We need to directly call __init_rwsem() or the name gets expanded
to SEM(lock-name). This is safe and correct for the support arches
x86/x86_64/ppc/ppc64.
For a generic explanation of why mutexs needed to be reimplemented
to work with the kernel lock profiling see commits:
e811949a57 and
d28db80fd0
The specific changes made to the mutex implemetation are as follows.
The Linux mutex structure is now directly embedded in the kmutex_t.
This allows a kmutex_t to be directly case to a mutex struct and
passed directly to the Linux primative.
Just like with the rwlocks it is critical that these functions be
implemented as '#defines to ensure the location information is
preserved. The preprocessor can then do a direct replacement of
the Solaris primative with the linux primative.
Just as with the rwlocks we need to track the lock owner. Here
things get a little more interesting because depending on your
kernel version, and how you've built your kernel Linux may already
do this for you. If your running a 2.6.29 or newer kernel on a
SMP system the lock owner will be tracked. This was added to Linux
to support adaptive mutexs, more on that shortly. Alternately, your
kernel might track the lock owner if you've set CONFIG_DEBUG_MUTEXES
in the kernel build. If neither of the above things is true for
your kernel the kmutex_t type will include and track the lock owner
to ensure correct behavior. This is all handled by a new autoconf
check called SPL_AC_MUTEX_OWNER.
Concerning adaptive mutexs these are a very recent development and
they did not make it in to either the latest FC11 of SLES11 kernels.
Ideally, I'd love to see this kernel change appear in one of these
distros because it does help performance. From Linux kernel commit:
0d66bf6d3514b35eb6897629059443132992dbd7
"Testing with Ingo's test-mutex application...
gave a 345% boost for VFS scalability on my testbox"
However, if you don't want to backport this change yourself you
can still simply export the task_curr() symbol. The kmutex_t
implementation will use this symbol when it's available to
provide it's own adaptive mutexs.
Finally, DEBUG_MUTEX support was removed including the proc handlers.
This was done because now that we are cleanly integrated with the
kernel profiling all this information and much much more is available
in debug kernel builds. This code was now redundant.
Update mutexs validated on:
- SLES10 (ppc64)
- SLES11 (x86_64)
- CHAOS4.2 (x86_64)
- RHEL5.3 (x86_64)
- RHEL6 (x86_64)
- FC11 (x86_64)
The behavior of RW_*_HELD was updated because it was not quite right.
It is not sufficient to return non-zero when the lock is help, we must
only do this when the current task in the holder.
This means we need to track the lock owner which is not something
tracked in a Linux semaphore. After some experimentation the
solution I settled on was to embed the Linux semaphore at the start
of a larger krwlock_t structure which includes the owner field.
This maintains good performance and allows us to cleanly intergrate
with the kernel lock analysis tools. My reasons:
1) By placing the Linux semaphore at the start of krwlock_t we can
then simply cast krwlock_t to a rw_semaphore and pass that on to
the linux kernel. This allows us to use '#defines so the preprocessor
can do direct replacement of the Solaris primative with the linux
equivilant. This is important because it then maintains the location
information for each rw_* call point.
2) Additionally, by adding the owner to krwlock_t we can keep this
needed extra information adjacent to the lock itself. This removes
the need for a fancy lookup to get the owner which is optimal for
performance. We can also leverage the existing spin lock in the
semaphore to ensure owner is updated correctly.
3) All helper functions which do not need to strictly be implemented
as a define to preserve location information can be done as a static
inline function.
4) Adding the owner to krwlock_t allows us to remove all memory
allocations done during lock initialization. This is good for all
the obvious reasons, we do give up the ability to specific the lock
name. The Linux profiling tools will stringify the lock name used
in the code via the preprocessor and use that.
Update rwlocks validated on:
- SLES10 (ppc64)
- SLES11 (x86_64)
- CHAOS4.2 (x86_64)
- RHEL5.3 (x86_64)
- RHEL6 (x86_64)
- FC11 (x86_64)
It turns out that the previous rwlock implementation worked well but
did not integrate properly with the upstream kernel lock profiling/
analysis tools. This is a major problem since it would be awfully
nice to be able to use the automatic lock checker and profiler.
The problem is that the upstream lock tools use the pre-processor
to create a lock class for each uniquely named locked. Since the
rwsem was embedded in a wrapper structure the name was always the
same. The effect was that we only ended up with one lock class for
the entire SPL which caused the lock dependency checker to flag
nearly everything as a possible deadlock.
