PM: domains: Delete usage ofdriver_deferred_probe_check_state()
pinctrl: devicetree: Delete usage ofdriver_deferred_probe_check_state()
net: mdio: Delete usage ofdriver_deferred_probe_check_state()Quoted 8 lines
driver core: Add
wait_for_init_devices_probehelper function
net: ipconfig: Relaxfw_devlinkif we need to mount a network rootfs Revert "driver core: Set defaultdeferred_probe_timeoutback to 0." driver core: Set fw_devlink.strict=1 by default
iommu/of: Delete usage ofdriver_deferred_probe_check_state()driver core: Deletedriver_deferred_probe_check_state()driver core:fw_devlink: Allow firmware to mark devices as best effort
of: base: Avoid console probe delay when fw_devlink.strict=1The last patch in this list regresses my HoneyComb LX2K (ironically the machine I do maintainer work on). It stops PCIe from probing, but without a single message indicating why.
The reason seems to be that the iommu-maps property doesn't get patched up by my (older) u-boot, and thus isn't a valid reference. System works fine without IOMMU, which is how I've ran it for a couple of years.
It's also extremely hard to diagnose out of the box because there are *no error messages*. And there were no warnings leading up to this strict enforcement.
This "feature" seems to have been done backwards. The checks should have been running (and not skipped due to the "optional" flag), but also not causing errors, just warnings. That would have given users a chance to know that this is something that needs to be fixed.
Quoted 8 lines
Quoted 5 lines
And when you flip the switch, at least report what failed so that people don't need to spend a whole night bisecting kernels, please.
Greg, mind reverting just the last one? If I hit this, I presume others would too.
Apologies, wrong patch pointed out. The culprit is "driver core: Set fw_devlink.strict=1 by default",
71066545b48e42.Is this still an issue in -rc5? A number of patches in the above series was just reverted and hopefully should have resolved the issue you are seeing.
Quoted 6 lines
Quoted 2 lines
Unfortunately, I discovered this regression with -rc5 in the first place, so it's still there.
Ick, ok, Saravana, any thoughts? I know you're at the conference this week with me, maybe you can give Olof a hint as to what to look for here?
Quoted 5 lines
I'm not sure what you want me to look for. The patch turns on enforcement of DT contents that never used to be enforced, so now my computer no longer boots. And it does it in a way that makes it impossible for someone not rebuilding kernels to debug to figure out what happened.
Hi Olof,
Sorry for the trouble. It doesn't print any error messages because there are cases where it's block the probe where it wouldn't be an error. If I printed it every time
fw_devlinkblocked a probe, it'd be a ton of messages.Quoted 4 lines
Btw, when I enabled fw_devlink.strict=1, it was AFTER making changes that'll stop indefinitely blocking probes. So what you are seeing shouldn't be happening. After about 10 seconds (configurable), it should stop blocking the probes.
"Shouldn't be happening" is a pretty bold statement. It's not actually stuck on timeout in my case, and doesn't recover.
Instead, what seems to be happening is that the PCIe driver, which registers as a
platform_driverhere:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/pci/controller/mobiveil/pcie-layerscape-gen4.c#n255ends up registering, and the driver core now refuses to try to probe the device matches, since they no longer have their suppliers fulfilled (the smmu suppliers would not be tracked since they are optional here:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/of/property.c#n1449So what happens is that the driver registration succeeds, but there have been no devices matched to it. So when it returns to the platform core, it thinks there are no devices bound to this driver, so it should be unregistered:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/base/platform.c#n951
That explains why the pcie core doesn't retry and just disappears, and stops retrying.
This is what it looks like with
CONFIG_DEBUG_DRIVERandCONFIG_DEBUG_DEVRES:[ 5.178538] bus: 'platform': add driver layerscape-pcie-gen4 [ 5.184301] bus: 'platform': __driver_probe_device: matched device 3600000.pcie with driver layerscape-pcie-gen4 [ 5.194498] platform 3600000.pcie: error -EPROBE_DEFER: supplier 5000000.iommu not ready [ 5.202607] platform 3600000.pcie: Added to deferred list [ 5.208024] bus: 'platform': __driver_probe_device: matched device 3800000.pcie with driver layerscape-pcie-gen4 [ 5.218227] platform 3800000.pcie: error -EPROBE_DEFER: supplier 5000000.iommu not ready [ 5.226333] platform 3800000.pcie: Added to deferred list [ 5.231814] bus: 'platform': remove driver layerscape-pcie-gen4 [ 5.237761] driver: 'layerscape-pcie-gen4': driver_releaseNote that the platform driver registration sets flags to disable async probing, supposedly so it can assume that any matching devices would be found by the time registration returns:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/base/platform.c#n917
:/*
* We have to run our probes synchronously because we check if
* we find any devices to bind to and exit with error if there
* are any.
*/drv->driver.probe_type=PROBE_FORCE_SYNCHRONOUS;/*
* Prevent driver from requesting probe deferral to avoid further
* futile probe attempts.
*/drv->prevent_deferred_probe= true;
Quoted 5 lines
Quoted 3 lines
Bottom line: How was this code tested? This seems far from mature, this doesn't seem like that of an obscure condition to occur and it could create minefields for others down the road if it's fragile.
I've reverted it for now, let's get this worked out for later releases.
Thanks Greg!
-Olof