xlog_tic_add_region() is used to trace the regions being added to a
log ticket to provide information in the situation where a ticket
reservation overrun occurs. The information gathered is stored int
the ticket, and dumped if xlog_print_tic_res() is called.
For a front end struct xfs_trans overrun, the ticket only contains
reservation tracking information - the ticket is never handed to the
log so has no regions attached to it. The overrun debug information in this
case comes from xlog_print_trans(), which walks the items attached
to the transaction and dumps their attached formatted log vectors
directly. It also dumps the ticket state, but that only contains
reservation accounting and nothing else. Hence xlog_print_tic_res()
never dumps region or overrun information from this path.
xlog_tic_add_region() is actually called from xlog_write(), which
means it is being used to track the regions seen in a
CIL checkpoint log vector chain. In looking at CIL behaviour
recently, I've seen 32MB checkpoints regularly exceed 250,000
regions in the LV chain. The log ticket debug code can track *15*
regions. IOWs, if there is a ticket overrun in the CIL code, the
ticket region tracking code is going to be completely useless for
determining what went wrong. The only thing it can tell us is how
much of an overrun occurred, and we really don't need extra debug
information in the log ticket to tell us that.
Indeed, the main place we call xlog_tic_add_region() is also adding
up the number of regions and the space used so that xlog_write()
knows how much will be written to the log. This is exactly the same
information that log ticket is storing once we take away the useless
region tracking array. Hence xlog_tic_add_region() is not useful,
but can be called 250,000 times a CIL push...
Just strip all that debug "information" out of the of the log ticket
and only have it report reservation space information when an
overrun occurs. This also reduces the size of a log ticket down by
about 150 bytes...
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Chandan Babu R <chandan.babu@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Current xlog_write() adds op headers to the log manually for every
log item region that is in the vector passed to it. While
xlog_write() needs to stamp the transaction ID into the ophdr, we
already know it's length, flags, clientid, etc at CIL commit time.
This means the only time that xlog write really needs to format and
reserve space for a new ophdr is when a region is split across two
iclogs. Adding the opheader and accounting for it as part of the
normal formatted item region means we simplify the accounting
of space used by a transaction and we don't have to special case
reserving of space in for the ophdrs in xlog_write(). It also means
we can largely initialise the ophdr in transaction commit instead
of xlog_write, making the xlog_write formatting inner loop much
tighter.
xlog_prepare_iovec() is now too large to stay as an inline function,
so we move it out of line and into xfs_log.c.
Object sizes:
text data bss dec hex filename
1125934 305951 484 1432369 15db31 fs/xfs/built-in.a.before
1123360 305951 484 1429795 15d123 fs/xfs/built-in.a.after
So the code is a roughly 2.5kB smaller with xlog_prepare_iovec() now
out of line, even though it grew in size itself.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chandan Babu R <chandan.babu@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
To include log op headers directly into the log iovec regions that
the ophdrs wrap, we need to move the buffer alignment code from
xlog_finish_iovec() to xlog_prepare_iovec(). This is because the
xlog_op_header is only 12 bytes long, and we need the buffer that
the caller formats their data into to be 8 byte aligned.
Hence once we start prepending the ophdr in xlog_prepare_iovec(), we
are going to need to manage the padding directly to ensure that the
buffer pointer returned is correctly aligned.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Chandan Babu R <chandan.babu@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
We currently set the log ticket client ID when we reserve a
transaction. This client ID is only ever written to the log by
a CIL checkpoint or unmount records, and so anything using a high
level transaction allocated through xfs_trans_alloc() does not need
a log ticket client ID to be set.
