kexec verbose dumps with 6.8 [was: [PATCH v4 1/7] kexec_file: add kexec_file flag to control debug printing]
Jiri Slaby
jirislaby at kernel.org
Tue Mar 12 22:58:57 PDT 2024
Hi,
On 13. 03. 24, 1:48, Baoquan He wrote:
> Hi Jiri,
>
> On 03/12/24 at 10:58am, Jiri Slaby wrote:
>> On 13. 12. 23, 6:57, Baoquan He wrote:
> ... snip...
>>> --- a/include/linux/kexec.h
>>> +++ b/include/linux/kexec.h
>> ...
>>> @@ -500,6 +500,13 @@ static inline int crash_hotplug_memory_support(void) { return 0; }
>>> static inline unsigned int crash_get_elfcorehdr_size(void) { return 0; }
>>> #endif
>>> +extern bool kexec_file_dbg_print;
>>> +
>>> +#define kexec_dprintk(fmt, ...) \
>>> + printk("%s" fmt, \
>>> + kexec_file_dbg_print ? KERN_INFO : KERN_DEBUG, \
>>> + ##__VA_ARGS__)
>>
>> This means you dump it _always_. Only with different levels.
>
> It dumped always too with pr_debug() before, I just add a switch to
> control it's pr_info() or pr_debug().
Not really, see below.
>>
>> And without any prefix whatsoever, so people see bloat like this in their
>> log now:
>> [ +0.000001] 0000000000001000-000000000009ffff (1)
>> [ +0.000002] 000000007f96d000-000000007f97efff (3)
>> [ +0.000002] 0000000000800000-0000000000807fff (4)
>> [ +0.000001] 000000000080b000-000000000080bfff (4)
>> [ +0.000002] 0000000000810000-00000000008fffff (4)
>> [ +0.000001] 000000007f97f000-000000007f9fefff (4)
>> [ +0.000001] 000000007ff00000-000000007fffffff (4)
>> [ +0.000002] 0000000000000000-0000000000000fff (2)
>
> On which arch are you seeing this? There should be one line above these
> range printing to tell what they are, like:
>
> E820 memmap:
Ah this is there too. It's a lot of output, so I took it out of context,
apparently.
> 0000000000000000-000000000009a3ff (1)
> 000000000009a400-000000000009ffff (2)
> 00000000000e0000-00000000000fffff (2)
> 0000000000100000-000000006ff83fff (1)
> 000000006ff84000-000000007ac50fff (2)
It should all be prefixed like kdump: or kexec: in any way.
>> without actually knowing what that is.
>>
>> There should be nothing logged if that is not asked for and especially if
>> kexec load went fine, right?
>
> Right. Before this patch, those pr_debug() were already there. You need
> enable them to print out like add '#define DEBUG' in *.c file, or enable
> the dynamic debugging of the file or function.
I think it's perfectly fine for DEBUG builds to print this out. And many
(all major?) distros use dyndbg, so it used to print nothing by default.
> With this patch applied,
> you only need specify '-d' when you execute kexec command with
> kexec_file load interface, like:
>
> kexec -s -l -d /boot/vmlinuz-xxxx.img --initrd xxx.img --reuse-cmdline
Perhaps our (SUSE) tooling passes -d? But I am seeing this every time I
boot.
No, it does not seem so:
load.sh[915]: Starting kdump kernel load; kexec cmdline: /sbin/kexec -p
/var/lib/kdump/kernel --append=" loglevel=7 console=tty0 console=ttyS0
video=1920x1080,1024x768,800x600 oops=panic
lsm=lockdown,capability,integrity,selinux sysrq=yes reset_devices
acpi_no_memhotplug cgroup_disable=memory nokaslr numa=off irqpoll
nr_cpus=1 root=kdump rootflags=bind rd.udev.children-max=8
disable_cpu_apicid=0 panic=1" --initrd=/var/lib/kdump/initrd -a
> For kexec_file load, it is not logging if not specifying '-d', unless
> you take way to make pr_debug() work in that file.
So is -d detection malfunctioning under some circumstances?
>> Can this be redesigned, please?
>
> Sure, after making clear what's going on with this, I will try.
>
>>
>> Actually what was wrong on the pr_debug()s? Can you simply turn them on from
>> the kernel when -d is passed to kexec instead of all this?
>
> Joe suggested this during v1 reviewing:
> https://lore.kernel.org/all/1e7863ec4e4ab10b84fd0e64f30f8464d2e484a3.camel@perches.com/T/#u
>
>>
>> ...
>>> --- a/kernel/kexec_core.c
>>> +++ b/kernel/kexec_core.c
>>> @@ -52,6 +52,8 @@ atomic_t __kexec_lock = ATOMIC_INIT(0);
>>> /* Flag to indicate we are going to kexec a new kernel */
>>> bool kexec_in_progress = false;
>>> +bool kexec_file_dbg_print;
>>
>> Ugh, and a global flag for this?
>
> Yeah, kexec_file_dbg_print records if '-d' is specified when 'kexec'
> command executed. Anything wrong with the global flag?
Global variables are frowned upon. To cite coding style: unless you
**really** need them. Here, it looks like you do not.
thanks,
--
js
suse labs
More information about the kexec
mailing list