The solution was to directly map a krwlock to a Linux rwsem using
a typedef there by eliminating the wrapper structure. This was not
done initially because the rwsem implementation is specific to the arch.
To fully implement the Solaris krwlock API using only the provided rwsem
API is not possible. It can only be done by directly accessing some of
the internal data member of the rwsem structure.
For example, the Linux API provides a different function for dropping
a reader vs writer lock. Whereas the Solaris API uses the same function
and the caller does not pass in what type of lock it is. This means to
properly drop the lock we need to determine if the lock is currently a
reader or writer lock. Then we need to call the proper Linux API function.
Unfortunately, there is no provided API for this so we must extracted this
information directly from arch specific lock implementation. This is
all do able, and what I did, but it does complicate things considerably.
The good news is that in addition to the profiling benefits of this
change. We may see performance improvements due to slightly reduced
overhead when creating rwlocks and manipulating them.
The only function I was forced to sacrafice was rw_owner() because this
information is simply not stored anywhere in the rwsem. Luckily this
appears not to be a commonly used function on Solaris, and it is my
understanding it is mainly used for debugging anyway.
In addition to the core rwlock changes, extensive updates were made to
the rwlock regression tests. Each class of test was extended to provide
more API coverage and to be more rigerous in checking for misbehavior.
This is a pretty significant change and with that in mind I have been
careful to validate it on several platforms before committing. The full
SPLAT regression test suite was run numberous times on all of the following
platforms. This includes various kernels ranging from 2.6.16 to 2.6.29.
- SLES10 (ppc64)
- SLES11 (x86_64)
- CHAOS4.2 (x86_64)
- RHEL5.3 (x86_64)
- RHEL6 (x86_64)
- FC11 (x86_64)
The run time stack overflow checking is being disabled by default
because it is not safe for use with 2.6.29 and latter kernels. These
kernels do now have their own stack overflow checking so this support
has become redundant anyway. It can be re-enabled for older kernels or
arches without stack overflow checking by redefining CHECK_STACK().
Basically everything we need to monitor the global memory state of
the system is now cleanly available via global_page_state(). The
problem is that this interface is still fairly recent, and there
has been one change in the page state enum which we need to handle.
These changes basically boil down to the following:
- If global_page_state() is available we should use it. Several
autoconf checks have been added to detect the correct enum names.
- If global_page_state() is not available check to see if
get_zone_counts() symbol is available and use that.
- If the get_zone_counts() symbol is not exported we have no choice
be to dynamically aquire it at load time. This is an absolute
last resort for old kernel which we don't want to patch to
cleanly export the symbol.
The previous credential implementation simply provided the needed types and
a couple of dummy functions needed. This update correctly ties the basic
Solaris credential API in to one of two Linux kernel APIs.
Prior to 2.6.29 the linux kernel embeded all credentials in the task
structure. For these kernels, we pass around the entire task struct as if
it were the credential, then we use the helper functions to extract the
credential related bits.
As of 2.6.29 a new credential type was added which we can and do fairly
cleanly layer on top of. Once again the helper functions nicely hide
the implementation details from all callers.
Three tests were added to the splat test framework to verify basic
correctness. They should be extended as needed when need credential
functions are added.
used to scale the number of threads based on the number of online
CPUs. As CPUs are added/removed we should rescale the thread
count appropriately, but currently this is only done at create.
- Allow checking for exported symbols in both Module.symvers
and Module.symvers. My stock SLES kernel ships an objects
directory with Module.symvers, yet produces a Module.symvers
in the local build directory.
- Properly honor --prefix in build system and rpm spec file.
- Add '--define require_kdir' to spec file to support building
rpms against kernel sources installed in non-default locations.
- Add '--define require_kobj' to spec file to support building
rpms against kernel object installed in non-default locations.
- Stop suppressing errors in autogen.sh script.
- Improved logic to detect missing kernel objects when they are
not located with the source. This is the common case for SLES
as well as in-tree chaos kernel builds and is done to simply
support for multiple arches.
- Moved spl-devel build products to /usr/src/spl-<version>, a
spl symlink is created to reference the last installed version.
- Proper ioctl() 32/64-bit binary compatibility. We need to ensure the
ioctl data itself is always packed the same for 32/64-bit binaries.
Additionally, the correct thing to do is encode this size in bytes
as part of the command using _IOC_SIZE().
- Minor formatting changes to respect the 80 character limit.
- Move all SPLAT_SUBSYSTEM_* defines in to splat-ctl.h.
- Increase SPLAT_SUBSYSTEM_UNKNOWN because we were getting close
to accidentally using it for a real registered subsystem.
