[Development] Fight with SWAP module. - XPERIA X8 Android Development

I will be off by next two weeks, so I want to show where I'm and also I have hope that someone with fresh look can point me what or where is the problem.
I'm fighting with module since for two weeks so probably I don't see something obvious apart from this is impossible to do .
I know that messing in so fragile thing like memory management is highly risky.
What I have now:
When I hijack some of the methods (hijacking shmem methods works fine) - I got kernel panic :
Code:
Kernel panic - not syncing: Aiee, killing interrupt handler!
this occurs here:
Code:
if (unlikely(in_interrupt()))
panic("Aiee, killing interrupt handler!");
and in_interrupt() is defined:
Code:
#define in_interrupt() (irq_count())
#define irq_count() (preempt_count() & (HARDIRQ_MASK | SOFTIRQ_MASK))
So I figured out that preempt_count value is decreased too many times, and I compiled two version of module one with swap enabled and one without and with full PREEMPT debug. Both disassembed versions are in attachment.
I count increasing and decreasing preempt_count value in methods and there are results:
Code:
<__remove_mapping>: add:1 sub:3 ---- add:1 sub:2 DC DSC
<do_wp_page>: add:4 sub:5 ---- add:-1 sub:-1 DC DAC DSC
<exit_mmap>: add:0 sub:1 ---- add:1 sub:1 DC DAC
<handle_mm_fault>: add:6 sub:6 ---- add:8 sub:9 DC DAC DSC
<shmem_getpage>: add:7 sub:33 ---- add:5 sub:19 DC DAC DSC
<shmem_writepage>: add:1 sub:4 ---- add:0 sub:0 DC DAC DSC
<shrink_zone>: add:2 sub:2 ---- add:0 sub:0 DAC DSC
<try_to_unmap_one>: add:1 sub:2 ---- add:0 sub:1 DC DAC DSC
<unmap_region>: add:0 sub:1 ---- add:1 sub:1 DC DAC
<unmap_vmas>: add:1 sub:2 ---- add:2 sub:2 DC DAC
<zap_page_range>: add:0 sub:1 ---- add:1 sub:1 DC DAC
add:141 sub:207 ---- add:146 sub:192
Last line shows overall count. First two number are from my module, second two number are from module without swap enabled. -1 means that method was not in module and is part of other method.
Without deep looking there are differences - but every method should be checked comparing source and both compiled versions.
Once again - if someone wants to look - be my guest . Sources are in GIT.

---

Andyx your back..new devz again.. thumbs up!
btw what is this SWAP module do?

---

Its for adding more RAM
Sent from my X8 using XDA Premium App

---

as-mario said:
Its for adding more RAM
Sent from my X8 using XDA Premium App
Click to expand...
Click to collapse
To add on, basically it is a memory on the sd card that can store memory that is supposed to be on RAM. This can somewhat increase RAM as this can store memory that goes on RAM but swap memory has its disadvantages because it is slow memory and causes the sd card to wear out faster. It is virtual RAM in simple terms.

---

AnDyX said:
I will be off by next two weeks, so I want to show where I'm and also I have hope that someone with fresh look can point me what or where is the problem.
I'm fighting with module since for two weeks so probably I don't see something obvious apart from this is impossible to do .
I know that messing in so fragile thing like memory management is highly risky.
What I have now:
When I hijack some of the methods (hijacking shmem methods works fine) - I got kernel panic :
Code:
Kernel panic - not syncing: Aiee, killing interrupt handler!
this occurs here:
Code:
if (unlikely(in_interrupt()))
panic("Aiee, killing interrupt handler!");
and in_interrupt() is defined:
Code:
#define in_interrupt() (irq_count())
#define irq_count() (preempt_count() & (HARDIRQ_MASK | SOFTIRQ_MASK))
So I figured out that preempt_count value is decreased too many times, and I compiled two version of module one with swap enabled and one without and with full PREEMPT debug. Both disassembed versions are in attachment.
I count increasing and decreasing preempt_count value in methods and there are results:
Code:
<__remove_mapping>: add:1 sub:3 ---- add:1 sub:2 DC DSC
<do_wp_page>: add:4 sub:5 ---- add:-1 sub:-1 DC DAC DSC
<exit_mmap>: add:0 sub:1 ---- add:1 sub:1 DC DAC
<handle_mm_fault>: add:6 sub:6 ---- add:8 sub:9 DC DAC DSC
<shmem_getpage>: add:7 sub:33 ---- add:5 sub:19 DC DAC DSC
<shmem_writepage>: add:1 sub:4 ---- add:0 sub:0 DC DAC DSC
<shrink_zone>: add:2 sub:2 ---- add:0 sub:0 DAC DSC
<try_to_unmap_one>: add:1 sub:2 ---- add:0 sub:1 DC DAC DSC
<unmap_region>: add:0 sub:1 ---- add:1 sub:1 DC DAC
<unmap_vmas>: add:1 sub:2 ---- add:2 sub:2 DC DAC
<zap_page_range>: add:0 sub:1 ---- add:1 sub:1 DC DAC
add:141 sub:207 ---- add:146 sub:192
Last line shows overall count. First two number are from my module, second two number are from module without swap enabled. -1 means that method was not in module and is part of other method.
Without deep looking there are differences - but every method should be checked comparing source and both compiled versions.
Once again - if someone wants to look - be my guest . Sources are in GIT.
Click to expand...
Click to collapse
Btw devs, it is possible to allocate flash memory (e.g. on /cache) to RAM? It may be a stupid question but just wanna know.

---

Oh its like USB PenDrive device as Ram on PC. Sugoi..
Thanks..

