Hello,
* Andrew, I think this one is best routed through -mm together with
the previous series. Please read on.
* Bruce, I couldn't convert nfsd. Can you please help? More on it
later.
* Stanislav, Eric, James, can you please take a look at ipc/util.c
conversion? I moved allocation inside ipc_addid(). I *think* it's
correct but I'm not too familiar with the code.
* Jens, 0006 is fix for block layer regarding devt allocation. Given
how long this has been broken unnoticed and how late we're in the
release cycle, I think it would be better to route the fix with the
rest and let it get backported through -stable later on.
Currently, idr interface is a pain in the ass to use. It has a
mandatory pre-alloc mechanism which doesn't even guarantee the
following allocation wouldn't fail from memory shortage requiring its
users to loop on -EAGAIN. Also, there's no way to specify upper
limit. Combined, a user who wants allocate an ID with upper limit
ends up doing something like the following.
do {
if (!idr_pre_get(idr, GFP_KERNEL))
return -ENOMEM;
spin_lock(lock);
ret = idr_get_new_above(idr, ptr, lower_limit, &id);
if (!ret && id < upper_limit) {
idr_remove(idr, id);
ret = -ENOSPC;
}
spin_unlock(lock);
} while (ret == -EAGAIN);
if (ret)
return ret;
which is just crazy. Preloading is also very inefficient - each idr
ends up holding MAX_IDR_FREE idr_layers. Even on a fairly small
machine (my x230) w/o anything too heavy going on, it ends up
allocating over 1200 idr_layers with three quarters of them unused.
This also makes it difficult to make each layer larger to reduce the
depth of idr trees.
This patchset implements a new set of allocation interface for idr.
void idr_preload(gfp_t gfp_mask);
void idr_preload_end(void);
int idr_alloc(struct idr *idr, void *ptr, int start, int end, gfp_t gfp_mask);
idr_alloc() can be used w/o preloading if permissive enough @gfp_mask
can be used directly. If not, preloading, which uses per-cpu layer
buffer, can be used. Note that idr_preload() doesn't fail. It simply
guarnatees that the following idr_alloc() would behave as if it were
called with @gfp_mask specified at preloading time. The above example
with the old interface would become the following using the new
interface.
idr_preload(GFP_KERNEL);
spin_lock(lock);
id = idr_alloc(idr, ptr, lower_limit, upper_limit, GFP_NOWAIT);
spin_unlock(lock);
idr_preload_end();
if (id < 0)
return id;
Which is simpler and less error-prone. Also, there are many cases
where preloading isn't necessary (e.g. the section is protected by
outer mutex). For those, idr_alloc() can be used by itself.
id = idr_alloc(idr, ptr, lower_limit, upper_limit, GFP_KERNEL);
if (id < 0)
return id;
I converted all in-kernel users except nfsd and staging drivers. nfsd
splits preloading and actual id allocation in a way that per-cpu
preloading can't be used. I couldn't follow the control flow to
verify whether the current code is correct either. I think the best
way would be allocating ID upfront without installing the handle and
then later using idr_replace() to install the pointer when the ID
actually gets used. Bruce, would something like that be possible?
This patchset contains the following 62 patches.