- Initial SLES testing uncovered a long standing bug in the debug
tracing. The tcd_for_each() macro expected a NULL to terminate
the trace_data[i] array but this was only ever true due to luck.
All trace_data[] iterators are now properly capped by TCD_TYPE_MAX.
- SPLAT_MAJOR 229 conflicted with a 'hvc' device on my SLES system.
Since this was always an arbitrary choice I picked something else.
- The HAVE_PGDAT_LIST case should set pgdat_list_addr to the value stored
at the address of the memory location returned by kallsyms_lookup_name().
- Prior to 2.6.17 there were no *_pgdat helper functions in mm/mmzone.c.
Instead for_each_zone() operated directly on pgdat_list which may or
may not have been exported depending on how your kernel was compiled.
Now new configure checks determine if you have the helpers or not, and
if the needed symbols are exported. If they are not exported then they
are dynamically aquired at runtime by kallsyms_lookup_name().
- Enable builds for powerpc ISA type.
- Add DIV_ROUND_UP and roundup macros if unavailable.
- Cast 64-bit values for %lld format string to (long long) to
quiet compile warning.
- Configure check for mutex_lock_nested(). This function was introduced
as part of the mutex validator in 2.6.18, but if it's unavailable then
it's safe to fallback to a plain mutex_lock().
- Configure check, the div64_64() function was renamed to
div64_u64() as of 2.6.26.
- Configure check, the global_page_state() fuction was introduced
in 2.6.18 kernels. The earlier 2.6.16 based SLES10 must not try
and use it, thankfully get_zone_counts() is still available.
- To simplify debugging poison all symbols aquired dynamically
using spl_kallsyms_lookup_name() with SYMBOL_POISON.
- Add console messages when the user mode helpers fail.
- spl_kmem_init_globals() use bit shifts instead of division.
- When the monotonic clock is unavailable __gethrtime() must perform
the HZ division as an 'unsigned long long' because the SPL only
implements __udivdi3(), and not __divdi3() for 'long long' division
on 32-bit arches.
We need dependent packages to be able to include spl_config.h so they
can leverage the configure checks the SPL has done. This is important
because several of the spl headers need the results of these checks to
work properly. Unfortunately, the autoheader build product is always
private to a particular build and defined certain common things.
(PACKAGE, VERSION, etc). This prevents other packages which also use
autoheader from being include because the definitions conflict. To
avoid this problem the SPL build system leverage AH_BOTTOM to include
a spl_unconfig.h at the botton of the autoheader build product. This
custom include undefs all known shared symbols to prevent the confict.
This does however mean that those definition are also not availble
to the SPL package either. The SPL package therefore uses the
equivilant SPL_META_* definitions.
In the interests of portability I have added a FC10/i686 box to
my list of development platforms. The hope is this will allow me
to keep current with upstream kernel API changes, and at the same
time ensure I don't accidentally break x86 support. This patch
resolves all remaining issues observed under that environment.
1) SPL_AC_ZONE_STAT_ITEM_FIA autoconf check added. As of 2.6.21
the kernel added a clean API for modules to get the global count
for free, inactive, and active pages. The SPL attempts to detect
if this API is available and directly map spl_global_page_state()
to global_page_state(). If the full API is not available then
spl_global_page_state() is implemented as a thin layer to get
these values via get_zone_counts() if that symbol is available.
2) New kmem:vmem_size regression test added to validate correct
vmem_size() functionality. The test case acquires the current
global vmem state, allocates from the vmem region, then verifies
the allocation is correctly reflected in the vmem_size() stats.
3) Change splat_kmem_cache_thread_test() to always use KMC_KMEM
based memory. On x86 systems with limited virtual address space
failures resulted due to exhaustig the address space. The tests
really need to problem exhausting all memory on the system thus
we need to use the physical address space.
4) Change kmem:slab_lock to cap it's memory usage at availrmem
instead of using the native linux nr_free_pages(). This provides
additional test coverage of the SPL Linux VM integration.
5) Change kmem:slab_overcommit to perform allocation of 256K
instead of 1M. On x86 based systems it is not possible to create
a kmem backed slab with entires of that size. To compensate for
this the number of allocations performed in increased by 4x.
6) Additional autoconf documentation for proposed upstream API
changes to make additional symbols available to modules.
7) Console error messages added when spl_kallsyms_lookup_name()
fails to locate an expected symbol. This causes the module to fail
to load and we need to know exactly which symbol was not available.
I'm very surprised this has not surfaced until now. But the taskq_wait()
implementation work only wait successfully the first time it was called.
