Use large stacks when available

While stack size will vary by architecture it has historically defaulted to
8K on x86_64 systems.  However, as of Linux 3.15 the default thread stack
size was increased to 16K.  These kernels are now the default in most non-
enterprise distributions which means we no longer need to assume 8K stacks.

This patch takes advantage of that fact by appropriately reverting stack
conservation changes which were made to ensure stability.  Changes which
may have had a negative impact on performance for certain workloads.  This
also has the side effect of bringing the code slightly more in line with
upstream.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Closes #4059
This commit is contained in:
Brian Behlendorf 2015-12-02 11:53:37 -08:00
parent f40926795c
commit b58986eebf
4 changed files with 67 additions and 19 deletions

View File

@ -7,7 +7,8 @@ AM_CFLAGS += ${NO_BOOL_COMPARE}
AM_CFLAGS += -fno-strict-aliasing AM_CFLAGS += -fno-strict-aliasing
AM_CPPFLAGS = -D_GNU_SOURCE -D__EXTENSIONS__ -D_REENTRANT AM_CPPFLAGS = -D_GNU_SOURCE -D__EXTENSIONS__ -D_REENTRANT
AM_CPPFLAGS += -D_POSIX_PTHREAD_SEMANTICS -D_FILE_OFFSET_BITS=64 AM_CPPFLAGS += -D_POSIX_PTHREAD_SEMANTICS -D_FILE_OFFSET_BITS=64
AM_CPPFLAGS += -D_LARGEFILE64_SOURCE -DTEXT_DOMAIN=\"zfs-linux-user\" AM_CPPFLAGS += -D_LARGEFILE64_SOURCE -DHAVE_LARGE_STACKS=1
AM_CPPFLAGS += -DTEXT_DOMAIN=\"zfs-linux-user\"
AM_CPPFLAGS += -DLIBEXECDIR=\"$(libexecdir)\" AM_CPPFLAGS += -DLIBEXECDIR=\"$(libexecdir)\"
AM_CPPFLAGS += -DRUNSTATEDIR=\"$(runstatedir)\" AM_CPPFLAGS += -DRUNSTATEDIR=\"$(runstatedir)\"
AM_CPPFLAGS += -DSBINDIR=\"$(sbindir)\" AM_CPPFLAGS += -DSBINDIR=\"$(sbindir)\"

View File

@ -460,9 +460,35 @@ AC_DEFUN([ZFS_AC_KERNEL_CONFIG], [
], [ ], [
]) ])
ZFS_AC_KERNEL_CONFIG_THREAD_SIZE
ZFS_AC_KERNEL_CONFIG_DEBUG_LOCK_ALLOC ZFS_AC_KERNEL_CONFIG_DEBUG_LOCK_ALLOC
]) ])
dnl #
dnl # Check configured THREAD_SIZE
dnl #
dnl # The stack size will vary by architecture, but as of Linux 3.15 on x86_64
dnl # the default thread stack size was increased to 16K from 8K. Therefore,
dnl # on newer kernels and some architectures stack usage optimizations can be
dnl # conditionally applied to improve performance without negatively impacting
dnl # stability.
dnl #
AC_DEFUN([ZFS_AC_KERNEL_CONFIG_THREAD_SIZE], [
AC_MSG_CHECKING([whether kernel was built with 16K or larger stacks])
ZFS_LINUX_TRY_COMPILE([
#include <linux/module.h>
],[
#if (THREAD_SIZE < 16384)
#error "THREAD_SIZE is less than 16K"
#endif
],[
AC_MSG_RESULT([yes])
AC_DEFINE(HAVE_LARGE_STACKS, 1, [kernel has large stacks])
],[
AC_MSG_RESULT([no])
])
])
dnl # dnl #
dnl # Check CONFIG_DEBUG_LOCK_ALLOC dnl # Check CONFIG_DEBUG_LOCK_ALLOC
dnl # dnl #
@ -572,7 +598,7 @@ dnl #
dnl # ZFS_LINUX_CONFIG dnl # ZFS_LINUX_CONFIG
dnl # dnl #
AC_DEFUN([ZFS_LINUX_CONFIG], AC_DEFUN([ZFS_LINUX_CONFIG],
[AC_MSG_CHECKING([whether Linux was built with CONFIG_$1]) [AC_MSG_CHECKING([whether kernel was built with CONFIG_$1])
ZFS_LINUX_TRY_COMPILE([ ZFS_LINUX_TRY_COMPILE([
#include <linux/module.h> #include <linux/module.h>
],[ ],[