0001-idr-cosmetic-updates-to-struct-initializer-definitio.patch
0002-idr-relocate-idr_for_each_entry-and-reorganize-id-r-.patch
0003-idr-remove-_idr_rc_to_errno-hack.patch
0004-idr-refactor-idr_get_new_above.patch
0005-idr-implement-idr_preload-_end-and-idr_alloc.patch
0006-block-fix-synchronization-and-limit-check-in-blk_all.patch
0007-block-convert-to-idr_alloc.patch
0008-block-loop-convert-to-idr_alloc.patch
0009-atm-nicstar-convert-to-idr_alloc.patch
0010-drbd-convert-to-idr_alloc.patch
0011-dca-convert-to-idr_alloc.patch
0012-dmaengine-convert-to-idr_alloc.patch
0013-firewire-convert-to-idr_alloc.patch
0014-gpio-convert-to-idr_alloc.patch
0015-drm-convert-to-idr_alloc.patch
0016-drm-exynos-convert-to-idr_alloc.patch
0017-drm-i915-convert-to-idr_alloc.patch
0018-drm-sis-convert-to-idr_alloc.patch
0019-drm-via-convert-to-idr_alloc.patch
0020-drm-vmwgfx-convert-to-idr_alloc.patch
0021-i2c-convert-to-idr_alloc.patch
0022-infiniband-core-convert-to-idr_alloc.patch
0023-infiniband-amso1100-convert-to-idr_alloc.patch
0024-infiniband-cxgb3-convert-to-idr_alloc.patch
0025-infiniband-cxgb4-convert-to-idr_alloc.patch
0026-infiniband-ehca-convert-to-idr_alloc.patch
0027-infiniband-ipath-convert-to-idr_alloc.patch
0028-infiniband-mlx4-convert-to-idr_alloc.patch
0029-infiniband-ocrdma-convert-to-idr_alloc.patch
0030-infiniband-qib-convert-to-idr_alloc.patch
0031-dm-convert-to-idr_alloc.patch
0032-memstick-convert-to-idr_alloc.patch
0033-mfd-convert-to-idr_alloc.patch
0034-misc-c2port-convert-to-idr_alloc.patch
0035-misc-tifm_core-convert-to-idr_alloc.patch
0036-mmc-convert-to-idr_alloc.patch
0037-mtd-convert-to-idr_alloc.patch
0038-i2c-convert-to-idr_alloc.patch
0039-ppp-convert-to-idr_alloc.patch
0040-power-convert-to-idr_alloc.patch
0041-pps-convert-to-idr_alloc.patch
0042-remoteproc-convert-to-idr_alloc.patch
0043-rpmsg-convert-to-idr_alloc.patch
0044-scsi-bfa-convert-to-idr_alloc.patch
0045-scsi-convert-to-idr_alloc.patch
0046-target-iscsi-convert-to-idr_alloc.patch
0047-scsi-lpfc-convert-to-idr_alloc.patch
0048-thermal-convert-to-idr_alloc.patch
0049-uio-convert-to-idr_alloc.patch
0050-vfio-convert-to-idr_alloc.patch
0051-dlm-convert-to-idr_alloc.patch
0052-inotify-convert-to-idr_alloc.patch
0053-ocfs2-convert-to-idr_alloc.patch
0054-ipc-convert-to-idr_alloc.patch
0055-cgroup-convert-to-idr_alloc.patch
0056-events-convert-to-idr_alloc.patch
0057-posix-timers-convert-to-idr_alloc.patch
0058-net-9p-convert-to-idr_alloc.patch
0059-mac80211-convert-to-idr_alloc.patch
0060-sctp-convert-to-idr_alloc.patch
0061-nfs4client-convert-to-idr_alloc.patch
0062-idr-deprecate-idr_pre_get-and-idr_get_new-_above.patch
0001-0005 implement the new idr interface. 0006 is idr related block
fix. 0007-0061 convert in-kernel users to idr_alloc(). 0062 marks
the old interface deprecated (note that this makes nfsd compilation
generates warning).
While I tried to be careful and reviewed the whole series a couple
times, my test coverage is unavoidably quite limited. I don't think
fallouts would be too difficult to spot during testing.
This patchset is on top of
[1] [PATCH] idr: fix a subtle bug in idr_get_next()
+ [2] [PATCHSET] idr: deprecate idr_remove_all()
and available in the following git branch.
git://git.kernel.org/pub/scm/linux/kernel/git/tj/misc.git convert-to-idr_alloc
Excluding idr.[hc], which should become smaller once we remove the old
interface, this conversion sheds 482 lines across the kernel.
diffstat follows.
