The SPL kmem_cache implementation provides a mechanism, `skc_reclaim`,
whereby individual caches can register a callback to be invoked when
there is memory pressure. This mechanism is used in only one place: the
ARC registers the `hdr_recl()` reclaim function. This function wakes up
the `arc_reap_zthr`, whose job is to call `kmem_cache_reap()` and
`arc_reduce_target_size()`.
The `skc_reclaim` callbacks are invoked only by shrinker callbacks and
`arc_reap_zthr`, and only callback only wakes up `arc_reap_zthr`. When
called from `arc_reap_zthr`, waking `arc_reap_zthr` is a no-op. When
called from shrinker callbacks, we are already aware of memory pressure
and responding to it. Therefore there is little benefit to ever calling
the `hdr_recl()` `skc_reclaim` callback.
The `arc_reap_zthr` also wakes once a second, and if memory is low when
allocating an ARC buffer. Therefore, additionally waking it from the
shrinker calbacks has little benefit.
The shrinker callbacks can be invoked very frequently, e.g. 10,000 times
per second. Additionally, for invocation of the shrinker callback,
skc_reclaim is invoked many times. Therefore, this mechanism consumes
significant amounts of CPU time.
The kmem_cache shrinker calls `spl_kmem_cache_reap_now()`, which,
in addition to invoking `skc_reclaim()`, does two things to attempt to
free pages for use by the system:
1. Return free objects from the magazine layer to the slab layer
2. Return entirely-free slabs to the page layer (i.e. free pages)
These actions apply only to caches implemented by the SPL, not those
that use the underlying kernel SLAB/SLUB caches. The SPL caches are
used for objects >=32KB, which are primarily linear ABD's cached in the
DBUF cache.
These actions (freeing objects from the magazine layer and returning
entirely-free slabs) are also taken whenever a `kmem_cache_free()` call
finds a full magazine. So there would typically be zero entirely-free
slabs, and the number of objects in magazines is limited (typically no
more than 64 objects per magazine, and there's one magazine per CPU).
Therefore the benefit of `spl_kmem_cache_reap_now()`, while nonzero, is
modest.
We also call `spl_kmem_cache_reap_now()` from the `arc_reap_zthr`, when
memory pressure is detected. Therefore, calling
`spl_kmem_cache_reap_now()` from the kmem_cache shrinker is not needed.
This commit removes the `skc_reclaim` mechanism, its only callback
`hdr_recl()`, and the kmem_cache shrinker callback.
Reviewed-By: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Wilson <gwilson@delphix.com>
Reviewed-by: Pavel Zakharov <pavel.zakharov@delphix.com>
Signed-off-by: Matthew Ahrens <mahrens@delphix.com>
Closes#10576
Move/add include of <linux/percpu_compat.h> to satisfy missing
requirements.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Romain Dolbeau <romain@dolbeau.org>
Closes#10568Closes#10569
OS-specific code (e.g. under `module/os/linux`) does not need to share
its code structure with any other operating systems. In particular, the
ARC and kmem code need not be similar to the code in illumos, because we
won't be syncing this OS-specific code between operating systems. For
example, if/when illumos support is added to the common repo, we would
add a file `module/os/illumos/zfs/arc_os.c` for the illumos versions of
this code.
Therefore, we can simplify the code in the OS-specific ARC and kmem
routines.
These changes do not impact system behavior, they are purely code
cleanup. The changes are:
Arenas are not used on Linux or FreeBSD (they are always `NULL`), so
`heap_arena`, `zio_arena`, and `zio_alloc_arena` can be removed, along
with code that uses them.
In `arc_available_memory()`:
* `desfree` is unused, remove it
* rename `freemem` to avoid conflict with pre-existing `#define`
* remove checks related to arenas
* use units of bytes, rather than converting from bytes to pages and
then back to bytes
`SPL_KMEM_CACHE_REAP` is unused, remove it.
`skc_reap` is unused, remove it.
The `count` argument to `spl_kmem_cache_reap_now()` is unused, remove
it.
`vmem_size()` and associated type and macros are unused, remove them.
In `arc_memory_throttle()`, use a less confusing variable name to store
the result of `arc_free_memory()`.
Reviewed-by: George Wilson <gwilson@delphix.com>
Reviewed-by: Pavel Zakharov <pavel.zakharov@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Ryan Moeller <ryan@ixsystems.com>
Signed-off-by: Matthew Ahrens <mahrens@delphix.com>
Closes#10499
The SPL provides a wrapper for the kernel's shrinker callbacks, which
enables the ZFS code to interface with multiple versions of the shrinker
API's from different kernel versions. Specifically, Linux kernels 3.0 -
3.11 has a single "combined" callback, and Linux kernels 3.12 and later
have two "split" callbacks. The SPL provides a wrapper function so that
the ZFS code only needs to implement one version of the callbacks.