For the CIL checkpoint, the client ID written to the journal is
always XFS_TRANSACTION, and for the unmount record it is always
XFS_LOG, and nothing else writes to the log. All of these operations
tell xlog_write() exactly what they need to write to the log (the
optype) and build their own opheaders for start, commit and unmount
records. Hence we no longer need to set the client id in either the
log ticket or the xfs_trans.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Chandan Babu R <chandan.babu@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Remove the final case where xlog_write() has to prepend an opheader
to a log transaction. Similar to the start record, the commit record
is just an empty opheader with a XLOG_COMMIT_TRANS type, so we can
just make this the payload for the region being passed to
xlog_write() and remove the special handling in xlog_write() for
the commit record.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chandan Babu R <chandan.babu@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Remove another case where xlog_write() has to prepend an opheader to
a log transaction. The unmount record + ophdr is smaller than the
minimum amount of space guaranteed to be free in an iclog (2 *
sizeof(ophdr)) and so we don't have to care about an unmount record
being split across 2 iclogs.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Chandan Babu R <chandan.babu@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
So move the one-off start record writing in xlog_write() out into
the static header that the CIL push builds to write into the log
initially. This simplifes the xlog_write() logic a lot.
pahole on x86-64 confirms that the xlog_cil_trans_hdr is correctly
32 bit aligned and packed for copying the log op and transaction
headers directly into the log as a single log region copy.
struct xlog_cil_trans_hdr {
struct xlog_op_header oph[2]; /* 0 24 */
struct xfs_trans_header thdr; /* 24 16 */
struct xfs_log_iovec lhdr[2]; /* 40 32 */
/* size: 72, cachelines: 2, members: 3 */
/* last cacheline: 8 bytes */
};
A wart is needed to handle the fact that length of the region the
opheader points to doesn't include the opheader length. hence if
we embed the opheader, we have to substract the opheader length from
the length written into the opheader by the generic copying code.
This will eventually go away when everything is converted to
embedded opheaders.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chandan Babu R <chandan.babu@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
It is static code deep in the middle of the CIL push logic. Factor
it out into a helper so that it is clear and easy to modify
separately.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Chandan Babu R <chandan.babu@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
These functions are currently only available on architectures that have
my_syscall6() macro implemented. Since these functions use malloc(),
malloc() uses mmap(), mmap() depends on my_syscall6() macro.
On architectures that don't support my_syscall6(), these function will
always return NULL with errno set to ENOSYS.
Acked-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
size_t strnlen(const char *str, size_t maxlen);
The strnlen() function returns the number of bytes in the string
pointed to by sstr, excluding the terminating null byte ('\0'), but at
most maxlen. In doing this, strnlen() looks only at the first maxlen
characters in the string pointed to by str and never beyond str[maxlen-1].
The first use case of this function is for determining the memory
allocation size in the strndup() function.
Link: https://lore.kernel.org/lkml/CAOG64qMpEMh+EkOfjNdAoueC+uQyT2Uv3689_sOr37-JxdJf4g@mail.gmail.com
Suggested-by: Alviro Iskandar Setiawan <alviro.iskandar@gnuweeb.org>
Acked-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Implement basic dynamic allocator functions. These functions are
currently only available on architectures that have nolibc mmap()
syscall implemented. These are not a super-fast memory allocator,
but at least they can satisfy basic needs for having heap without
libc.
Cc: David Laight <David.Laight@ACULAB.COM>
Acked-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Implement `offsetof()` and `container_of()` macro. The first use case
of these macros is for `malloc()`, `realloc()` and `free()`.
Acked-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Implement mmap() and munmap(). Currently, they are only available for
architecures that have my_syscall6 macro. For architectures that don't
have, this function will return -1 with errno set to ENOSYS (Function
not implemented).
This has been tested on x86 and i386.
Notes for i386:
1) The common mmap() syscall implementation uses __NR_mmap2 instead
of __NR_mmap.
2) The offset must be shifted-right by 12-bit.
Acked-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
On i386, the 6th argument of syscall goes in %ebp. However, both Clang
and GCC cannot use %ebp in the clobber list and in the "r" constraint
without using -fomit-frame-pointer. To make it always available for
any kind of compilation, the below workaround is implemented.
1) Push the 6-th argument.
2) Push %ebp.
3) Load the 6-th argument from 4(%esp) to %ebp.
4) Do the syscall (int $0x80).
5) Pop %ebp (restore the old value of %ebp).
6) Add %esp by 4 (undo the stack pointer).
