Following the recent introduction of for_each_andnot(), add some tests to
ensure for_each_cpu_and(not) results in the same as iterating over the
result of cpumask_and(not)().
Suggested-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Valentin Schneider <vschneid@redhat.com>
In preparation of introducing for_each_cpu_andnot(), add a variant of
find_next_bit() that negate the bits in @addr2 when ANDing them with the
bits in @addr1.
Signed-off-by: Valentin Schneider <vschneid@redhat.com>
The helper is better optimized for the worst case: in case of empty
cpumask, current code traverses 2 * size:
next = cpumask_next_and(prev, src1p, src2p);
if (next >= nr_cpu_ids)
next = cpumask_first_and(src1p, src2p);
At bitmap level we can stop earlier after checking 'size + offset' bits.
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Add cpumask_nth_{,and,andnot} as wrappers around corresponding
find functions, and use it in cpumask_local_spread().
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Kernel lacks for a function that searches for Nth bit in a bitmap.
Usually people do it like this:
for_each_set_bit(bit, mask, size)
if (n-- == 0)
return bit;
We can do it more efficiently, if we:
1. find a word containing Nth bit, using hweight(); and
2. find the bit, using a helper fns(), that works similarly to
__ffs() and ffz().
fns() is implemented as a simple loop. For x86_64, there's PDEP instruction
to do that: ret = clz(pdep(1 << idx, num)). However, for large bitmaps the
most of improvement comes from using hweight(), so I kept fns() simple.
New find_nth_bit() is ~70 times faster on x86_64/kvm in find_bit benchmark:
find_nth_bit: 7154190 ns, 16411 iterations
for_each_bit: 505493126 ns, 16315 iterations
With all that, a family of 3 new functions is added, and used where
appropriate in the following patches.
Signed-off-by: Yury Norov <yury.norov@gmail.com>
The function calculates Hamming weight of (bitmap1 & bitmap2). Now we
have to do like this:
tmp = bitmap_alloc(nbits);
bitmap_and(tmp, map1, map2, nbits);
weight = bitmap_weight(tmp, nbits);
bitmap_free(tmp);
This requires additional memory, adds pressure on alloc subsystem, and
way less cache-friendly than just:
weight = bitmap_weight_and(map1, map2, nbits);
The following patches apply it for cpumask functions.
Signed-off-by: Yury Norov <yury.norov@gmail.com>
__bitmap_weight() is not to be used directly in the kernel code because
it's a helper for bitmap_weight(). Switch everything to bitmap_weight().
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Over the past couple years, the function _find_next_bit() was extended
with parameters that modify its behavior to implement and- zero- and le-
flavors. The parameters are passed at compile time, but current design
prevents a compiler from optimizing out the conditionals.
As find_next_bit() API grows, I expect that more parameters will be added.
Current design would require more conditional code in _find_next_bit(),
which would bloat the helper even more and make it barely readable.
This patch replaces _find_next_bit() with a macro FIND_NEXT_BIT, and adds
a set of wrappers, so that the compile-time optimizations become possible.
The common logic is moved to the new macro, and all flavors may be
generated by providing a FETCH macro parameter, like in this example:
#define FIND_NEXT_BIT(FETCH, MUNGE, size, start) ...
find_next_xornot_and_bit(addr1, addr2, addr3, size, start)
{
return FIND_NEXT_BIT(addr1[idx] ^ ~addr2[idx] & addr3[idx],
/* nop */, size, start);
}
The FETCH may be of any complexity, as soon as it only refers the bitmap(s)
and an iterator idx.
MUNGE is here to support _le code generation for BE builds. May be
empty.
I ran find_bit_benchmark 16 times on top of 6.0-rc2 and 16 times on top
of 6.0-rc2 + this series. The results for kvm/x86_64 are:
v6.0-rc2 Optimized Difference Z-score
Random dense bitmap ns ns ns %
find_next_bit: 787735 670546 117189 14.9 3.97
find_next_zero_bit: 777492 664208 113284 14.6 10.51
find_last_bit: 830925 687573 143352 17.3 2.35
find_first_bit: 3874366 3306635 567731 14.7 1.84
find_first_and_bit: 40677125 37739887 2937238 7.2 1.36
find_next_and_bit: 347865 304456 43409 12.5 1.35
Random sparse bitmap
find_next_bit: 19816 14021 5795 29.2 6.10
find_next_zero_bit: 1318901 1223794 95107 7.2 1.41
find_last_bit: 14573 13514 1059 7.3 6.92
find_first_bit: 1313321 1249024 64297 4.9 1.53
find_first_and_bit: 8921 8098 823 9.2 4.56
find_next_and_bit: 9796 7176 2620 26.7 5.39
Where the statistics is significant (z-score > 3), the improvement
is ~15%.
