[Kernel, init.d] flawed init.d support stemming from doomlord kernels - Sony Ericsson Xperia Mini, Mini Pro, Xperia Pro, A

It seems that all Xperia Kernels that stem back to the DoomLord kernel start the init.d execution twice. While you may think "better twice than never" the nearly parallel execution of scripts can create problems if they concurrently manipulate CPU related tables - or fail to do so due to security mechanisms built in. I was hunting the problem that cpu-clock manipulation from init.d did not work for the scripts generated by System Tuner - finally resulting in this finding.
I checked for sirkay 587c and 587d and for fly-kernel 0.8 as the latest of their breed, quite sure nobody has ever cared about this quirk.
The duplicate execution could be tracked back to the init.rc entries:
Code:
[COLOR=SeaGreen]#DooMLoRD: init.d scripts support
start sysinitsupport
class_start core
class_start main
#DooMLoRD: new init.d scripts support
service sysinitsupport /sbin/sysinitsupport.sh
class main
disabled
oneshot[/COLOR]
Which does:
Code:
#!/sbin/sh
# DooMLoRD: init.d support script (v1)
# [START] setting up
echo "[START] remounting system" > /data/local/tmp/sysinitsupportlog.txt
/sbin/busybox mount -o remount,rw /system >> /data/local/tmp/sysinitsupportlog.txt
# make init.d directory
echo "
[*] make init.d directory" >> /data/local/tmp/sysinitsupportlog.txt
/sbin/busybox mkdir -p /system/etc/init.d >> /data/local/tmp/sysinitsupportlog.txt
# correcting permissions of files in init.d directory
echo "
[*] correcting permissions of files in init.d directory" >> /data/local/tmp/sysinitsupportlog.txt
/sbin/busybox chmod 777 /system/etc/init.d/*
# [COLOR=DarkOrchid]make [/COLOR]init.d directory
echo "
[*] [COLOR=DarkOrchid]make [/COLOR]init.d directory" >> /data/local/tmp/sysinitsupportlog.txt
[COLOR=Red]/system/bin/logwrapper /sbin/busybox run-parts /system/etc/init.d[/COLOR]
# [DONE] all done exiting
echo "[DONE] all done exiting" >> /data/local/tmp/sysinitsupportlog.txt
And later also in init.rc:
Code:
[COLOR=SeaGreen]#DooMLoRD: run my mods
service mymods /sbin/execute_mods.sh
class main
oneshot[/COLOR]
Which then does:
Code:
#!/sbin/sh
# starting
echo "[ START ]" > /data/local/tmp/log_doom-mods.log
echo "" >> /data/local/tmp/log_doom-mods.log
[COLOR=Red]# execute tweaks
/system/bin/logwrapper /sbin/busybox run-parts /etc/init.d[/COLOR]
# execute FPS limit remove
/sbin/mount -t debugfs debugfs /sys/kernel/debug
/sbin/echo '0' > /sys/kernel/debug/msm_fb/0/vsync_enable
/sbin/umount /sys/kernel/debug
echo "FPS limit successfully removed " >> /data/local/tmp/log_doom-mods.log
# DONE
echo "" >> /data/local/tmp/log_doom-mods.log
echo "[ DONE ]" >> /data/local/tmp/log_doom-mods.log
You see that the execution of
Code:
[COLOR=Red]/system/bin/logwrapper /sbin/busybox run-parts /etc/init.d[/COLOR]
is actually done twice - from sysinitsupport.sh first and then again from execute_mods.sh
Also mind that the log-entry in the first is leading in the wrong direction (copy error from above). It should better read "execute init.d directory"
The related logs are found in \data\local\tmp.
You can check yourself with this little script in \etc\init.d:
Code:
#!/system/bin/sh
echo [] > /data/local/tmp/$PPID-exec-done
echo $PPID "init.d executed" >> /data/local/tmp/$PPID-exec-done
date >> /data/local/tmp/$PPID-exec-done
Mind the $PPID which is the parent PID of the executing command (the busybox "run-parts"). Per boot you should just get 1 file <PPID>-exec-done containing the timestamp if you get 2 then you know why...
I have attached the script wrapped in a zip file. Unpack it, copy to \etc\init.d (or if not sym-linked to \system\etc\init.d) and change attributes to "777". Reboot and look what you get in \data\local\tmp.
Once you know, remove the script again and delete the created files in \data\local\tmp.
Mind that the scripts referenced from init.rc are copied over again from the kernel part so any change of the scripts in the \system\sbin folder is useless. The kernel has to fix that, no way out here.

