Messages in this thread | | | From | Marco Elver <> | Date | Mon, 8 Mar 2021 19:23:34 +0100 | Subject | Re: [PATCH 2/2] lib/vsprintf: reduce space taken by no_hash_pointers warning |
| |
On Mon, 8 Mar 2021 at 18:23, Petr Mladek <pmladek@suse.com> wrote: [...] > > I'm actually concerned about both. Platforms (and boot loaders) may > > have limitations for kernel image size, too. > > Static memory consumption is also more easily measured, so I tend > > to run bloat-o-meter, and dive into anything that adds more than 1 KiB. > > And yes, this message is a low-hanging fruit... > > OK, I wondered how big trick does the __initconst on its own. > > 1. I compiled kernel without this patchset: > > $# ll /boot/vmlinux-5.12.0-rc2-default+.bz2 > -rwxr-xr-x 1 root root 18911364 Mar 8 15:58 /boot/vmlinux-5.12.0-rc2-default+.bz2 > > 2. With this patchset: > > $# ll /boot/vmlinux-5.12.0-rc2-default+.bz2 > -rwxr-xr-x 1 root root 18910767 Mar 8 16:16 /boot/vmlinux-5.12.0-rc2-default+.bz2 > $# echo $((18910767 - 18911364)) > -597 > > 3. With the patch below: > > $# ll /boot/vmlinux-5.12.0-rc2-default+.bz2 > -rwxr-xr-x 1 root root 18910906 Mar 8 16:58 /boot/vmlinux-5.12.0-rc2-default+.bz2 > $# echo $((18910906 - 18911364)) > -458 > > This patchset saves 139B more than a simple array. > > > Well, I am a bit confused. I have tried to keep the strings as a > static variable outside the function: > > static const char *no_hash_pointers_warning[] __initconst = { > ... > > and I got the following build error: > > CC lib/vsprintf.o > lib/vsprintf.c:2097:20: error: no_hash_pointers_warning causes a section type conflict with __setup_str_no_hash_pointers_enable > static const char *no_hash_pointers_warning[] __initconst = {
This does not place the strings themselves into the initconst section, but only the array of pointers to them. So, with 13 lines, we're merely saving 13*sizeof(char*) after init, which does not resolve Geert's problem of runtime overhead.
To dealloc the string text itself (remove the section), each line must be placed into a 'char[N] __initconst' (or 'char [M][N] __initconst' if we split the lines).
> ^~~~~~~~~~~~~~~~~~~~~~~~ > In file included from ./include/linux/printk.h:6:0, > from ./include/linux/kernel.h:16, > from ./include/linux/clk.h:13, > from lib/vsprintf.c:22: > ./include/linux/init.h:315:20: note: ‘__setup_str_no_hash_pointers_enable’ was declared here > static const char __setup_str_##unique_id[] __initconst \ > ^ > ./include/linux/init.h:330:2: note: in expansion of macro ‘__setup_param’ > __setup_param(str, fn, fn, 1) > ^~~~~~~~~~~~~ > lib/vsprintf.c:2127:1: note: in expansion of macro ‘early_param’ > early_param("no_hash_pointers", no_hash_pointers_enable); > ^~~~~~~~~~~ > > > I solved this be defining the array inside the function that is marked > __init. But I am not sure if it is the correct solution. And I wonder > why the original patch did not have this problem. > > Also I am curious why the array reduced the size of the binary so > significantly in compare with the const strings used as pr_warn() > arguments. It might depend on the compression method or??? > > > Anyway, here is the patch that works for me and reduced the size of > the binary considerably: > > diff --git a/lib/vsprintf.c b/lib/vsprintf.c > index 4a14889ccb35..af01edae0d86 100644 > --- a/lib/vsprintf.c > +++ b/lib/vsprintf.c > @@ -2096,24 +2096,30 @@ EXPORT_SYMBOL_GPL(no_hash_pointers); > > static int __init no_hash_pointers_enable(char *str) > { > + int i; > + const char *no_hash_pointers_warning[] = { > + "**********************************************************", > + "** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **", > + "** **", > + "** This system shows unhashed kernel memory addresses **", > + "** via the console, logs, and other interfaces. This **", > + "** might reduce the security of your system. **", > + "** **", > + "** If you see this message and you are not debugging **", > + "** the kernel, report this immediately to your system **", > + "** administrator! **", > + "** **", > + "** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **", > + "**********************************************************", > + }; > +
This has no __initconst optimization (no runtime savings), and the compiler places these strings into the data section and the above array is just an array of pointers to them.
> if (no_hash_pointers) > return 0; > > no_hash_pointers = true; > > - pr_warn("**********************************************************\n"); > - pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n"); > - pr_warn("** **\n"); > - pr_warn("** This system shows unhashed kernel memory addresses **\n"); > - pr_warn("** via the console, logs, and other interfaces. This **\n"); > - pr_warn("** might reduce the security of your system. **\n"); > - pr_warn("** **\n"); > - pr_warn("** If you see this message and you are not debugging **\n"); > - pr_warn("** the kernel, report this immediately to your system **\n");
While we're here: This paragraph can be shortened by saying what kernel/trace/trace.c says ("..., report this immediately to your vendor!") which avoids the "administrator! <lots of wasted spaces>".
> - pr_warn("** administrator! **\n"); > - pr_warn("** **\n"); > - pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n"); > - pr_warn("**********************************************************\n"); > + for (i = 0; i < ARRAY_SIZE(no_hash_pointers_warning); i++) > + pr_warn("%s\n", no_hash_pointers_warning[i]);
My guess is that the savings came from repeated calls to pr_warn() and reduction in code-size and compression working better.
> return 0; > } > > > Honestly, I do not want to spend much more time on this. I made the > test out of curiosity. > > Feel free to provide the patch using the array, ideally with some > numbers how it helps. But please _avoid_ the indirection via > > const int lines[] = { 0, 1, -1, 2, 3, 4, -1, 5, 6, 7, -1, 1, 0 };
We can probably do without this, but we'll have duplicated lines stored in the initconst section.
> and also _avoid_ all the hardcoded constants, like: > > no_hash_pointers_warning[8][55]
We'll need this if we want __initconst. But perhaps we do not have to split it by lines, so we can get away with a char[].
> and > > pr_warn("**%54s**\n" > > They are error prone and hard to maintain. Such tricks are not > worth it from my POV.
I can send the version with a single 'static char[] __initconst', but that version doesn't dedup lines and requires more init memory. I don't know if we can make everybody happy here, we have to sacrifice something: readability or space.
| |