---

i think that kernel we are using does not support swap,
i've tried some application for making swap file, but i was failed..
i forgot the application name, it creates like page file ,custom sized..
but affected nothing, just created a big junk file in my SD card..
i think that we better wait till our developers succeed bypassing bootloader,
with new custom kernel, may be we can enable swap and anything else linux can do

---

alphine.zzz said:
i think that kernel we are using does not support swap,
i've tried some application for making swap file, but i was failed..
i forgot the application name, it creates like page file ,custom sized..
but affected nothing, just created a big junk file in my SD card..
i think that we better wait till our developers succeed bypassing bootloader,
with new custom kernel, may be we can enable swap and anything else linux can do
Click to expand...
Click to collapse
He's trying to make a module for our kernel... that can enable SWAP
Our kernel doesn't support a lot of stuff... that's why we have modules

---

I will be off by next two weeks, so I want to show where I'm and also I have hope that someone with fresh look can point me what or where is the problem.
I'm fighting with module since for two weeks so probably I don't see something obvious apart from this is impossible to do .
I know that messing in so fragile thing like memory management is highly risky.
What I have now:
When I hijack some of the methods (hijacking shmem methods works fine) - I got kernel panic :
Kernel panic - not syncing: Aiee, killing interrupt handler!
this occurs here:
if (unlikely(in_interrupt()))
panic("Aiee, killing interrupt handler!");
and in_interrupt() is defined:
#define in_interrupt() (irq_count())
#define irq_count() (preempt_count() & (HARDIRQ_MASK | SOFTIRQ_MASK))
So I figured out that preempt_count value is decreased too many times, and I compiled two version of module one with swap enabled and one without and with full PREEMPT debug. Both disassembed versions are in attachment.
I count increasing and decreasing preempt_count value in methods and there are results:
<__remove_mapping>: add:1 sub:3 ---- add:1 sub:2 DC DSC
<do_wp_page>: add:4 sub:5 ---- add:-1 sub:-1 DC DAC DSC
<exit_mmap>: add:0 sub:1 ---- add:1 sub:1 DC DAC
<handle_mm_fault>: add:6 sub:6 ---- add:8 sub:9 DC DAC DSC
<shmem_getpage>: add:7 sub:33 ---- add:5 sub:19 DC DAC DSC
<shmem_writepage>: add:1 sub:4 ---- add:0 sub:0 DC DAC DSC
<shrink_zone>: add:2 sub:2 ---- add:0 sub:0 DAC DSC
<try_to_unmap_one>: add:1 sub:2 ---- add:0 sub:1 DC DAC DSC
<unmap_region>: add:0 sub:1 ---- add:1 sub:1 DC DAC
<unmap_vmas>: add:1 sub:2 ---- add:2 sub:2 DC DAC
<zap_page_range>: add:0 sub:1 ---- add:1 sub:1 DC DAC
add:141 sub:207 ---- add:146 sub:192
Last line shows overall count. First two number are from my module, second two number are from module without swap enabled. -1 means that method was not in module and is part of other method.
Without deep looking there are differences - but every method should be checked comparing source and both compiled versions.
Once again - if someone wants to look - be my guest . Sources are in GIT.
Click to expand...
Click to collapse
AndyX ...u mean u success hijacking our x8 and now it has swap modules? ?;D..
Or still on development? ?..
Furthermore ...is a class 2 memory card recommended for swap modules? ?><…
Sent from my X8 using Tapatalk

---

lukewong01 said:
AndyX ...u mean u success hijacking our x8 and now it has swap modules? ?;D..
Or still on development? ?..
Furthermore ...is a class 2 memory card recommended for swap modules? ?><…
Sent from my X8 using Tapatalk
Click to expand...
Click to collapse
U don't have to qoute whole ****ng post

---

spyder12345 said:
To add on, basically it is a memory on the sd card that can store memory that is supposed to be on RAM. This can somewhat increase RAM as this can store memory that goes on RAM but swap memory has its disadvantages because it is slow memory and causes the sd card to wear out faster. It is virtual RAM in simple terms.
Click to expand...
Click to collapse
This looks promising as RAM is the main problem with our phone.

---

lukewong01 said:
AndyX ...u mean u success hijacking our x8 and now it has swap modules? ?;D..
Or still on development? ?..
Furthermore ...is a class 2 memory card recommended for swap modules? ?><…
Sent from my X8 using Tapatalk
Click to expand...
Click to collapse
I don´t think class 2 will give you good performance, it´s gonna be very slow

---

biscoitu said:
I don´t think class 2 will give you good performance, it´s gonna be very slow
Click to expand...
Click to collapse
how about class 4 ?

---

Even with hard disk, it's slow.

---

Dare-Devil Inside said:
U don't have to qoute whole ****ng post
Click to expand...
Click to collapse
Don't f*cking swear
@doixanh, That is quite a bad news, even using hdd as memory card(impossible)
Still slow at SWAP

---

How about compcache? Maybe that will help. Just a question .

---

Btw..i just wanna swap those RAM on those background services and widgets....so about 20MB of swap...i dun tink if use on widgets and background will slow.....right?...

---

Swap was never intended to for speed up any machine . It's only for having more memory for price of slower performance when swap memory is used. Windows users should know this best

---

I think swap better usage is for storing not active apps so you can have more space for active apps that are using much ram but still having great multitasking,right?

---

as-mario said:
I think swap better usage is for storing not active apps so you can have more space for active apps that are using much ram but still having great multitasking,right?
Click to expand...
Click to collapse
I doubt this will be the case. On the default linux kernel's memory management, yes, but the android kernel is not designed with swap in mind.