Subsequent usage of taskq_wait() on the taskq would not wait.
The issue was caused by tq->tq_lowest_id being set to MAX_INT after the
first wait completed. This caused subsequent waits which check that the
waiting id is less than the lowest taskq id to always succeed. The fix
is to ensure that tq->tq_lowest_id is never set larger than tq->tq_next.id.
Additional fixes which were added to this patch include:
1) Fix a race by placing the taskq_wait_check() in the tq->tq_lock spinlock.
2) taskq_wait() should wait for the largest outstanding id.
3) Multiple spelling corrections.
4) Added taskq wait regression test to validate correct behavior.
As of 2.6.27 kernels the device_create() API changed to include
a private data argument. This check detects which version of
device_create() function the kernel has and properly defines
spl_device_create() to use the correct prototype.
An update to the build system to properly support all commonly
used Makefile targets these include:
make all # Build everything
make install # Install everything
make clean # Clean up build products
make distclean # Clean up everything
make dist # Create package tarball
make srpm # Create package source RPM
make rpm # Create package binary RPMs
make tags # Create ctags and etags for everything
Extra care was taken to ensure that the source RPMs are fully
rebuildable against Fedora/RHEL/Chaos kernels. To build binary
RPMs from the source RPM for your system simply run:
rpmbuild --rebuild spl-x.y.z-1.src.rpm
This will produce two binary RPMs with correct 'requires'
dependencies for your kernel. One will contain all spl modules
and support utilities, the other is a devel package for compiling
additional kernel modules which are dependant on the spl.
spl-x.y.z-1_<kernel version>.x86_64.rpm
spl-devel-x.y.2-1_<kernel version>.x86_64.rpm
Remove all instances of functions being reimplemented in the SPL.
When the prototypes are available in the linux headers but the
function address itself is not exported use kallsyms_lookup_name()
to find the address. The function name itself can them become a
define which calls a function pointer. This is preferable to
reimplementing the function in the SPL because it ensures we get
the correct version of the function for the running kernel. This
is actually pretty safe because the prototype is defined in the
headers so we know we are calling the function properly.
This patch also includes a rhel5 kernel patch we exports the needed
symbols so we don't need to use kallsyms_lookup_name(). There are
autoconf checks to detect if the symbol is exported and if so to
use it directly. We should add patches for stock upstream kernels
as needed if for no other reason than so we can easily track which
additional symbols we needed exported. Those patches can also be
used by anyone willing to rebuild their kernel, but this should
not be a requirement. The rhel5 version of the export-symbols
patch has been applied to the chaos kernel.
Additional fixes:
1) Implement vmem_size() function using get_vmalloc_info()
2) SPL_CHECK_SYMBOL_EXPORT macro updated to use $LINUX_OBJ instead
of $LINUX because Module.symvers is a build product. When
$LINUX_OBJ != $LINUX we will not properly detect exported symbols.
3) SPL_LINUX_COMPILE_IFELSE macro updated to add include2 and
$LINUX/include search paths to allow proper compilation when
the kernel target build directory is not the source directory.
Minimal support added for the zone_get_hostid() function. Only
global zones are supported therefore this function must be called
with a NULL argumment. Additionally, I've added the HW_HOSTID_LEN
define and updated all instances where a hard coded magic value
of 11 was used; "A good riddance of bad rubbish!"
Because vmem_free() was implemented as a macro using the ','
operator to evaluate both arguments and we performed the free
before evaluating size we would deference the free'd pointer.
To resolve the problem we just invert the ordering and evaluate
size first just as if it was evaluated by the caller when being
passed to this function. This ensure that if the caller is
doing something reckless like performing an assignment as
part of the size argument we still perform it and it simply
doesn't get removed by the macro. Oh course nobody should
be doing this sort of thing, but just in case.
When SPLAT_TEST_INIT() initialized SPLAT_KMEM_TEST11_NAME the short
short test name overran the static length buffer of SPLAT_NAME_SIZE.
This was fixed by increasing the buffer length from 16 to 20 bytes.
- The previous magazine ageing sceme relied on the on_each_cpu()
function to call spl_magazine_age() on each cpu. It turns out
this could deadlock with do_flush_tlb_all() which also relies
on the IPI based on_each_cpu(). To avoid this problem a per-
magazine delayed work item is created and indepentantly
scheduled to the correct cpu removing the need for on_each_cpu().
- Additionally two unused fields were removed from the type
spl_kmem_cache_t, they were hold overs from previous cleanup.