block/bsg.c | 26 --
block/genhd.c | 22 --
drivers/atm/nicstar.c | 27 +-
drivers/block/drbd/drbd_main.c | 29 +-
drivers/block/loop.c | 23 --
drivers/dca/dca-sysfs.c | 23 --
drivers/dma/dmaengine.c | 16 -
drivers/firewire/core-cdev.c | 16 -
drivers/firewire/core-device.c | 5
drivers/gpio/gpiolib.c | 11 -
drivers/gpu/drm/drm_context.c | 17 -
drivers/gpu/drm/drm_crtc.c | 15 -
drivers/gpu/drm/drm_gem.c | 35 +--
drivers/gpu/drm/drm_stub.c | 19 -
drivers/gpu/drm/exynos/exynos_drm_ipp.c | 16 -
drivers/gpu/drm/i915/i915_gem_context.c | 21 --
drivers/gpu/drm/sis/sis_mm.c | 13 -
drivers/gpu/drm/via/via_mm.c | 13 -
drivers/gpu/drm/vmwgfx/vmwgfx_resource.c | 17 -
drivers/i2c/i2c-core.c | 45 ----
drivers/infiniband/core/cm.c | 22 +-
drivers/infiniband/core/cma.c | 24 --
drivers/infiniband/core/sa_query.c | 15 -
drivers/infiniband/core/ucm.c | 16 -
drivers/infiniband/core/ucma.c | 32 ---
drivers/infiniband/core/uverbs_cmd.c | 17 -
drivers/infiniband/hw/amso1100/c2_qp.c | 19 +
drivers/infiniband/hw/cxgb3/iwch.h | 24 +-
drivers/infiniband/hw/cxgb4/iw_cxgb4.h | 27 +-
drivers/infiniband/hw/ehca/ehca_cq.c | 27 --
drivers/infiniband/hw/ehca/ehca_qp.c | 34 +--
drivers/infiniband/hw/ipath/ipath_driver.c | 16 -
drivers/infiniband/hw/mlx4/cm.c | 32 +--
drivers/infiniband/hw/ocrdma/ocrdma_main.c | 14 -
drivers/infiniband/hw/qib/qib_init.c | 21 --
drivers/md/dm.c | 54 +----
drivers/memstick/core/memstick.c | 21 --
drivers/memstick/core/mspro_block.c | 17 -
drivers/mfd/rtsx_pcr.c | 13 -
drivers/misc/c2port/core.c | 22 --
drivers/misc/tifm_core.c | 11 -
drivers/mmc/core/host.c | 11 -
drivers/mtd/mtdcore.c | 9
drivers/net/macvtap.c | 21 --
drivers/net/ppp/ppp_generic.c | 33 ---
drivers/power/bq2415x_charger.c | 11 -
drivers/power/bq27x00_battery.c | 9
drivers/power/ds2782_battery.c | 9
drivers/pps/kapi.c | 2
drivers/pps/pps.c | 36 +--
drivers/remoteproc/remoteproc_core.c | 10
drivers/rpmsg/virtio_rpmsg_bus.c | 30 +-
drivers/scsi/bfa/bfad_im.c | 15 -
drivers/scsi/ch.c | 21 --
drivers/scsi/lpfc/lpfc_init.c | 12 -
drivers/scsi/sg.c | 43 +---
drivers/scsi/st.c | 27 --
drivers/target/iscsi/iscsi_target.c | 15 -
drivers/target/iscsi/iscsi_target_login.c | 15 -
drivers/thermal/cpu_cooling.c | 17 -
drivers/thermal/thermal_sys.c | 17 -
drivers/uio/uio.c | 19 -
drivers/vfio/vfio.c | 18 -
fs/dlm/lock.c | 18 -
fs/dlm/recover.c | 27 +-
fs/nfs/nfs4client.c | 13 -
fs/notify/inotify/inotify_user.c | 24 +-
fs/ocfs2/cluster/tcp.c | 32 +--
include/linux/idr.h | 135 +++++++++----
ipc/util.c | 30 --
kernel/cgroup.c | 27 --
kernel/events/core.c | 10
kernel/posix-timers.c | 18 -
lib/idr.c | 291 ++++++++++++++++++-----------
net/9p/util.c | 17 -
net/mac80211/main.c | 2
net/mac80211/tx.c | 18 -
net/sctp/associola.c | 27 +-
78 files changed, 816 insertions(+), 1160 deletions(-)
--
tejun
[1] https://lkml.org/lkml/2013/2/2/145
[2] https://lkml.org/lkml/2013/1/25/759
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