According to the bloat-o-meter, the Image size is 10-11K less:
x86_64/defconfig:
add/remove: 32/14 grow/shrink: 61/782 up/down: 6344/-16521 (-10177)
arm64/defconfig:
add/remove: 3/2 grow/shrink: 50/714 up/down: 608/-11556 (-10948)
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
find_first_zero_bit_le() is an alias to find_next_zero_bit_le(),
despite that 'next' is known to be slower than 'first' version.
Now that we have common FIND_FIRST_BIT() macro helper, it's trivial
to implement find_first_zero_bit_le() as a real function.
Reviewed-by: Valentin Schneider <vschneid@redhat.com>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Now that we have many flavors of find_first_bit(), and expect even more,
it's better to have one macro that generates optimal code for all and makes
maintaining of slightly different functions simpler.
The logic common to all versions is moved to the new macro, and all the
flavors are generated by providing an FETCH macro-parameter, like
in this example:
#define FIND_FIRST_BIT(FETCH, MUNGE, size) ...
find_first_ornot_and_bit(addr1, addr2, addr3, size)
{
return FIND_FIRST_BIT(addr1[idx] | ~addr2[idx] & addr3[idx], /* nop */, size);
}
The FETCH may be of any complexity, as soon as it only refers
the bitmap(s) and an iterator idx.
MUNGE is here to support _le code generation for BE builds. May be
empty.
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Reviewed-by: Valentin Schneider <vschneid@redhat.com>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
The size of cpumasks is hard-limited by compile-time parameter NR_CPUS,
but defined at boot-time when kernel parses ACPI/DT tables, and stored in
nr_cpu_ids. In many practical cases, number of CPUs for a target is known
at compile time, and can be provided with NR_CPUS.
In that case, compiler may be instructed to rely on NR_CPUS as on actual
number of CPUs, not an upper limit. It allows to optimize many cpumask
routines and significantly shrink size of the kernel image.
This patch adds FORCE_NR_CPUS option to teach the compiler to rely on
NR_CPUS and enable corresponding optimizations.
If FORCE_NR_CPUS=y, kernel will not set nr_cpu_ids at boot, but only check
that the actual number of possible CPUs is equal to NR_CPUS, and WARN if
that doesn't hold.
The new option is especially useful in embedded applications because
kernel configurations are unique for each SoC, the number of CPUs is
constant and known well, and memory limitations are typically harder.
For my 4-CPU ARM64 build with NR_CPUS=4, FORCE_NR_CPUS=y saves 46KB:
add/remove: 3/4 grow/shrink: 46/729 up/down: 652/-46952 (-46300)
Signed-off-by: Yury Norov <yury.norov@gmail.com>
dependency on XOR_BLOCKS.