- struct work_struct work
- struct timer_list timer
- Default SPL_KMEM_CACHE_DELAY changed to 15 to match Solaris.
- Aged out slab checking occurs every SPL_KMEM_CACHE_DELAY / 3.
- skc->skc_reap tunable added whichs allows callers of
spl_slab_reclaim() to cap the number of slabs reclaimed.
On Solaris all eligible slabs are always reclaimed, and this
is still the default behavior. However, I suspect that is
not always wise for reasons such as in the next comment.
- spl_slab_reclaim() added cond_resched() while walking the
slab/object free lists. Soft lockups were observed when
freeing large numbers of vmalloc'd slabs/objets.
- spl_slab_reclaim() 'sks->sks_ref > 0' check changes from
incorrect 'break' to 'continue' to ensure all slabs are
checked.
- spl_cache_age() reworked to avoid a deadlock with
do_flush_tlb_all() which occured because we slept waiting
for completion in spl_cache_age(). To waiting for magazine
reclamation to finish is not required so we no longer wait.
- spl_magazine_create() and spl_magazine_destroy() shifted
back to using for_each_online_cpu() instead of the
spl_on_each_cpu() approach which was of course a bad idea
due to memory allocations which Ricardo pointed out.
Added support for Solaris swapfs_minfree, and swapfs_reserve tunables.
In additional availrmem is now available and return a reasonable value
which is reasonably analogous to the Solaris meaning. On linux we
return the sun of free and inactive pages since these are all easily
reclaimable.
All tunables are available in /proc/sys/kernel/spl/vm/* and they may
need a little adjusting once we observe the real behavior. Some of
the defaults are mapped to similar linux counterparts, others are
straight from the OpenSolaris defaults.
Support added to provide reasonable values for the global Solaris
VM variables: minfree, desfree, lotsfree, needfree. These values
are set to the sum of their per-zone linux counterparts which
should be close enough for Solaris consumers.
When a non-GPL app links against the SPL we cannot use the udev
interfaces, which means non of the device special files are created.
Because of this I had added a poor mans udev which cause the SPL
to invoke an upcall and create the basic devices when a minor
is registered. When a minor is unregistered we use the vnode
interface to unlink the special file.
- Added SPL_AC_3ARGS_ON_EACH_CPU configure check to determine
if the older 4 argument version of on_each_cpu() should be
used or the new 3 argument version. The retry argument was
dropped in the new API which was never used anyway.
- Updated work queue compatibility wrappers. The old way this
worked was to pass a data point when initialized the workqueue.
The new API assumed the work item is embedding in a structure
and we us container_of() to find that data pointer.
- Updated skc->skc_flags to be an unsigned long which is now
type checked in the bit operations. This silences the warnings.
- Updated autogen products and splat tests accordingly
- Added slab work queue task which gradually ages and free's slabs
from the cache which have not been used recently.
- Optimized slab packing algorithm to ensure each slab contains the
maximum number of objects without create to large a slab.
- Fix deadlock, we can never call kv_free() under the skc_lock. We
now unlink the objects and slabs from the cache itself and attach
them to a private work list. The contents of the list are then
subsequently freed outside the spin lock.
- Move magazine create/destroy operation on to local cpu.
- Further performace optimizations by minimize the usage of the large
per-cache skc_lock. This includes the addition of KMC_BIT_REAPING
bit mask which is used to prevent concurrent reaping, and to defer
new slab creation when reaping is occuring.
- Add KMC_BIT_DESTROYING bit mask which is set when the cache is being
destroyed, this is used to catch any task accessing the cache while
it is being destroyed.
- Add comments to all the functions and additional comments to try
and make everything as clear as possible.
- Major cleanup and additions to the SPLAT kmem tests to more
rigerously stress the cache implementation and look for any problems.
This includes correctness and performance tests.
- Updated portable work queue interfaces
udev support from sunddi implementation because it uses GPL-only
symbols. This support is optionally available for SPL consumers
if they define HAVE_GPL_ONLY_SYMBOLS and license their module as
GPL using the MODULE_LICENSE("GPL") macro.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@179 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
This fixes an oops when unloading the modules, in the case where memory
tracking was enabled and there were memory leaks. The comment in the
code explains what was the problem.
* spl-10-fix-assert-verify-ndebug.patch
This fixes ASSERT*() and VERIFY*() macros in non-debug builds. VERIFY*()
macros are supposed to check the condition and panic even in production
builds, and ASSERT*() macros don't need to evaluate the arguments.
