Meta's resource control demo project[0] includes a benchmark tool that can
be used to calculate the best iocost solutions for a given SSD.
[0]: https://github.com/facebookexperimental/resctl-demo
A project[1] has now been started to create a publicly available database
of results that can be used to apply them automatically.
[1]: https://github.com/iocost-benchmark/iocost-benchmarks
This change adds a new tool that gets triggered by a udev rule for any
block device and queries the hwdb for known solutions. The format for
the hwdb file that is currently generated by the github action looks like
this:
# This file was auto-generated on Tue, 23 Aug 2022 13:03:57 +0000.
# From the following commit:
# ca82acfe93
#
# Match key format:
# block:<devpath>:name:<model name>:
# 12 points, MOF=[1.346,1.346], aMOF=[1.249,1.249]
block:*:name:HFS256GD9TNG-62A0A:fwver:*:
IOCOST_SOLUTIONS=isolation isolated-bandwidth bandwidth naive
IOCOST_MODEL_ISOLATION=rbps=1091439492 rseqiops=52286 rrandiops=63784 wbps=192329466 wseqiops=12309 wrandiops=16119
IOCOST_QOS_ISOLATION=rpct=0.00 rlat=8807 wpct=0.00 wlat=59023 min=100.00 max=100.00
IOCOST_MODEL_ISOLATED_BANDWIDTH=rbps=1091439492 rseqiops=52286 rrandiops=63784 wbps=192329466 wseqiops=12309 wrandiops=16119
IOCOST_QOS_ISOLATED_BANDWIDTH=rpct=0.00 rlat=8807 wpct=0.00 wlat=59023 min=100.00 max=100.00
IOCOST_MODEL_BANDWIDTH=rbps=1091439492 rseqiops=52286 rrandiops=63784 wbps=192329466 wseqiops=12309 wrandiops=16119
IOCOST_QOS_BANDWIDTH=rpct=0.00 rlat=8807 wpct=0.00 wlat=59023 min=100.00 max=100.00
IOCOST_MODEL_NAIVE=rbps=1091439492 rseqiops=52286 rrandiops=63784 wbps=192329466 wseqiops=12309 wrandiops=16119
IOCOST_QOS_NAIVE=rpct=99.00 rlat=8807 wpct=99.00 wlat=59023 min=75.00 max=100.00
The IOCOST_SOLUTIONS key lists the solutions available for that device
in the preferred order for higher isolation, which is a reasonable
default for most client systems. This can be overriden to choose better
defaults for custom use cases, like the various data center workloads.
The tool can also be used to query the known solutions for a specific
device or to apply a non-default solution (say, isolation or bandwidth).
Co-authored-by: Santosh Mahto <santosh.mahto@collabora.com>
In 5118e8e71d, the rules were changed to add
OPTIONS="string_escape=replace" to creation of
ENV{ID_SERIAL}="$env{ID_MODEL}_$env{ID_SERIAL_SHORT}", so that "/" would be
escaped. But this also changes how the symlink looks for devices that do not
have "/". This adds back the old symlink for compat, except when a slash
is present.
In the meantime, we changed the symlink format to include ${ND_NSID}. Since
the symlink with unescaped characters are older than that, for compat we
only need to cover the older type. (Symlinks without escaping and with ${ND_NSID}
were never created.) This makes it slightly easier on users: the non-deprecated
symlinks are with "_${ND_NSID}", so they are easier to distinguish.
Fixes#27155.
Mostly untested :( I only have a boring nvme device with no special characters
in the id, and the symlinks are unchanged for it by this patch.
The nvme by-id symlink changes to the latest namespace when a new namespace gets
added, for example by connecting multiple NVMe/TCP host controllers via nvme
connect-all.
That is incorrect for persistent device links.
The persistent symbolic device link should continue to point to the same NVMe
namespace throughout the lifetime of the current boot.
Therefore the namespace id needs to be added to the link name.
This adds symlinks that allow accessing loopback block devices via stable
names that reference their backing block devices, make the unpredictable
naming of loopback devices less of an issue.
Example:
1. Create a loopback block device for a file $F
losetup --find $F
2. Reference the backing block device via its inode:
L="$(stat -c '/dev/loop/by-inode/%Hd:%Ld-%i' $F)"
fdisk $L
In the above the loop device name (which might be /dev/loop47 or any
other name) is not used at all.
When built without blkid, then udev-builtin-blkid is not built,
and the verifier warns about the unknown builtin:
60-persistent-storage.rules:114 Unknown builtin command: blkid --hint=session_offset=$env{ID_CDROM_MEDIA_SESSION_LAST_OFFSET}
60-persistent-storage.rules:117 Unknown builtin command: blkid --noraid
60-persistent-storage.rules:120 Unknown builtin command: blkid
60-persistent-storage.rules: udev rules check failed
We would execute up to four hwdb match patterns (+ the keyboard builtin):
After the first hit, we would skip the other patterns, because of the GOTO="evdev_end"
action.