-----BEGIN PGP SIGNATURE-----
iQIzBAABCgAdFiEEn51F/lCuNhUwmDeSxycdCkmxi6cFAmMIp9QACgkQxycdCkmx
i6eYhRAAs4XrSzpSSWaTaf3JZuQIqqs9KoABLvxeYfemEwgnd28NLQrea2qnM5yW
NT5TSL/OYpTmTTD9Q/uHm9TBpRchuQMto7rDLwB129ie1WdeSqwOKjk9zMb7SzkL
5mgO2L05jkyvo4NI4KxUyeBtMNdZc8eu4/iAhOejdnZYIgxFQUAuNVmFcy1VG/tq
kFOOLaXLOtrs9y+IswyoLoP7LdLbYCtUO8B4wBzMKZmRHXRKYtpXAnz/ytUhfSEM
JL6Vrb4jwAUK29B8A6Nk49gl5CHSlXOhnTrv6RK2qhwpLXQxsWWrDRinsJ6xQJk0
3pWVRDm2MAhpqsuk6Q/Cj0pFpFSiLCPRPSiH0Du2w6W3Fm8KGcQoE87bYIgBSTX/
vpBdGU0ItVokMB0OqNIHOLUFyi4wMM20wOKt5znVLeVlY0adye6SdcBGm/qXurR4
midv7HNyBZkjh/H9wtC//hdkpQDq8l87ygL/F44IBFlXItUFt+Vrhoi61m8AGSVb
5nkGwSQMe/VXBKV3gpBEM7LzVTwUsw0poG73rxFBByrAzVVd9Y9F4FXJ5xUmcRvf
uYl95Zw6vk5IBu/ijdqaJPyHBXLqVWg7FyAlZjtJxCIO4gIBMl/KCNY8yHE6f8Nd
6kpvL2GmFvYmpnu04ay+tFAEB0KgbzDZSNrZissZN0ilMwvykzM=
=dIEI
-----END PGP SIGNATURE-----
Merge tag 'v6.0-p2' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Pull crypto fix from Herbert Xu:
"Fix a boot performance regression due to an unnecessary dependency on
XOR_BLOCKS"
* tag 'v6.0-p2' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
crypto: lib - remove unneeded selection of XOR_BLOCKS
Hi Linus,
Please pull (hopefully) the last portion of fixes from Sander for his
UP rework series. The original series came from -mm tree, and it was
not the latest version, that's why we need follow-ups. It fixes only
a test introduced by that series. The test fails under certain configs.
From Sander:
This series fixes the reported issues, and implements the suggested
improvements, for the version of the cpumask tests [1] that was merged
with commit c41e8866c2 ("lib/test: introduce cpumask KUnit test
suite").
These changes include fixes for the tests, and better alignment with the
KUnit style guidelines.
Thanks,
Yury
-----BEGIN PGP SIGNATURE-----
iQGzBAABCgAdFiEEi8GdvG6xMhdgpu/4sUSA/TofvsgFAmMKYcoACgkQsUSA/Tof
vsiceAv/TY+HTn1gmrNQwi7xC6VUD7mFYlVNZMtyMpZ23UYildz5SjFfuQV3UbXI
H5yKgSao9VFsbwyDUXbhySgOaNR8auq17Ey3jSJuR2A76qO2u2d79Gdt4IjIkq5N
IGOPv/pNOur7J+KSbiVhXasFeZGJ6Xi+xAobp5CK1uPCUI3oU1pAcm1iKkI+eWZ3
tPsM3aWcYGCDec7tqtqcsiWO2x9imPnrpI+C91Pwwr+N40ObkMc4IPzuPrQRn2T2
ECY9pgIWKOwOJ41jzgCVwZIHmuOn9dEgmaEGvE9Ah57OwuDlS43M4Ok3xy2+xS3t
3naLG3p02sJy7sXabC+xH4VJVPNT9/qauMW27cntPeeI2i/+yZXuQSLlVOllrY7/
LYxI8lVb1j50A90I/WrwXoDV0E68cfjhkiqhkgV33t1EamhSJvTG8GwCnF46WG8o
LzLukvoohA9uIrPAH2YpkZtrvsuT6iQccCY0M+kXv6TuYTgygdE16muVHffDKvsG
EIVdBGu6
=oNmV
-----END PGP SIGNATURE-----
Merge tag 'bitmap-6.0-rc3' of github.com:/norov/linux
Pull bitmap fixes from Yury Norov:
"Fix the reported issues, and implements the suggested improvements,
for the version of the cpumask tests [1] that was merged with commit
c41e8866c2 ("lib/test: introduce cpumask KUnit test suite").
These changes include fixes for the tests, and better alignment with
the KUnit style guidelines"
* tag 'bitmap-6.0-rc3' of github.com:/norov/linux:
lib/cpumask_kunit: add tests file to MAINTAINERS
lib/cpumask_kunit: log mask contents
lib/test_cpumask: follow KUnit style guidelines
lib/test_cpumask: fix cpu_possible_mask last test
lib/test_cpumask: drop cpu_possible_mask full test
CRYPTO_LIB_CHACHA_GENERIC doesn't need to select XOR_BLOCKS. It perhaps
was thought that it's needed for __crypto_xor, but that's not the case.
Enabling XOR_BLOCKS is problematic because the XOR_BLOCKS code runs a
benchmark when it is initialized. That causes a boot time regression on
systems that didn't have it enabled before.