View File

@ -67,7 +67,7 @@ typedef struct dump_bytes_io {
} dump_bytes_io_t; } dump_bytes_io_t;
static void static void
dump_bytes_strategy(void *arg) dump_bytes_cb(void *arg)
{ {
dump_bytes_io_t *dbi = (dump_bytes_io_t *)arg; dump_bytes_io_t *dbi = (dump_bytes_io_t *)arg;
dmu_sendarg_t *dsp = dbi->dbi_dsp; dmu_sendarg_t *dsp = dbi->dbi_dsp;
@ -94,6 +94,9 @@ dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
dbi.dbi_buf = buf; dbi.dbi_buf = buf;
dbi.dbi_len = len; dbi.dbi_len = len;
#if defined(HAVE_LARGE_STACKS)
dump_bytes_cb(&dbi);
#else
/* /*
* The vn_rdwr() call is performed in a taskq to ensure that there is * The vn_rdwr() call is performed in a taskq to ensure that there is
* always enough stack space to write safely to the target filesystem. * always enough stack space to write safely to the target filesystem.
@ -101,7 +104,8 @@ dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
* them and they are used in vdev_file.c for a similar purpose. * them and they are used in vdev_file.c for a similar purpose.
*/ */
spa_taskq_dispatch_sync(dmu_objset_spa(dsp->dsa_os), ZIO_TYPE_FREE, spa_taskq_dispatch_sync(dmu_objset_spa(dsp->dsa_os), ZIO_TYPE_FREE,
ZIO_TASKQ_ISSUE, dump_bytes_strategy, &dbi, TQ_SLEEP); ZIO_TASKQ_ISSUE, dump_bytes_cb, &dbi, TQ_SLEEP);
#endif /* HAVE_LARGE_STACKS */
return (dsp->dsa_err); return (dsp->dsa_err);
} }

View File

@ -1401,6 +1401,31 @@ zio_execute(zio_t *zio)
spl_fstrans_unmark(cookie); spl_fstrans_unmark(cookie);
} }
/*
* Used to determine if in the current context the stack is sized large
* enough to allow zio_execute() to be called recursively. A minimum
* stack size of 16K is required to avoid needing to re-dispatch the zio.
*/
boolean_t
zio_execute_stack_check(zio_t *zio)
{
#if !defined(HAVE_LARGE_STACKS)
dsl_pool_t *dp = spa_get_dsl(zio->io_spa);
/* Executing in txg_sync_thread() context. */
if (dp && curthread == dp->dp_tx.tx_sync_thread)
return (B_TRUE);
/* Pool initialization outside of zio_taskq context. */
if (dp && spa_is_initializing(dp->dp_spa) &&
!zio_taskq_member(zio, ZIO_TASKQ_ISSUE) &&
!zio_taskq_member(zio, ZIO_TASKQ_ISSUE_HIGH))
return (B_TRUE);
#endif /* HAVE_LARGE_STACKS */
return (B_FALSE);
}
__attribute__((always_inline)) __attribute__((always_inline))
static inline void static inline void
__zio_execute(zio_t *zio) __zio_execute(zio_t *zio)
@ -1410,8 +1435,6 @@ __zio_execute(zio_t *zio)
while (zio->io_stage < ZIO_STAGE_DONE) { while (zio->io_stage < ZIO_STAGE_DONE) {
enum zio_stage pipeline = zio->io_pipeline; enum zio_stage pipeline = zio->io_pipeline;
enum zio_stage stage = zio->io_stage; enum zio_stage stage = zio->io_stage;
dsl_pool_t *dp;
boolean_t cut;
int rv; int rv;
ASSERT(!MUTEX_HELD(&zio->io_lock)); ASSERT(!MUTEX_HELD(&zio->io_lock));
@ -1424,10 +1447,6 @@ __zio_execute(zio_t *zio)
ASSERT(stage <= ZIO_STAGE_DONE); ASSERT(stage <= ZIO_STAGE_DONE);
dp = spa_get_dsl(zio->io_spa);
cut = (stage == ZIO_STAGE_VDEV_IO_START) ?
zio_requeue_io_start_cut_in_line : B_FALSE;
/* /*
* If we are in interrupt context and this pipeline stage * If we are in interrupt context and this pipeline stage
* will grab a config lock that is held across I/O, * will grab a config lock that is held across I/O,
@ -1439,21 +1458,19 @@ __zio_execute(zio_t *zio)
*/ */
if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL &&
zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) { zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) {
boolean_t cut = (stage == ZIO_STAGE_VDEV_IO_START) ?
zio_requeue_io_start_cut_in_line : B_FALSE;
zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut);
return; return;
} }
/* /*
* If we executing in the context of the tx_sync_thread, * If the current context doesn't have large enough stacks
* or we are performing pool initialization outside of a * the zio must be issued asynchronously to prevent overflow.
* zio_taskq[ZIO_TASKQ_ISSUE|ZIO_TASKQ_ISSUE_HIGH] context.
* Then issue the zio asynchronously to minimize stack usage
* for these deep call paths.
*/ */
if ((dp && curthread == dp->dp_tx.tx_sync_thread) || if (zio_execute_stack_check(zio)) {
(dp && spa_is_initializing(dp->dp_spa) && boolean_t cut = (stage == ZIO_STAGE_VDEV_IO_START) ?
!zio_taskq_member(zio, ZIO_TASKQ_ISSUE) && zio_requeue_io_start_cut_in_line : B_FALSE;
!zio_taskq_member(zio, ZIO_TASKQ_ISSUE_HIGH))) {
zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut);
return; return;
} }