57bb707d48 (rules: Add extended evdev/input match
rules for event nodes with the same name), added an additional match with
":phys:<phys>:ev:<ev>" inserted. This breaks backwards compatibility for user
hwdb patterns, because we quit after the first match.
In general hwdb properties are "additive". We often have a general rule that
matches a wider class and then some specific overrides. E.g. in this particular
case, we have a match for all trackpoints, and then a bunch of model-specific
settings.
So let's change the rules to try all the match patterns and combine the
received properties. We execute builtin-keyboard once at the end, if there was
at least one match.
Fixes#25698. Fixes https://bugzilla.redhat.com/show_bug.cgi?id=2152226.
This also impacts other cases which I think would be very confusing for users.
Since we quit after a first successful match, if we had e.g. a match for
'evdev:input:b*v*p*' in out database, and the user added a match using
'evdev:name:*', which is the approach we document in the .hwdb files and which
users quite often use, it would be silently ignored. What's worse, if we added
our 'evdev:input:b*v*p*' match at a later point, user's match would stop
working. If we combine all the properties, we get more stable behaviour.
After 479da1107a, the usb_id builtin
command does not set ID_SERIAL if ID_BUS is already set.
Before the commit, all properties set based on pci bus were overwritten
by the usb_id, hence now it is sufficient setting them only when ID_BUS is
not set yet.
Fixes#25238.
Previously, ata_id might not be able to retrieve attributes correctly,
and properties from usb_id were used as a fallback. See issue #24921
and PR #24923. To keep backward compatibility, still we need to create
symlinks based on USB serial.
Fixes#25179.
If no module name is provided, then try to load modules based on the
device modealias.
Previously, MODALIAS property is passed as an argument, but it may
contain quotation. Hence, unfortunately the modalias may be modified
and cannot load expected modules.
Fixes#24715.
The uverbs devices are sequentially numbered and are not guarranteed to
stay stable across reboot.
At least one good person was disappointed by this, because they couldn't
find their device: https://bugzilla.redhat.com/show_bug.cgi?id=2036515
Let's add a few helpful links.
If multipath feature is enabled, nvme block devices may belong to the
"nvme-subsystem" subsystem, instead of "nvme" subsystem.
(What a confusing name...)
Then, the syspath is something like the following,
/sys/devices/virtual/nvme-subsystem/nvme-subsys0/nvme0n1
Hence, we need to find the 'real parent' device, such as
/sys/devices/pci0000:00/0000:00:1c.4/0000:3c:00.0/nvme/nvme0
Fixes https://bugzilla.redhat.com/show_bug.cgi?id=2031810.
Fixes https://bugzilla.redhat.com/show_bug.cgi?id=2124964.
Replaces #24748.
We can't get any FS meta-data from a suspended device. Hence defer
making any plugged/unplugged decisions, i.e. we just import whatever was
previous state and skip processing all other rules.
Thanks Lennart Poettering <lennart@poettering.net> for suggesting this
solution.
The cros-ec-accel devices report their mounting location by the 'label'
sysfs file only since Linux v6.0. With earlier kernels, a nonstandard
'location' file reports this, but slightly differently (lid instead of
display) [1].
Add udev rules to import the correct hwdb entries based on this
'location' file for cros-ec-accel devices, so that the base-mounted
accel matrix has the correct value for older kernels as well.
[1] https://kernel.org/doc/Documentation/ABI/testing/sysfs-bus-iio-cros-ec
The IIO subsystem exposes a 'label' sysfs file to help userspace better
identify its devices [1]. Standardized labels include the sensor type
along with its location, including 'accel-base' and 'accel-display'.
Most Chrome OS boards have two accelerometers that are indistinguishable
except for this label (or a 'location' sysfs file before Linux v6.0),
and need different mounting matrix corrections based on their location.
Add a udev rule that matches hwdb entries using this label, so we can
correct both accelerometers on these devices with hwdb entries. The
existing rules and hwdb entries are not modified to keep potential
out-of-tree entries working, but new entries in this form will override
existing ones. Also add currently standardized labels to parse-hwdb.py.
[1] https://www.kernel.org/doc/Documentation/ABI/testing/sysfs-bus-iio
This can be useful for users of slow block devices.
For example, the persistent-storage rules are needed for USB floppy
drives be recognized by udisks2, but the extra blkid calls cause
thrashing for 25+ seconds after every disk change.