---

I've been looking into this issue as well with the help of dk_zero_cool (mounts2sd amongst other things) as no matter what I tried I could not get m2sd to run in anything other than Safe Mode because init.d was being run as a service and not being executed in full before the init continued. So far I have edited init.rc to remove the second instance you quote above, and edited the first instance to:
Code:
exec /sbin/sysinitsupport.sh
so it executes rather than running a service. I then edited sysinitsupport.sh to contain just this:
Code:
#!/system/bin/sh
export PATH=/sbin:/system/sbin:/system/bin:/system/xbin
/system/bin/logwrapper busybox run-parts /system/etc/init.d
and also removed the line in execute_mods.sh that relates to init.d, so now theoretically the only thing that should happen in relation to init.d is that the exec command runs sysinitsupport.sh which in turn runs the init.d scripts before anything else in init.rc happens. IN THEORY!!! Because even with all of that done (and I have searched through the whole ramdisk, plus looked at the git for the init binary used in the Lupus ICS kernel I am using to make sure I haven't missed anything) the m2sd init.d script is still running in Safe Mode because it is detecting /system/bin/servicemanager running at the point it tries to execute. So something somewhere is still starting before the execution of init.d. I even tried with the init binary from the latest CM9 build for the Ray to rule out something wrong in there too.
I have sent all my files to dk_zero_cool to see if he can shed any further light on what else may be wrong. I have checked four different Ray kernels and they all use the same DoomLord methods so I doubt whether there are any Ray kernels that are running init.d correctly. It would be great to find a fix for this. Hope more people chip in!
Thanks for opening the discussion

---

As I understood, you have made some changes in the kernel assembly (not the code) to circumvent the effects you have outlined. I admit that I have not fully understood YOUR concern - but for my double execution of the "run-parts" the deactivation of the relevant line in either of the 2 scripts should do it, or not?
Is your concern related to the situation that init.d cannot do "everything" at the time it is executing and so it cannot achieve what some scripts intend to?
I am far too little educated in the details of kernel execution privileges so cannot further comment on that

---

Yeah, pretty much - to avoid possible issues with the m2sd script moving stuff around while something else is trying to make use of it the first thing it does is check if servicemanager is running, and if it is it disables the ability to move things like /data and dalvik-cache to sd-ext. The changes that we made in the scripts SHOULD have changed the init.d implementation from it running as a service whilst the rest of the init process carried on, to being executed as a command allowing any init.d scripts to be executed prior to any other service being started - as I understand it this is how init.d was intended to be used (ie the user scripts in /etc/init.d are run fully before anything else). However as I said, even with these changes and everything else relating to init.d having been removed something is still starting servicemanager, and until the source of that can be isolated scripts like m2sd cannot run fully/safely.

---

I guess the strategy to check on servicemanager is not right here. This is a process that starts several services and should not depend on anything in the init.d. So if you say that the boot sequence would have init.d completed BEFORE any service is started via servicemanager - THEN this could be a flaw in the kernel.
However is that really true? Is there no option to check if certain "dangerous" (for your move purposes) services are active already instead of checking on the servicemanager? I had found a nice overview on the Linux boot process here and I think that somewhere as part of the various init.x excuted scripts the servicemanager simply MUST be started - init.d is just a part of init - and for sure not the first part of it.
Off Topic:
I wonder where I can get some more insight in the Xperia Kernels and how they are assembled - especially which trace of that is noticeable in the filesystem. I noticed that the ICS Kernel have roughly 340MB for the user space while the JB Kernel has 360MB. Device should have 512 MB RAM, then some MB go away for radio and possibly video buffer, but his should be the same for all (accross ICS/JB). So what kind of memory tweaking makes this 20MB difference? I have not found a good place to discuss this - where could I go to?

---

tobbbie said:
It seems that all Xperia Kernels that stem back to the DoomLord kernel start the init.d execution twice. While you may think "better twice than never" the nearly parallel execution of scripts can create problems if they concurrently manipulate CPU related tables - or fail to do so due to security mechanisms built in. I was hunting the problem that cpu-clock manipulation from init.d did not work for the scripts generated by System Tuner - finally resulting in this finding.
Click to expand...
Click to collapse
How about Radeon kernel? Is there an issue like you said?
Sent from my ST18i using Tapatalk 2

---

frogerra said:
How about Radeon kernel? Is there an issue like you said?
Sent from my ST18i using Tapatalk 2
Click to expand...
Click to collapse
Just try out (see the guide on 01test.zip) - nothing you can harm doing that.

---

kernel behaviour trick
I noticed some kernels (GB and ICS) do overwrite settings from /etc/init.d ....
So if you make an init.d script with specific cpu values, or VM settings...some kernel overwrite these with their compiled/preferred values "after" init.d scripts are executed. This way it looks as if your script is not completely accepted by the rom...but the truth is that the kernel applies its own preferred values afterwards.
Thats why my init.d script contains a wait of 60 seconds in the beginning of the scripts......how ? add:
sleep 60
So your script looks like this (example):
#!/system/bin/sh
# this init.d script is for when you apply doomlord kernel supplied with repack 2013
sleep 60
That will run your init.d goodies after one minute.
As you see I'm working on a revived repack4pda 2013 (GB) , which will be released soon in repack4pda thread.
Br.
Michel

---

I guess that these kernels may be doing the same what you are proposing - just sleep the shell process before action starts. So you need to lookup the call tree from the init process along the various init.*.rc scripts if this is the case. For duplicate execution of init.d content any timed scripts willl not help - the duplicate execution will just happen later as the same script will pause the same amount of time.
Not sure if the init process script execution is synchronous or not - so if you create scripts which sleep, the final signal for "boot completed" may just also delay and the whole boot process may take longer by that sleep time. As well would you just stack the delays after each other and so nothing is gained finally. Synchronous execution would make it impossible that a part of the init process could postpone beyond boot completed.
It may differ if you run the scripts via the "exec" command or let it execute via the servicemanager. I guess the latter may run them asynchronously - not sure here as well.