forums.ps2dev.org

[Cpasjuste]

Hi, i'm not sure how to 64b align an allocation made with "sceKernelAllocPartitionMemory".

I want to use the "sceKernelLoadModuleBuffer" function from a kernel mode module, so i need to extract and read the binary (data.psp) into user memory with sceKernelAllocPartitionMemory, then load it. Some help would be great.

Here is the non-working code i writed for now, crashing when the module is loaded :

main.c :

[J.F.]

If all you need is a 64 byte aligned buffer, allocate (size + 63), then use (ptr + 63) & ~63.

[Cpasjuste]

Thanks for the advice JF. Unfortunately i'm not able to allocate it correctly with sceKernelAllocPartitionMemory while i have no problem with malloc.

Here is the function i'm trying to use :

[pspZorba]

HI Cpasjuste,

could you try (change the first arg)

[Cpasjuste]

Hi pspZorba.

My main problem is that i'm in kernel mode, so i can't allocate 2048Kb of memory directly (it's why i'm using sceKernelAllocPartitionMemory to alloc the memory in user partition).

From my (poor) point of view, i think something like this should be better :

[pspZorba]

it's strange because the code I wrote (above) is an extraction of one of my prx in kernel mode.

so I was able to do allocate the 2048 (but you can set less than that), here is an extract of my prx, with Kmalloc, Kfree and Kmemset and a sample of use wihich works fine for me.

[J.F.]

[Jim]

need cast to uint in last line
_________________
http://www.dbfinteractive.com

[moonlight]

What gives error is probably the "&". Juse use some casts:

[pspZorba]

I was wondering, since the allocated address is in the user space, do you need to use the k1 stuff ?


[edit] if yes, that could be the reason of your crash Cpasjuste[/edit]
_________________
--pspZorba--
NO to K1.5 !

[Cpasjuste]

Thanks for all your help.

Unfortunatly, i'm still unable to use the sceKernelLoadModuleBuffer function.

Here is my final code which seems ok : http://rafb.net/p/nPBZl784.html

Here is my error :

[pspZorba]

Hi Cpasjuste,

1) I think your filename (char *) is coming from user space so you have to do the k1 stuff earlier in your code ( when I try your code, if I don't set k1 , I can't even do the first sceIoRead).


2) for me the code crash in the sceIoRead line 141 in the last loop. if you read less bytes the code doesn't crash (but I can't go further since I didn't read all datas).
so it looks like an overflow. It could come from a stack pb, actually sceIoRead certainly needs a certain amount of the stack (like printf).


ps: btw you should initialized the variable 'done' , because in the 5 first loops when you do sceKernelFreePartitionMemory it is not and you may have a problem. And you should test sceKernelAllocPartitionMemory as well rather than the buffer to check if the alocation was successful
_________________
--pspZorba--
NO to K1.5 !

[Cpasjuste]

Thanks again pspZorba :)

I'm not sure the problem is in sceIoRead, when i uncomment the sceKernelLoadFromBuffer function, the executable (data.psp) is always correctly loaded in memory, i verify it under psplink. I have dumped the memory where it is loaded and it's 100% the same as the DATA.PSP binary extracted from the eboot with pbpunpacker, and can be loaded under psplink.

The thing that seems strange for me is that with or without the 64b aligned buffer, the executable is always loaded at the same address (0x8804000), is it normal ?

[pspZorba]

well I don't know if 0x8804000 is a valid address, but it is a 64-memaligned address, so it is normal that 0x8804000 is the same as (((0x8804000)+ 63) & ~63).

either you are lucky or sceKernelAllocPartitionMemory gives always a memaligned allocation.

for me it really crashs in the secIoRead but that could be normal since I don't use the same prx loader nor the same makefile (don't use kubridge, just stick to pspsdk) as you.




ps: rather doing a loop to reach the part you read in the pbp, you should do something like :

[TyRaNiD]

The reason for the crash dump you posted is probably cause the load buffer function has its parameters wrong, for some strange reason sony played with order of those functions so sometimes it is buffer, length and sometimes length, buffer

Try prototyping it as :
SceUID sceKernelLoadModuleBuffer(SceSize bufsize, void *buf, int flags, SceKernelLMOption *option);

Instead and see if it fixed it ;)

[Cpasjuste]

Many thanks to all of you, i can now load and start an executable from a buffer in kernel mode. Tyranid was right, lenght and buffer was not in the good order :)

Now i still encounter another problem, with some executable loaded/started from a buffer, i get a 0x80020132 error (partition mismatch) while it's working great with other modules. Any idea ?

I will post the final code soon if someone come on the forum for this ...

pspZorba, i think you where right, buffer allocated with sceKernelAllocPartitionMemory seems to always be 64b aligned.

[pspZorba]

It is a pure supposition but I would bet that the ones that doesn't work are eboot(s) that run in kernel mode.
As you load them in the user partition they probably create this error.

You can easily check that.

If it is the case the solution is to check the eboot (I don't know how but it must be in the header ) and load it in the kernel or user partition...

A solution for the check is may be just this error code ! if you get it try to load the eboot in the kernel partition
_________________
--pspZorba--
NO to K1.5 !