Also some 32-bit fixes.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@165 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
That said when working with a finite resource like memory failure really
is always a possibility. It would be far better longer term if the ZFS
code could be weened off this assumption and properly handle the cases
where an allocation fails. Still I've applied the patch to spl-0.3.4
since this layer is supposed to emulate Solaris as closely as possible.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@164 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
already supprt atomic64_t types.
* spl-07-kmem-cleanup.patch
This moves all the debugging code from sys/kmem.h to spl-kmem.c, because
the huge macros were hard to debug and were bloating functions that
allocated memory. I also fixed some other minor problems, including
32-bit fixes and a reported memory leak which was just due to using the
wrong free function.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@163 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
spl-05-div64.patch
This is a much less intrusive fix for undefined 64-bit division symbols
when compiling the DMU in 32-bit kernels.
* spl-06-atomic64.patch
This is a workaround for 32-bit kernels that don't have atomic64_t.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@162 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
* spl-04-fix-taskq-spinlock-lockup.patch
Fixes a deadlock in the BIO completion handler, due to the taskq code
prematurely re-enabling interrupts when another spinlock had disabled
them in the IDE IRQ handler.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@161 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
from Ricardo which removes a dependency on the GPL-only symbol
set_cpus_allowed(). Using this symbol is simpler but in the name
of portability we are adopting a spinlock based solution here
to remove this dependency.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@160 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
were added to cover the 3 possible APIs from 2.6.9 to
2.6.26. We attempt to use the newest interfaces and if
not available fallback to the oldest. This a rework of
some changes proposed by Ricardo for RHEL4.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@150 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
at a time as I audit it. This chunk finishes moving the SPL entirely
off the linux slab on to the SPL implementation. It differs slightly
from the proposed version in that the spl continues to export to
all the Solaris types and functions. These do conflict with the
Linux slab so a module usings these interfaces must not include the
SPL slab if they also intend to use the linux slab. Or they must
explcitly #undef the macros which remap the functioin to their
spl_* equivilants.
A nice side of effect of dropping the entire linux slab is we
don't need to autoconf checks anymore. They kept messing with
the slab API endlessly!
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@148 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
based of the spl_kmem_obj_t tacked on the end of each object.
This actually isn't so back because we are now allocing large
chunks for the slab and partitioning it ourselves. So there's
not a ton of wasted space. We may suffer a performance hit
however due to alignment issues.
- Remove remaining depenancies on the linux slab implementation.
We're standing on our own now for better or worse.
- Rework slabs to be either kmem or vmem based. If neither
KMC_VMEM of KMC_KMEM are specified we make a decent guess
about what will work best for their based on the object
size. Additionally we provide a kmem_virt() function caller
can use to see if they have a virtual or physical address.
- Minor fixups in the test suite.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@141 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
cycle count which was costing me overhead. It was hurting
performance pretty badly for heavily used caches. I'm also
thinking the hash may be hurting me as well and it might
be worth sticking a pointer in to a little space after the
alloced object.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@140 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
based by vmalloc()'ed memory. I now alloc a slab which is
roughly 32*spl_obj_size and in this block of memory I place
the slab descriptor, slab object descriptors, and objects
themselves. This greatly reduces vmalloc lock contention.
Still some minor cleanup remains and fine tuning but
it's working pretty well.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@139 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
well for the expected workloads. Improvement in this commit include:
- Added DEBUG_KMEM_TRACKING #define which can optionally be set
when DEBUG_KMEM is defined to do per allocation tracking. This
allows us to get all the lightweight kmem debugging enabled by
default which is pretty light weight, and only when looking
for a memory leak we can briefly enable the per alloc tracking.
- Added set_normalized_timespec() in to SPL to simply using
the timespec() primatives from within a module.
- Added per-spinlock cycle counters to the slab in an attempt
to run down a lock contention issue. The contended lock
was in vmalloc() but I'm going to leave the cycle counters
in place for a little while until I'm convinced there arn't
other locking improvement possible in the slab.
- Added a proc interface to the slab to export per slab
cache statistics to /proc/spl/kmem/slab for analysis.
- Reworked spl_slab_alloc() function to allocate from kmem for
small allocation and vmem for large allocations. This improved
things considerably but futher work is needed.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@138 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
when repopulating it. Plus I fixed a few more suble races in
that part of the code which were catching me. Finally I fixed
a small race in kmem_test8.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@137 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
factor of 10x improvement on SMP system due to reduced lock contention.