Currently the SPL's wrappers are designed such that the ZFS code
implements the older, "combined" callback. There are a few downsides to
this approach:
* The general design within ZFS is for the latest Linux kernel to be
considered the "first class" API.
* The newer, "split" callback API is easier to understand, because each
callback has one purpose.
* The current wrappers do not completely abstract out the differing
API's, so ZFS code needs `#ifdef` code to handle the differing return
values required for different kernel versions.
This commit addresses these drawbacks by having the ZFS code provide the
latest, "split" callbacks, and the SPL provides a wrapping function for
the older, "combined" API.
Reviewed-by: Pavel Zakharov <pavel.zakharov@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Matthew Ahrens <mahrens@delphix.com>
Closes#10502
A previous commit enabled the tracking of object allocations
in Linux-backed caches from the SPL layer for debuggability.
The commit is: 9a170fc6fe
Unfortunately, it also introduced minor performance regressions
that were highlighted by the ZFS perf test-suite. Within Delphix
we found that the regression would be from -1%, all the way up
to -8% for some workloads.
This commit brings performance back up to par by creating a
separate counter for those caches and making it a percpu in
order to avoid lock-contention.
The initial performance testing was done by myself, and the
final round was conducted by @tonynguien who was also the one
that discovered the regression and highlighted the culprit.
Reviewed-by: Matt Ahrens <matt@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#10397
The `pgprot` argument has been removed from `__vmalloc` in Linux 5.8,
being `PAGE_KERNEL` always now [1].
Detect this during configure and define a wrapper for older kernels.
[1] https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/mm/vmalloc.c?h=next-20200605&id=88dca4ca5a93d2c09e5bbc6a62fbfc3af83c4fca
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Co-authored-by: Sebastian Gottschall <s.gottschall@dd-wrt.com>
Co-authored-by: Michael Niewöhner <foss@mniewoehner.de>
Signed-off-by: Sebastian Gottschall <s.gottschall@dd-wrt.com>
Signed-off-by: Michael Niewöhner <foss@mniewoehner.de>
Closes#10422
When growing the size of a (VMEM or KVMEM) kmem cache, spl_cache_grow()
always does taskq_dispatch(spl_cache_grow_work), and then waits for the
KMC_BIT_GROWING to be cleared by the taskq thread.
The taskq thread (spl_cache_grow_work()) does:
1. allocate new slab and add to list
2. wake_up_all(skc_waitq)
3. clear_bit(KMC_BIT_GROWING)
Therefore, the waiting thread can wake up before GROWING has been
cleared. It will see that the growing has not yet completed, and go
back to sleep until it hits the 100ms timeout.
This can have an extreme performance impact on workloads that alloc/free
more than fits in the (statically-sized) magazines. These workloads
allocate and free slabs with high frequency.
The problem can be observed with `funclatency spl_cache_grow`, which on
some workloads shows that 99.5% of the time it takes <64us to allocate
slabs, but we spend ~70% of our time in outliers, waiting for the 100ms
timeout.
The fix is to do `clear_bit(KMC_BIT_GROWING)` before
`wake_up_all(skc_waitq)`.
A future investigation should evaluate if we still actually need to
taskq_dispatch() at all, and if so on which kernel versions.
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Reviewed-by: Pavel Zakharov <pavel.zakharov@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Wilson <gwilson@delphix.com>
Signed-off-by: Matthew Ahrens <mahrens@delphix.com>
Closes#9989
On kernels with KASAN enabled the following failure can be observed as
soon as the zfs module is loaded:
VERIFY(IS_P2ALIGNED(ptr, PAGE_SIZE)) failed
PANIC at spl-kmem-cache.c:228:kv_alloc()
The problem is kmalloc() has never guaranteed aligned allocations; this
requirement resulted in zfsonlinux/spl@8b45dda which removed all
kmalloc() usage in kv_alloc().
Until a GFP_ALIGNED flag (or equivalent functionality) is provided by
the kernel this commit partially reverts 66955885 and 6d948c35 to
prevent k(v)malloc() allocations in kv_alloc().
Reviewed-by: Kjeld Schouten <kjeld@schouten-lebbing.nl>
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes#9813
As of cppcheck 1.82 warnings are issued when using the list_for_each_*
functions with an uninitialized variable. Functionally, this is fine
but to resolve the warning initialize these variables.
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#9732
This adds a new KMC_KVMEM flag was added to enforce use of the
kvmalloc allocator in kmem_cache_create even for large blocks, which
may also increase performance in some specific cases (e.g. zstd), too.