Therefore, remove this unnecessary and problematic selection.
Fixes: e56e189855 ("lib/crypto: add prompts back to crypto libraries")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Current release - new code bugs:
- dsa: don't dereference NULL extack in dsa_slave_changeupper()
- dpaa: fix <1G ethernet on LS1046ARDB
- neigh: don't call kfree_skb() under spin_lock_irqsave()
Previous releases - regressions:
- r8152: fix the RX FIFO settings when suspending
- dsa: microchip: keep compatibility with device tree blobs with
no phy-mode
- Revert "net: macsec: update SCI upon MAC address change."
- Revert "xfrm: update SA curlft.use_time", comply with RFC 2367
Previous releases - always broken:
- netfilter: conntrack: work around exceeded TCP receive window
- ipsec: fix a null pointer dereference of dst->dev on a metadata
dst in xfrm_lookup_with_ifid
- moxa: get rid of asymmetry in DMA mapping/unmapping
- dsa: microchip: make learning configurable and keep it off
while standalone
- ice: xsk: prohibit usage of non-balanced queue id
- rxrpc: fix locking in rxrpc's sendmsg
Misc:
- another chunk of sysctl data race silencing
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
-----BEGIN PGP SIGNATURE-----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=azH5
-----END PGP SIGNATURE-----
Merge tag 'net-6.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Pull networking fixes from Jakub Kicinski:
"Including fixes from ipsec and netfilter (with one broken Fixes tag).
Current release - new code bugs:
- dsa: don't dereference NULL extack in dsa_slave_changeupper()
- dpaa: fix <1G ethernet on LS1046ARDB
- neigh: don't call kfree_skb() under spin_lock_irqsave()
Previous releases - regressions:
- r8152: fix the RX FIFO settings when suspending
- dsa: microchip: keep compatibility with device tree blobs with no
phy-mode
- Revert "net: macsec: update SCI upon MAC address change."
- Revert "xfrm: update SA curlft.use_time", comply with RFC 2367
Previous releases - always broken:
- netfilter: conntrack: work around exceeded TCP receive window
- ipsec: fix a null pointer dereference of dst->dev on a metadata dst
in xfrm_lookup_with_ifid
- moxa: get rid of asymmetry in DMA mapping/unmapping
- dsa: microchip: make learning configurable and keep it off while
standalone
- ice: xsk: prohibit usage of non-balanced queue id
- rxrpc: fix locking in rxrpc's sendmsg
Misc:
- another chunk of sysctl data race silencing"
* tag 'net-6.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (87 commits)
net: lantiq_xrx200: restore buffer if memory allocation failed
net: lantiq_xrx200: fix lock under memory pressure
net: lantiq_xrx200: confirm skb is allocated before using
net: stmmac: work around sporadic tx issue on link-up
ionic: VF initial random MAC address if no assigned mac
ionic: fix up issues with handling EAGAIN on FW cmds
ionic: clear broken state on generation change
rxrpc: Fix locking in rxrpc's sendmsg
net: ethernet: mtk_eth_soc: fix hw hash reporting for MTK_NETSYS_V2
MAINTAINERS: rectify file entry in BONDING DRIVER
i40e: Fix incorrect address type for IPv6 flow rules
ixgbe: stop resetting SYSTIME in ixgbe_ptp_start_cyclecounter
net: Fix a data-race around sysctl_somaxconn.
net: Fix a data-race around netdev_unregister_timeout_secs.
net: Fix a data-race around gro_normal_batch.
net: Fix data-races around sysctl_devconf_inherit_init_net.
net: Fix data-races around sysctl_fb_tunnels_only_for_init_net.
net: Fix a data-race around netdev_budget_usecs.
net: Fix data-races around sysctl_max_skb_frags.
net: Fix a data-race around netdev_budget.
...
The cpumask test suite doesn't follow the KUnit style guidelines, as
laid out in Documentation/dev-tools/kunit/style.rst. The file is
renamed to lib/cpumask_kunit.c to clearly distinguish it from other,
non-KUnit, tests.