This may put me in the ballpark of what is needed. We can still further
improve things on NUMA systems by creating an additional L3 cache per
memory node instead of the current global pool. With luck this won't
be needed. I should also take another look at the locking now that
everything is working. There's a good chance I can tighten it up a
little bit and improve things a little more.
kmem_lock: time (sec) slabs objs hash
kmem_lock: tot/max/calc tot/max/calc size/depth
kmem_lock: 0.000999926 6/6/1 192/192/32 32768/0
kmem_lock: 0.000999926 4/4/2 128/128/64 32768/0
kmem_lock: 0.000999926 4/4/4 128/128/128 32768/0
kmem_lock: 0.000999926 4/4/8 128/128/256 32768/0
kmem_lock: 0.000999926 4/4/16 128/128/512 32768/0
kmem_lock: 0.000999926 4/4/32 128/128/1024 32768/0
kmem_lock: 0.000999926 4/4/64 128/128/2048 32768/0
kmem_lock: 0.000999926 8/8/128 256/256/4096 32768/0
kmem_lock: 0.003999704 24/23/256 768/736/8192 32768/1
kmem_lock: 0.012999038 44/41/512 1408/1312/16384 32768/1
kmem_lock: 0.051996153 96/93/1024 3072/2976/32768 32768/2
kmem_lock: 0.181986536 187/184/2048 5984/5888/65536 32768/3
kmem_lock: 0.655951469 342/339/4096 10944/10848/131072 32768/4
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@136 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
to be overly clever and the context switch when the semaphore was busy
was destroying performance. Converting to a simple spin lock bough me
a factor of 50 or so. That said it's still not good enough. Tests
show bad performance and we are still CPU bound. The logical fix is
I need to implement per-cpu hot caches to minimize the SMP contention.
Linux and Solaris both have this, I was hoping to do without but it
looks like that's not to be.
kmem_lock: time (sec) slabs objs hash
kmem_lock: tot/max/calc tot/max/calc size/depth
kmem_lock: 0.022000000 7/6/64 224/177/2048 32768/1
kmem_lock: 0.039000000 13/13/128 416/404/4096 32768/1
kmem_lock: 0.079000000 23/21/256 736/672/8192 32768/1
kmem_lock: 0.158000000 48/47/512 1536/1504/16384 32768/1
kmem_lock: 0.345000000 105/105/1024 3360/3358/32768 32768/2
kmem_lock: 0.760000000 202/200/2048 6464/6400/65536 32768/3
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@135 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
longer be based on the linux slab but to be its own complete
implementation. The new slab behaves much more like the
Solaris slab than the Linux slab.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@132 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
1) Ensure mutex_init() never fails in the case of ENOMEM by retrying
forever. I don't think I've ever seen this happen but it was clear
after code inspection that if it did we would immediately crash.
2) Enable full debugging in check.sh for sanity tests. Might as well
get as much debug as we can in the case of a failure.
3) Reworked list of kmem caches tracked by SPL in to a hash with the
key based on the address of the kmem_cache_t. This should speed
up the constructor/destructor/shrinker lookup needed now for newer
kernel which removed the destructor support.
4) Updated kmem_cache_create to handle the case where CONFIG_SLUB
is defined. The slub would occasionally merge slab caches which
resulted in non-unique keys for our hash lookup in 3). To fix this
we detect if the slub is enabled and then set the needed flag
to prevent this merging from ever occuring.
5) New kernels removed the proc_dir_entry pointer from items
registered by sysctl. This means we can no long be sneaky and
manually insert things in to the sysctl tree simply by walking
the proc tree. So I'm forced to create a seperate tree for
all the things I can't easily support via sysctl interface.
I don't like it but it will do for now.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@124 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
working on this branch for the next few days I suggested you work
off of the 0.3.1 tag. The following changes are fairly extensive
and are designed to make the SPL compatible with all kernels in
the range of 2.6.18-2.6.25. There were 13 relevant API changes
between these releases and I have added the needed autoconf tests
to check for them. However, this has not all been tested extensively.
I'll sort of the breakage on Fedora Core 9 and RHEL5 this week.
SPL_AC_TYPE_UINTPTR_T
SPL_AC_TYPE_KMEM_CACHE_T
SPL_AC_KMEM_CACHE_DESTROY_INT
SPL_AC_ATOMIC_PANIC_NOTIFIER
SPL_AC_3ARGS_INIT_WORK
SPL_AC_2ARGS_REGISTER_SYSCTL
SPL_AC_KMEM_CACHE_T
SPL_AC_KMEM_CACHE_CREATE_DTOR
SPL_AC_3ARG_KMEM_CACHE_CREATE_CTOR
SPL_AC_SET_SHRINKER
SPL_AC_PATH_IN_NAMEIDATA
SPL_AC_TASK_CURR
SPL_AC_CTL_UNNUMBERED
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@119 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
compiled out when doing performance runs.