With this change, a user wishing to avoid the extra blkid invocation(s)
could create /etc/udev/rules.d/55-floppy-noprobe.rules as follows:
# Don't probe PC floppy drives
SUBSYSTEM=="block", KERNEL=="fd*", \
ENV{UDEV_DISABLE_PERSISTENT_STORAGE_BLKID_FLAG}="1"
# Don't probe USB floppy drives
SUBSYSTEM=="block", SUBSYSTEMS=="usb", \
ATTRS{bInterfaceClass}=="08", ATTRS{bInterfaceSubClass}=="04", \
ENV{UDEV_DISABLE_PERSISTENT_STORAGE_BLKID_FLAG}="1"
I didn't exclude floppies by default in this change, because floppy
devices are also emulated by some BIOSes/hypervisors in some cases, and
I don't know how many systems would fail to boot if /dev/disk/by-uuid/*
became unavailable for 'floppy disks' on those systems.
GIT_VERSION is not available as a config.h variable, because it's rendered
into version.h during builds. Let's rework jinja2 rendering to also
parse version.h. No functional change, the new variable is so far unused.
I guess this will make partial rebuilds a bit slower, but it's useful
to be able to use the full version string.
This adds another symlink for block devices:
/dev/disk/by-diskseq/<number>
where the number is the diskseq number as exposed by the kernel. It's
useful for apps because they can use it to open a device by diskseq, in
a way that is safe against device node reuse. I.e. if a device node path
like this is passed to an app it could open the device node via the
symlink and also parse the diskseq from the path. Once the device is
opened it could compare the parsed diskseq with the one returned by
BLKGETDISKSEQ on the open node, and if it matches they know they are
talking to the right device.
Fixes: #22906
Some SCSI tape devices use the same device ID (NAA registered device
designator) for the SCSI tape changer device and the first actual tape
device. For example, this one:
https://docs.oracle.com/en/storage/tape-storage/storagetek-sl150-modular-tape-library/slofs/bridged-tape-drives.html
You must connect the bridged drive to an HBA supporting multiple
LUNs (also referred to as LUN scanning). The SL150 Library uses a
single SCSI ID and two logical unit numbers (LUN). LUN 0 controls
the tape drive and LUN 1 which is configured as a SCSI medium
changer device controls the robotics. Data is sent to the remaining
LUN on the bridged drive or to LUNs on the other, unbridged drives
in the partition, all of which are configured as SCSI
sequential-access (tape) devices.
This may lead to errors because /dev/tape/by-id symlinks may sometimes
point to the st device representing the tape, and sometimes to the sg
device representing the changer.
Fix this by assigning an increased priority to the tape device, and creating
a separate -changer link for the SCSI tape changer.
Co-developed-by: Martin Wilck <mwilck@suse.com>
The approach to use '''…'''.split() instead of a list of strings was initially
used when converting from automake because it allowed identical blocks of lines
to be used for both, making the conversion easier.
But over the years we have been using normal lists more and more, especially
when there were just a few filenames listed. This converts the rest.
No functional change.
This reverts commit 94cb45d57f.
This rule set up a duplicate import:
$ udevadm test /devices/pci0000:00/0000:00:14.0/usb2/2-4/2-4.1/2-4.1.3
...
2-4.1.3: /usr/lib/udev/rules.d/40-libgphoto2.rules:9 Importing properties from results of builtin command 'usb_id'
2-4.1.3: /usr/lib/udev/rules.d/50-udev-default.rules:13 Skipping builtin 'usb_id' in IMPORT key
2-4.1.3: /usr/lib/udev/rules.d/50-udev-default.rules:13 Importing properties from results of builtin command 'hwdb --subsystem=usb'
2-4.1.3: hwdb modalias key: "usb:v17EFp3054:OneLink+ Giga"
2-4.1.3: /usr/lib/udev/rules.d/50-udev-default.rules:15 Importing properties from results of builtin command 'hwdb 'usb:v17efp3054''
2-4.1.3: No entry found from hwdb.
2-4.1.3: /usr/lib/udev/rules.d/50-udev-default.rules:15 Failed to run builtin 'hwdb 'usb:v17efp3054'': No data available
2-4.1.3: /usr/lib/udev/rules.d/50-udev-default.rules:52 MODE 0664
except that the existing one was done with uppercase digits and the full match pattern,
and the second one was done with lowercase digits.
With the previous commit we only have uppercase digits in our match patterns, so we can
drop the duplicate import. (Some other projects might have rules that used the lowercase
match patterns, and people might have some local rules that did that too. But the second
import was only added recently so I think it's better to rip off the bandaid quickly.)
Enable /dev/sgx_vepc access for the group 'sgx', which allows KVM-backed VMs
to host Intel Software Guard eXtension (SGX) enclaves. The upcoming QEMU
6.2 uses /dev/sgx_vepc to reserve portions of Enclave Page Cache (EPC) for
VMs. EPC is the reserved physical memory used for hosting enclaves.
When using "capture : true" in custom_target()s the mode of the source
file is not preserved when the generated file is not installed and so
needs to be tweaked manually. Switch from output capture to creating the
target file and copy the permissions from the input file.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
systemd-udevd.service listens to kernel uevents and is needed for device
units to be available.
systemd-udevd.service is misspelled as systemd-udev.service in a couple places.
Fixing typo.