Default to KVMEM instead of VMEM in spl_kmem_cache_create.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matt Ahrens <matt@delphix.com>
Signed-off-by: Sebastian Gottschall <s.gottschall@dd-wrt.com>
Signed-off-by: Michael Niewöhner <foss@mniewoehner.de>
Closes#9034
This patch implements use of kvmalloc for GFP_KERNEL allocations, which
may increase performance if the allocator is able to allocate physical
memory, if kvmalloc is available as a public kernel interface (since
v4.12). Otherwise it will simply fall back to virtual memory (vmalloc).
Also fix vmem_alloc implementation which can lead to slow allocations
since the first attempt with kmalloc does not make use of the noretry
flag but tells the linux kernel to retry several times before it fails.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matt Ahrens <matt@delphix.com>
Signed-off-by: Sebastian Gottschall <s.gottschall@dd-wrt.com>
Signed-off-by: Michael Niewöhner <foss@mniewoehner.de>
Closes#9034
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matt Ahrens <matt@delphix.com>
Signed-off-by: Sebastian Gottschall <s.gottschall@dd-wrt.com>
Signed-off-by: Michael Niewöhner <foss@mniewoehner.de>
Closes#9034
Increase the minimum supported kernel version from 2.6.32 to 3.10.
This removes support for the following Linux enterprise distributions.
Distribution | Kernel | End of Life
---------------- | ------ | -------------
Ubuntu 12.04 LTS | 3.2 | Apr 28, 2017
SLES 11 | 3.0 | Mar 32, 2019
RHEL / CentOS 6 | 2.6.32 | Nov 30, 2020
The following changes were made as part of removing support.
* Updated `configure` to enforce a minimum kernel version as
specified in the META file (Linux-Minimum: 3.10).
configure: error:
*** Cannot build against kernel version 2.6.32.
*** The minimum supported kernel version is 3.10.
* Removed all `configure` kABI checks and matching C code for
interfaces which solely predate the Linux 3.10 kernel.
* Updated all `configure` kABI checks to fail when an interface is
missing which was in the 3.10 kernel up to the latest 5.1 kernel.
Removed the HAVE_* preprocessor defines for these checks and
updated the code to unconditionally use the verified interface.
* Inverted the detection logic in several kABI checks to match
the new interface as it appears in 3.10 and newer and not the
legacy interface.
* Consolidated the following checks in to individual files. Due
the large number of changes in the checks it made sense to handle
this now. It would be desirable to group other related checks in
the same fashion, but this as left as future work.
- config/kernel-blkdev.m4 - Block device kABI checks
- config/kernel-blk-queue.m4 - Block queue kABI checks
- config/kernel-bio.m4 - Bio interface kABI checks
* Removed the kABI checks for sops->nr_cached_objects() and
sops->free_cached_objects(). These interfaces are currently unused.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#9566
Currently, for certain sizes and classes of allocations we use
SPL caches that are backed by caches in the Linux Slab allocator
to reduce fragmentation and increase utilization of memory. The
way things are implemented for these caches as of now though is
that we don't keep any statistics of the allocations that we
make from these caches.
This patch enables the tracking of allocated objects in those
SPL caches by making the trade-off of grabbing the cache lock
at every object allocation and free to update the respective
counter.
Additionally, this patch makes those caches visible in the
/proc/spl/kmem/slab special file.
As a side note, enabling the specific counter for those caches
enables SDB to create a more user-friendly interface than
/proc/spl/kmem/slab that can also cross-reference data from
slabinfo. Here is for example the output of one of those
caches in SDB that outputs the name of the underlying Linux
cache, the memory of SPL objects allocated in that cache,
and the percentage of those objects compared to all the
objects in it:
```
> spl_kmem_caches | filter obj.skc_name == "zio_buf_512" | pp
name ... source total_memory util
----------- ... ----------------- ------------ ----
zio_buf_512 ... kmalloc-512[SLUB] 16.9MB 8
```
Reviewed-by: Matt Ahrens <matt@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#9474
Make the metaslab platform agnostic again by adding
accessor functions which can be implemented by each
platform.
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Jorgen Lundman <lundman@lundman.net>
Reviewed-by: Ryan Moeller <ryan@ixsystems.com>
Signed-off-by: Matt Macy <mmacy@FreeBSD.org>
Closes#9404
Move platform specific Linux source under module/os/linux/
and update the build system accordingly. Additional code
restructuring will follow to make the common code fully
portable.
Reviewed-by: Jorgen Lundman <lundman@lundman.net>
Reviewed-by: Igor Kozhukhov <igor@dilos.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Matthew Macy <mmacy@FreeBSD.org>
Closes#9206
Matthew Ahrens
Romain Dolbeau
Serapheim Dimitropoulos
Michael Niewöhner
Matthew Ahrens
loli10K
Ubuntu
Michael Niewöhner
Michael Niewöhner
Brian Behlendorf
Matthew Macy
Matthew Macy