Hi all,
When printk-indexing is enabled, each printk() invocation emits a
pi_entry structure, containing the format string and other information
related to its location in the kernel sources. This is even true when
the printk() is protected by an always-false check, as is typically the
case for debug messages: while the actual code to print the message is
optimized out by the compiler, the pi_entry structure is still emitted.
Hence when debugging is disabled, this leads to the inclusion in the
index of lots of printk formats that cannot be emitted by the current
kernel.
This series fixes that for the common debug helpers under include/.
It reduces the size of an arm64 defconfig kernel with
CONFIG_PRINTK_INDEX=y by ca. 1.5 MiB, or 28% of the overhead of
enabling CONFIG_PRINTK_INDEX=y.
Notes:
- netdev_(v)dbg() and netif_(v)dbg() are not affected, as
net{dev,if}_printk() do not implement printk-indexing, except
for the single global internal instance of __netdev_printk().
- This series fixes only debug code in global header files under
include/. There are more cases to fix in subsystem-specific header
files and in sources files.
Thanks for your comments!
Geert Uytterhoeven (4):
printk: Let no_printk() use _printk()
dev_printk: Add and use dev_no_printk()
dyndbg: Use *no_printk() helpers
ceph: Use no_printk() helper
include/linux/ceph/ceph_debug.h | 18 +++++++-----------
include/linux/dev_printk.h | 25 +++++++++++++------------
include/linux/dynamic_debug.h | 4 ++--
include/linux/printk.h | 2 +-
4 files changed, 23 insertions(+), 26 deletions(-)
--
2.34.1
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