Cc: x86@kernel.org
Cc: llvm@lists.linux.dev
Link: https://lore.kernel.org/lkml/2e335ac54db44f1d8496583d97f9dab0@AcuMS.aculab.com
Suggested-by: David Laight <David.Laight@ACULAB.COM>
Acked-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Building with clang yields the following error:
```
<inline asm>:3:1: error: _start changed binding to STB_GLOBAL
.global _start
^
1 error generated.
```
Make sure only specify one between `.global _start` and `.weak _start`.
Remove `.global _start`.
Cc: llvm@lists.linux.dev
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Acked-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Replace `asm` with `__asm__` to support compilation with -std flag.
Using `asm` with -std flag makes GCC think `asm()` is a function call
instead of an inline assembly.
GCC doc says:
For the C language, the `asm` keyword is a GNU extension. When
writing C code that can be compiled with `-ansi` and the `-std`
options that select C dialects without GNU extensions, use
`__asm__` instead of `asm`.
Link: https://gcc.gnu.org/onlinedocs/gcc/Basic-Asm.html
Reported-by: Alviro Iskandar Setiawan <alviro.iskandar@gnuweeb.org>
Acked-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
When building with gcc at -O0 we're seeing link errors due to the
"environ" variable being referenced by getenv(). The problem is that
at -O0 gcc will not inline getenv() and will not drop the external
reference. One solution would be to locally declare the variable as
weak, but then it would appear in all programs even those not using
it, and would be confusing to users of getenv() who would forget to
set environ to envp.
An alternate approach used in this patch consists in always inlining
the outer part of getenv() that references this extern so that it's
always dropped when not used. The biggest part of the function was
now moved to a new function called _getenv() that's still not inlined
by default.
Reported-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Tested-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
clang wants to use strlen() for __builtin_strlen() at -O0. We don't
really care about -O0 but it at least ought to build, so let's make
sure we don't choke on this, by dropping the optimizationn for
constant strings in this case.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The Makefile in tools/ is used to forward options to the makefiles
in the various subdirs. Let's add nolibc there so that it becomes
possible to make tools/nolibc_headers_standalone from the main tree
to simply create a completely usable sysroot.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This provides a target "headers_standalone" which installs the nolibc's
arch-specific headers with "arch.h" taken from the current arch (or a
concatenation of both i386 and x86_64 for arch=x86), then installs
kernel headers. This creates a convenient sysroot which is directly
usable by a bare-metal compiler to create any executable.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
- POLLIN etc were missing, so poll() could only be used with timeouts.
- WNOHANG was not defined and is convenient to check if a child is still
running
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This is essentially for completeness as it's not the most often used
in regtests.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
We need these functions all the time, including when checking environment
variables and parsing command-line arguments. These implementations were
optimized to show optimal code size on a wide range of compilers (22 bytes
return included for strcmp(), 33 for strncmp()).
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
%p remains quite useful in test code, and the code path can easily be
merged with the existing "%x" thus only adds ~50 bytes, thus let's
add it.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This implementation relies on an extern definition of the environ
variable, that the caller must declare and initialize from envp.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
It's often convenient to support this, especially in test programs where
a NULL may correspond to an allocation error or a non-existing value.
Let's make printf("%s") support being passed a NULL. In this case it
prints "(null)" like glibc's printf().
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
libgcc uses it for certain divide functions, so it must be exported. Like
for memset() we do that in its own section so that the linker can strip
it when not needed.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Now that a few basic include files are provided, some simple portable
programs may build, which will save them from having to surround their
includes with #ifndef NOLIBC. This patch mentions how to proceed, and
enumerates the list of files that are covered.
A comprehensive list of required include files is available here:
https://en.cppreference.com/w/c/header
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The time() syscall is used by a few simple applications, and is trivial
to implement based on gettimeofday() that we already have. Let's create
the file to ease porting and provide the function. It never returns any
error, though it may segfault in case of invalid pointer, like other
implementations relying on gettimeofday().
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This function is normally found in signal.h, and providing the file
eases porting of existing programs. Let's move it there.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This call is trivial to implement based on select() to complete sleep()
and msleep(), let's add it.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
These functions are normally provided by unistd.h. For ease of porting,
let's create the file and move them there.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This allows us to provide a minimal errno.h to ease porting applications
that use it.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
"clang -Os" and "gcc -Ofast" without -ffreestanding may ignore memset()
and memmove(), hoping to provide their builtin equivalents, and finally
not find them. Thus we must export these functions for these rare cases.