- Bite the bullet and fully autoconfize the debug options in the configure
time parameters. By default all the debug support is disable in the core
SPL build, but available to modules which enable it when building against
the SPL. To enable particular SPL debug support use the follow configure
options:
--enable-debug Internal ASSERTs
--enable-debug-kmem Detailed memory accounting
--enable-debug-mutex Detailed mutex tracking
--enable-debug_kstat Kstat info exported to /proc
--enable-debug-callb Additional callb debug
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@111 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
other primitive implementations. Additionally ensure that GFP_ATOMIC
is use for allocations when in interrupt context.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@108 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
may not fail. To get this behavior I'd added a retry to the shim layer
even though it is abusive to the VM, at least it should prevent the crash.
Additionally I added a proc counter so I can easily check how often this
is happening. It should be fairly rare, but likely will get worse and
worse the longer the machine has been up.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@104 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
not to support a few flags (we assert if they are used), and I
did not add the libkstat interface and instead exported everything
to proc for easy access.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@103 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
- Ensure the mutex_stats_sem and mutex_stats_list are initialized
- Only spin if you have to in mutex_init
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@97 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
- Detailed kmem memory allocation tracking. We can now get on
spl module unload a list of all memory allocations which were
not free'd and where the original alloc was. E.g.
SPL: 15554:632:(spl-kmem.c:442:kmem_fini()) kmem leaked 90/319332 bytes
SPL: 15554:648:(spl-kmem.c:451:kmem_fini()) address size data func:line
SPL: 15554:648:(spl-kmem.c:457:kmem_fini()) ffff8100734b68b8 32 0100000001005a5a __spl_mutex_init:70
SPL: 15554:648:(spl-kmem.c:457:kmem_fini()) ffff8100734b6148 13 &tl->tl_lock __spl_mutex_init:74
SPL: 15554:648:(spl-kmem.c:457:kmem_fini()) ffff81007ac43730 32 0100000001005a5a __spl_mutex_init:70
SPL: 15554:648:(spl-kmem.c:457:kmem_fini()) ffff81007ac437d8 13 &tl->tl_lock __spl_mutex_init:74
- Shift to using rwsems in kmem implmentation, to simply locking and
improve concurency.
- Shift to using rwsems in mutex implementation, additionally ensure we
never sleep in the init function if non-zero preempt_count or
interrupts are disabled as can happen in a slab cache ctor/dtor.
- Other minor formating fixes and such.
TODO:
- Finish the vmem memory allocation tracking
- Vet all other SPL primatives for potential sleeping during *_init. I
suspect the rwlock implemenation does this and should be fixes just
like the mutex implemenation.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@95 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
crashes but it's not clear to me yet if these are a problem with
the mutex implementation or ZFSs usage of it.
Minor taskq fixes to add new tasks to the end of the pending list.
Minor enhansements to the debug infrastructure.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@94 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
configurable number of threads like the Solaris version and almost
all of the options are supported. Unfortunately, it appears to have
made absolutely no difference to our performance numbers. I need
to keep looking for where we are bottle necking.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@93 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
do not have __GFP_ZERO set. Once the memory is allocated
then zero out the memory if __GFP_ZERO is passed to
__vmem_alloc.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@88 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
ensure I never add anything to the stack I don't absolutely need.
All this debug code could be removed from a production build
anyway so I'm not so worried about the performance impact. We
may also consider revisting the mutex and condvar implementation
to ensure no additional stack is used there.
Initial indications are I have reduced the worst case stack
usage to 9080 bytes. Still to large for the default 8k stacks
so I have been forced to run with 16k stacks until I can
reduce the worst offenders.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@83 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
process of destroying the stacks. Threshhold set fairly aggressively
top 80% of stack usage.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@82 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
- Replacing all BUG_ON()'s with proper ASSERT()'s
- Using ENTRY,EXIT,GOTO, and RETURN macro to instument call paths
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@78 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
changes bring over everything lustre had for debugging with
two exceptions. I dropped by the debug daemon and upcalls
just because it made things a little easier. They can be
readded easily enough if we feel they are needed.
Everything compiles and seems to work on first inspection
but I suspect there are a handful of issues still lingering
which I'll be sorting out right away. I just wanted to get
all these changes commited and safe. I'm getting a little
paranoid about losing them.
git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@75 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c
