When told to force load a module, we were removing only the value of
vermagic instead of the complete entry.
Philippe De Swert (philippe.deswert@jollamobile.com) sent a patch that
was additionally mangling also the last two chars of the key
("vermagic="). Instead of creating an invalid entry in .modinfo section
like this, this patch removes the complete entry, key + value, by
zeroing the entire string.
Much thanks to Philippe who found the issue and pointed to the fix.
If we are replacing a lower priority module (due to its location), we
already created a kmod_module, but didn't open the file for reading its
symbols. This means mod->kmod won't be NULL, and this is just ok. Since
all the functions freeing stuff below the previous assert already takes
NULL into consideration, it's safe to just unref mod->kmod and let the
right thing happens.
We index modules in depmod by it's uncompressed relative path, not
relative path. We didn't notice this bug before since this function is
only triggered if we release a module to be replaced by one of higher
priority.
Also fix a leftover log message referring to relpath instead of
uncrelpath.
This also fixes a bug in "e6996c5 rmmod: route all messages to syslog if
told to" in which "+ verbose" was removed. Instead of letting verbose
add to kmod_get_log_priority(), let it be similar to the other programs
instead.
When user supplied --help/-h, program should output to stdout the usage,
not to stderr. It's the expected behavior, what the user asked for,
not something to log or an error.
When we are logging to stderr we are previously relying on libkmod
sending it to the default location in case we are not asked to log to
syslog. The problem is that modprobe may be used in scripts that don't
want to log to syslog (since they are not daemons, like scripts to
generate initrd) and then it's difficult to know where the message comes
from.
This patch treats only the messages coming from libkmod.
Once we read all we need from a module, unref it so any resource taken
by it (including the mmap to access the file in libkmod) will be
dropped. This drastically reduces the number of open file descriptors
and also the memory needed, with no performance penalties. Rather,
there's a small speedup of ~2.6%.
Running depmod in a laptop with 2973 modules and comparing the number of
open file descriptors for kmod-10, before and after the last patches to
depmod (caaf438cb6 and HEAD) we have:
Before: 2980 simultaneously open fds
After: 7 simultaneously open fds
kmod-10: 7 simultaneously open fds
So now we have the speedup of caching the file in kmod_module without
the drawback of increasing the number of open file descriptors.
In depmod_module_add() we already called kmod_module_get_name() and
copied the string to our struct. Use it instead of calling again and
again the libkmod function.
The overall goal is to coalesce the accesses to a file that is the
backend of a module. This commit addresses the calls to
kmod_module_get_get_dependency_symbols(). Calling it earlier, while we
are iterating the modules allows us to free the struct kmod of each
module much sooner. We are still not freeing it since there are other
places that must be refactored first.
There's a performance penalty of ~2.5% from previous commit.
The overall goal is to coalesce the accesses to a file that is the
backend of a module. This commit addresses the calls to
kmod_module_get_info(). Calling it earlier, while we are iterating the
modules allows us to free the struct kmod of each module much sooner. We
are still not freeing it since there are other places that must be
refactored first.
A nice side effect is that this commit reduces in ~33% the calls to
malloc(), giving a speedup of ~6% for cold caches (reproduced on only 1
laptop).
Before this commit the build system failed at late state with
non-helpful message when xsltproc was not available.
Making all in man
GEN depmod.d.5
/bin/sh: --nonet: command not found
make[2]: *** [depmod.d.5] Error 127
make[1]: *** [all-recursive] Error 1
make: *** [all] Error 2
If we are accessing several times the modules and reading some sections
by sucessive calls to the functions below, we are incurring in a penalty
of having to open, parse the header and close the file. For each
function.
- kmod_module_get_info()
- kmod_module_get_versions()
- kmod_module_get_symbols()
- kmod_module_get_dependency_symbols()
These functions are particularly important to depmod. It calls all of
them, for each module. Moreover there's a huge bottleneck in the open
operation if we are using compression. Every time we open the module we
need to uncompress the file and after getting the information we need we
discard the result. This is clearly shown by profiling depmod with perf
(record + report), using compressed modules:
64.07% depmod libz.so.1.2.7 [.] 0x00000000000074b8 ◆
18.18% depmod libz.so.1.2.7 [.] crc32 ▒
2.42% depmod libz.so.1.2.7 [.] inflate ▒
1.17% depmod libc-2.16.so [.] __memcpy_ssse3_back ▒
0.96% depmod [kernel.kallsyms] [k] copy_user_generic_string ▒
0.89% depmod libc-2.16.so [.] __strcmp_sse42 ▒
0.82% depmod [kernel.kallsyms] [k] hrtimer_interrupt ▒
0.77% depmod libc-2.16.so [.] _int_malloc ▒
0.44% depmod kmod-nolib [.] kmod_elf_get_strings ▒
0.41% depmod kmod-nolib [.] kmod_elf_get_dependency_symbols ▒
0.37% depmod kmod-nolib [.] kmod_elf_get_section ▒
0.36% depmod kmod-nolib [.] kmod_elf_get_symbols
...
Average of running depmod 5 times, dropping caches between them, in a
slow spinning disk:
Before: 12.25 +- 0.20
After: 8.20 +- 0.21
m-i-t: 9.62 +- 0.27
So this patch leads to an improvement of ~33% over unpatched version,
ending up with 15% speedup over module-init-tools.
Although the hash table implementation allows passing a callback function
to free a value when it is removed from the hash table, hash_del() wasn't
freeing it if it was provided. Now it does.
As a bonus, it now checks if the callback is set in hash_add() as well.
This gives the test cases the ability to supply files that must be
checked after the test is run, rather than just checking stdout/stderr.
This is intended to be used with tools that generate files, like depmod.
It includes a poor's man implementation of a "check for differences in
files". Not really optimized, but it's simple enough and does what it
proposes to.
We now index the modules by uncompressed-relative-path instead of
relative-path. This is because the file modules.order, coming from
kernel, always comes with uncompressed paths. This fixes the issue of
not sorting the aliases correctly due to paths not matching when using
compressed modules.
This is a broken option that only leads to misery and incompatabilities
with other systems. Kbuild doesn't come close to supporting directories
other than /lib/modules with several targets simply failing without
hacky fixes. Simply remove the option and all traces of it, as it
doesn't make sense in today's world.
Clang doesn't treat unknown warnings flags as an error, but rather as a
warning. The result is that the detection for whic CFLAGS are supported
by this compiler will not work, since the compilation will succeed.
With this patch we now successfully detect clang doesn't support
-Wlogical-op, as opposed to previous behavior:
checking if clang supports flag -Wlogical-op in envvar CFLAGS... no
We use this macro only for LDFLAGS and CFLAGS, so it's safe to stash
-Werror there.
Lucas De Marchi
Lucas De Marchi
Sami Kerola
Leandro Pereira
Dave Reisner
Jan Engelhardt