Link: https://lore.kernel.org/lkml/346cb279-8e75-24b0-7d12-9803f2b41c73@riseup.net/
Suggested-by: Maíra Canal <mairacanal@riseup.net>
Signed-off-by: Sander Vanheule <sander@svanheule.net>
Reviewed-by: Maíra Canal <mairacanal@riseup.net>
Reviewed-by: David Gow <davidgow@google.com>
Acked-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Since cpumask_first() on the cpu_possible_mask must return at most
nr_cpu_ids - 1 for a valid result, cpumask_last() cannot return anything
larger than this value. As test_cpumask_weight() also verifies that the
total weight of cpu_possible_mask must equal nr_cpu_ids, the last bit
set in this mask must be at nr_cpu_ids - 1.
Fixes: c41e8866c2 ("lib/test: introduce cpumask KUnit test suite")
Link: https://lore.kernel.org/lkml/346cb279-8e75-24b0-7d12-9803f2b41c73@riseup.net/
Reported-by: Maíra Canal <mairacanal@riseup.net>
Signed-off-by: Sander Vanheule <sander@svanheule.net>
Tested-by: Maíra Canal <mairacanal@riseup.net>
Reviewed-by: David Gow <davidgow@google.com>
Acked-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
When the number of CPUs that can possibly be brought online is known at
boot time, e.g. when HOTPLUG is disabled, nr_cpu_ids may be smaller than
NR_CPUS. In that case, cpu_possible_mask would not be completely filled,
and cpumask_full(cpu_possible_mask) can return false for valid system
configurations.
Without this test, cpu_possible_mask contents are still constrained by
a check on cpumask_weight(), as well as tests in test_cpumask_first(),
test_cpumask_last(), test_cpumask_next(), and test_cpumask_iterators().
Fixes: c41e8866c2 ("lib/test: introduce cpumask KUnit test suite")
Link: https://lore.kernel.org/lkml/346cb279-8e75-24b0-7d12-9803f2b41c73@riseup.net/
Reported-by: Maíra Canal <mairacanal@riseup.net>
Signed-off-by: Sander Vanheule <sander@svanheule.net>
Tested-by: Maíra Canal <mairacanal@riseup.net>
Reviewed-by: David Gow <davidgow@google.com>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
While reading rs->interval and rs->burst, they can be changed
concurrently via sysctl (e.g. net_ratelimit_state). Thus, we
need to add READ_ONCE() to their readers.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since lib/cpumask.o is only built for CONFIG_SMP=y, NR_CPUS will always
be greater than 1 at compile time. This makes checking for that
condition unnecesarry, so it can be dropped.
Signed-off-by: Sander Vanheule <sander@svanheule.net>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
In the uniprocessor case, cpumask_next_wrap() can be simplified, as the
number of valid argument combinations is limited:
- 'start' can only be 0
- 'n' can only be -1 or 0
The only valid CPU that can then be returned, if any, will be the first
one set in the provided 'mask'.
For NR_CPUS == 1, include/linux/cpumask.h now provides an inline
definition of cpumask_next_wrap(), which will conflict with the one
provided by lib/cpumask.c. Make building of lib/cpumask.o again depend
on CONFIG_SMP=y (i.e. NR_CPUS > 1) to avoid the re-definition.
Suggested-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Sander Vanheule <sander@svanheule.net>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Commit 36d4b36b69 ("lib/nodemask: inline next_node_in() and
node_random()") removed the lib/nodemask.c file, but the remove didn't
happen when the patch was applied.
Reported-by: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* more new_sync_{read,write}() speedups - ITER_UBUF introduction
* ITER_PIPE cleanups
* unification of iov_iter_get_pages/iov_iter_get_pages_alloc and
switching them to advancing semantics
* making ITER_PIPE take high-order pages without splitting them
* handling copy_page_from_iter() for high-order pages properly
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQQqUNBr3gm4hGXdBJlZ7Krx/gZQ6wUCYvHI8QAKCRBZ7Krx/gZQ
62CQAPsGlbebqBeAT2pMulaGDxfLAsgz5Yf4BEaMLhPtRqFOQgD+KrZQId7Sd8O0
3IWucpTb2c4jvLlXhGMS+XWnusQH+AQ=
=pBux
-----END PGP SIGNATURE-----
Merge tag 'pull-work.iov_iter-rebased' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Pull more iov_iter updates from Al Viro:
- more new_sync_{read,write}() speedups - ITER_UBUF introduction
- ITER_PIPE cleanups
- unification of iov_iter_get_pages/iov_iter_get_pages_alloc and
switching them to advancing semantics
- making ITER_PIPE take high-order pages without splitting them
- handling copy_page_from_iter() for high-order pages properly
* tag 'pull-work.iov_iter-rebased' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (32 commits)
fix copy_page_from_iter() for compound destinations
hugetlbfs: copy_page_to_iter() can deal with compound pages
copy_page_to_iter(): don't split high-order page in case of ITER_PIPE
expand those iov_iter_advance()...