Note that as they're set in their own sections, they will be eliminated
by the linker if not used. In addition, they do not prevent gcc from
identifying them and replacing them with the shorter "rep movsb" or
"rep stosb" when relevant.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
These ones are often used and commonly set by applications to fallback
values. Let's fix them both to agree on PATH_MAX=4096 by default, as is
already present in linux/limits.h.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
By doing so we can link together multiple C files that have been compiled
with nolibc and which each have a _start symbol.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Some functions like raise() and memcpy() are permanently exported because
they're needed by libgcc on certain platforms. However most of the time
they are not needed and needlessly take space.
Let's move them to their own sub-section, called .text.nolibc_<function>.
This allows ld to get rid of them if unused when passed --gc-sections.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
While these functions are often dangerous, forcing the user to work
around their absence is often much worse. Let's provide small versions
of each of them. The respective sizes in bytes on a few architectures
are:
strncat(): x86:0x33 mips:0x68 arm:0x3c
strlcat(): x86:0x25 mips:0x4c arm:0x2c
The two are quite different, and strncat() is even different from
strncpy() in that it limits the amount of data it copies and will always
terminate the output by one zero, while strlcat() will always limit the
total output to the specified size and will put a zero if possible.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
These are minimal variants. strncpy() always fills the destination for
<size> chars, while strlcpy() copies no more than <size> including the
zero and returns the source's length. The respective sizes on various
archs are:
strncpy(): x86:0x1f mips:0x30 arm:0x20
strlcpy(): x86:0x17 mips:0x34 arm:0x1a
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The direction test inside the loop was not always completely optimized,
resulting in a larger than necessary function. This change adds a
direction variable that is set out of the loop. Now the function is down
to 48 bytes on x86, 32 on ARM and 68 on mips. It's worth noting that other
approaches were attempted (including relying on the up and down functions)
but they were only slightly beneficial on x86 and cost more on others.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Till now memcpy() relies on memmove(), but it's always included for libgcc,
so we have a larger than needed function. Let's implement two unidirectional
variants to copy from bottom to top and from top to bottom, and use the
former for memcpy(). The variants are optimized to be compact, and at the
same time the compiler is sometimes able to detect the loop and to replace
it with a "rep movsb". The new function is 24 bytes instead of 52 on x86_64.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
These syscalls never fail so there is no need to extract and set errno
for them.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
raise() doesn't set errno, so there's no point calling kill(), better
call sys_kill(), which also reduces the function's size.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The build of printf() on mips requires libgcc for functions __ashldi3 and
__lshrdi3 due to 64-bit shifts when scanning the input number. These are
not really needed in fact since we scan the number 4 bits at a time. Let's
arrange the loop to perform two 32-bit shifts instead on 32-bit platforms.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Let's pass a vararg to open() so that it remains compatible with existing
code. The arg is only dereferenced when flags contain O_CREAT. The function
is generally not inlined anymore, causing an extra call (total 16 extra
bytes) but it's still optimized for constant propagation, limiting the
excess to no more than 16 bytes in practice when open() is called without
O_CREAT, and ~40 with O_CREAT, which remains reasonable.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
It doesn't contain the text for the error codes, but instead displays
"errno=" followed by the errno value. Just like the regular errno, if
a non-empty message is passed, it's placed followed with ": " on the
output before the errno code. The message is emitted on stderr.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
These ones are found in some examples found in man pages and ease
portability tests.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This adds a minimal vfprintf() implementation as well as the commonly
used fprintf() and printf() that rely on it.
For now the function supports:
- formats: %s, %c, %u, %d, %x
- modifiers: %l and %ll
- unknown chars are considered as modifiers and are ignored
It is designed to remain minimalist, despite this printf() is 549 bytes
on x86_64. It would be wise not to add too many formats.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
We'll use it to write substrings. It relies on a simpler _fwrite() that
only takes one size. fputs() was also modified to rely on it.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
