00:36 <derek0883> anyone knows how to access kernel global variables?

00:36 <fche> @var("foo") or $foo from a probe context where $foo is in scope

00:39 <derek0883> cool, thanks @fche

00:46 <derek0883> @fche, printf("command = %p\n", @var("boot_command_line")); works. however. printf("command = %s\n", kernel_string(@var("boot_command_line"))); got runtime error. do you know why?

00:48 <fche> hm no, is the %p pointer value plausible?

00:55 <derek0883> yes, %p result exactly matches sudo cat /proc/kallsyms | grep boot_command_line

00:55 <fche> hm I wonder if it's some kernel pointer checking paranoia at work

00:56 <fche> what's your kernel version?

00:56 <derek0883> 4.15.18

00:57 <derek0883> ERROR: kernel string copy fault at 0x00000000f6e778ec [man error::fault] near identifier 'kernel_string' at /usr/local/share/systemtap/tapset/linux/conversions.stp:18:10

00:57 <fche> can you paste your stap script more completely?

00:57 <fche> stap -e 'probe kernel.function("do_exit") { printf("command = %s\n", kernel_string(@var("boot_command_line"))) }'

00:58 <fche> is what I'm using here (kernel probe to put it in kernel context), but I'm not getting errors, only empty strings

00:58 <fche> are these ordinary char*'s in the kernel or something different?

00:58 <derek0883> #! /usr/bin/env stap

00:58 <derek0883> probe begin {

00:58 <derek0883> printf("command = %p\n", @var("boot_command_line"));

00:58 <derek0883> printf("command = %s\n", kernel_string(@var("boot_command_line")));

00:58 <derek0883> exit();

00:58 <derek0883> }

00:59 <fche> (btw that won't work on 4.4+ stap; you'll need --compatible=4.3 or something, because probe begin is not a @var()capable context directly)

00:59 <fche> but anyway

00:59 <fche> I see no errors here 5.9.10-200.fc33.x86_64 as the kernel, just empty strings

01:00 <derek0883> got it, thanks

01:00 <fche> oh I'm also using stap 4.4, that MIGHT make a difference

01:00 <derek0883> Systemtap translator/driver (version 4.1/0.165

01:01 <derek0883> I'm using old version, maybe that's difference

01:01 <fche> conceivably .......

01:01 <fche> any other such global char*'s you can think of printing?

01:02 <fche> printf("command = %.*M\n", 20, @var("boot_command_line")); <<< hex-dumps that

01:02 <fche> and here on my box that's a bunch of zeroes

01:02 <fche> wonder why

01:02 <derek0883> another one is command_line, but I got same result. let me think other one.

01:03 <fche> yeah no read error hear, just \0's.

01:03 <fche> (I can trigger read faults with bad addresses like 0x0)

01:06 <derek0883> printf("command = %s\n", kernel_string(@var("last_unloaded_module")));

01:06 <derek0883> this last_unloaded_module works

01:06 <fche> ditto here

01:07 <fche> wonder if something's special about those few you chose - maybe copied at boot time somewhere, then cleared in this copy

01:07 <fche> dunno

01:08 <derek0883> I think you are right

01:08 <derek0883> printf("command = %s\n", kernel_string(@var("saved_command_line")));

01:08 <derek0883> this one works

01:08 <fche> wild

17:59 <derek0883> Hi @fche I can get address by @var("nmi_desc"), nmi_desc is a array name, do you know how can I access like an array? currently I can only access nmi_desc[0]. is there any better way that I don't have to manually calculate offset? thanks.

21:47 <fche> if the type is not forward-declared or something incomplete, you should be able to say @var("nmi_desc")[1] and so on