pipe_get_pages(): switch to append_pipe()
get rid of non-advancing variants
ceph: switch the last caller of iov_iter_get_pages_alloc()
9p: convert to advancing variant of iov_iter_get_pages_alloc()
af_alg_make_sg(): switch to advancing variant of iov_iter_get_pages()
iter_to_pipe(): switch to advancing variant of iov_iter_get_pages()
block: convert to advancing variants of iov_iter_get_pages{,_alloc}()
iov_iter: advancing variants of iov_iter_get_pages{,_alloc}()
iov_iter: saner helper for page array allocation
fold __pipe_get_pages() into pipe_get_pages()
ITER_XARRAY: don't open-code DIV_ROUND_UP()
unify the rest of iov_iter_get_pages()/iov_iter_get_pages_alloc() guts
unify xarray_get_pages() and xarray_get_pages_alloc()
unify pipe_get_pages() and pipe_get_pages_alloc()
iov_iter_get_pages(): sanity-check arguments
iov_iter_get_pages_alloc(): lift freeing pages array on failure exits into wrapper
...
now that we are advancing the iterator, there's no need to
treat the first page separately - just call append_pipe()
in a loop.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
mechanical change; will be further massaged in subsequent commits
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
All call sites of get_pages_array() are essenitally identical now.
Replace with common helper...
Returns number of slots available in resulting array or 0 on OOM;
it's up to the caller to make sure it doesn't ask to zero-entry
array (i.e. neither maxpages nor size are allowed to be zero).
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
... and don't mangle maxsize there - turn the loop into counting
one instead. Easier to see that we won't run out of array that
way. Note that special treatment of the partial buffer in that
thing is an artifact of the non-advancing semantics of
iov_iter_get_pages() - if not for that, it would be append_pipe(),
same as the body of the loop that follows it. IOW, once we make
iov_iter_get_pages() advancing, the whole thing will turn into
calculate how many pages do we want
allocate an array (if needed)
call append_pipe() that many times.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
same as for pipes and xarrays; after that iov_iter_get_pages() becomes
a wrapper for __iov_iter_get_pages_alloc().
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
The differences between those two are
* pipe_get_pages() gets a non-NULL struct page ** value pointing to
preallocated array + array size.
* pipe_get_pages_alloc() gets an address of struct page ** variable that
contains NULL, allocates the array and (on success) stores its address in
that variable.
Not hard to combine - always pass struct page ***, have
the previous pipe_get_pages_alloc() caller pass ~0U as cap for
array size.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
zero maxpages is bogus, but best treated as "just return 0";
NULL pages, OTOH, should be treated as a hard bug.
get rid of now completely useless checks in xarray_get_pages{,_alloc}().
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Incidentally, ITER_XARRAY did *not* free the sucker in case when
iter_xarray_populate_pages() returned 0...
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
All their callers are next to each other; all of them
want the total amount of pages and, possibly, the
offset in the partial final buffer.
Combine into a new helper (pipe_npages()), fix the
bogosity in pipe_space_for_user(), while we are at it.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
We often need to find whether the last buffer is anon or not, and
currently it's rather clumsy:
check if ->iov_offset is non-zero (i.e. that pipe is not empty)
if so, get the corresponding pipe_buffer and check its ->ops
if it's &default_pipe_buf_ops, we have an anon buffer.
Let's replace the use of ->iov_offset (which is nowhere near similar to
its role for other flavours) with signed field (->last_offset), with
the following rules:
empty, no buffers occupied: 0
anon, with bytes up to N-1 filled: N
zero-copy, with bytes up to N-1 filled: -N
That way abs(i->last_offset) is equal to what used to be in i->iov_offset
and empty vs. anon vs. zero-copy can be distinguished by the sign of
i->last_offset.
Checks for "should we extend the last buffer or should we start
a new one?" become easier to follow that way.
Note that most of the operations can only be done in a sane
state - i.e. when the pipe has nothing past the current position of
iterator. About the only thing that could be done outside of that
state is iov_iter_advance(), which transitions to the sane state by
truncating the pipe. There are only two cases where we leave the
sane state:
1) iov_iter_get_pages()/iov_iter_get_pages_alloc(). Will be
dealt with later, when we make get_pages advancing - the callers are
actually happier that way.
2) iov_iter copied, then something is put into the copy. Since
they share the underlying pipe, the original gets behind. When we
decide that we are done with the copy (original is not usable until then)
we advance the original. direct_io used to be done that way; nowadays
it operates on the original and we do iov_iter_revert() to discard
the excessive data. At the moment there's nothing in the kernel that
could do that to ITER_PIPE iterators, so this reason for insane state
is theoretical right now.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Fold pipe_truncate() into it, clean up. We can release buffers
in the same loop where we walk backwards to the iterator beginning
looking for the place where the new position will be.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
instead of setting ->iov_offset for new position and calling
pipe_truncate() to adjust ->len of the last buffer and discard
everything after it, adjust ->len at the same time we set ->iov_offset
and use pipe_discard_from() to deal with buffers past that.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
it's only used to get to the partial buffer we can add to,
and that's always the last one, i.e. pipe->head - 1.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Expand the only remaining call of push_pipe() (in
__pipe_get_pages()), combine it with the page-collecting loop there.
Note that the only reason it's not a loop doing append_pipe() is
that append_pipe() is advancing, while iov_iter_get_pages() is not.
As soon as it switches to saner semantics, this thing will switch
to using append_pipe().
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
New helper: append_pipe(). Extends the last buffer if possible,
allocates a new one otherwise. Returns page and offset in it
on success, NULL on failure. iov_iter is advanced past the
data we've got.
Use that instead of push_pipe() in copy-to-pipe primitives;
they get simpler that way. Handling of short copy (in "mc" one)
is done simply by iov_iter_revert() - iov_iter is in consistent
state after that one, so we can use that.
[Fix for braino caught by Liu Xinpeng <liuxp11@chinatelecom.cn> folded in]
[another braino fix, this time in copy_pipe_to_iter() and pipe_zero();
caught by testcase from Hugh Dickins]
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
There are only two kinds of pipe_buffer in the area used by ITER_PIPE.
1) anonymous - copy_to_iter() et.al. end up creating those and copying
data there. They have zero ->offset, and their ->ops points to
default_pipe_page_ops.
2) zero-copy ones - those come from copy_page_to_iter(), and page
comes from caller. ->offset is also caller-supplied - it might be
non-zero. ->ops points to page_cache_pipe_buf_ops.
Move creation and insertion of those into helpers - push_anon(pipe, size)
and push_page(pipe, page, offset, size) resp., separating them from
the "could we avoid creating a new buffer by merging with the current
head?" logics.
Acked-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
pipe_buffer instances of a pipe are organized as a ring buffer,
with power-of-2 size. Indices are kept *not* reduced modulo ring
size, so the buffer refered to by index N is
pipe->bufs[N & (pipe->ring_size - 1)].
Ring size can change over the lifetime of a pipe, but not while
the pipe is locked. So for any iov_iter primitives it's a constant.
Original conversion of pipes to this layout went overboard trying
to microoptimize that - calculating pipe->ring_size - 1, storing
it in a local variable and using through the function. In some
cases it might be warranted, but most of the times it only
obfuscates what's going on in there.
Introduce a helper (pipe_buf(pipe, N)) that would encapsulate
that and use it in the obvious cases. More will follow...
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Christian Brauner (Microsoft) <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Equivalent of single-segment iovec. Initialized by iov_iter_ubuf(),
checked for by iter_is_ubuf(), otherwise behaves like ITER_IOVEC
ones.
We are going to expose the things like ->write_iter() et.al. to those
in subsequent commits.
New predicate (user_backed_iter()) that is true for ITER_IOVEC and
ITER_UBUF; places like direct-IO handling should use that for
checking that pages we modify after getting them from iov_iter_get_pages()
would need to be dirtied.
DO NOT assume that replacing iter_is_iovec() with user_backed_iter()
will solve all problems - there's code that uses iter_is_iovec() to
decide how to poke around in iov_iter guts and for that the predicate
replacement obviously won't suffice.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
