FIXME

[Using QT on some OSs -- allocate N kernel threads, suspend,
then use their stacks.]

To: "Scott W. Bradley" <ScottWB@BeComm.com>
cc: Guy in Au <guy@clearwater.com.au>
From: pardo@pardo.net
Subject: Re: QuickThreads 
In-reply-to: Your message of "Wed, 10 Apr 2002 12:55:58 PDT."
             <A9DFD9380C01D4118E740050DA7A897056B045@becomm-mis2.becomm.com> 
Date: Thu, 11 Apr 2002 00:48:34 +0000
Sender: pardo

>[Allocate and suspend a theread, steal it's stack.]

Now that's using your noodle!  Cool!  Thanks for the note and I'll try
to keep it in mind.

	;-D on  ( Keep Hit Inn Minde )  Pardo




Date: Sun, 8 Dec 2002 22:06:08 +0100
Subject: incorrect pointer operation qt v2?
Content-Type: text/plain; charset=US-ASCII; format=flowed
Mime-Version: 1.0 (Apple Message framework v548)
From: Marco Bucci <marco.bucci@inwind.it>
To: qt-0001@pardo.net
Content-Transfer-Encoding: 7bit
In-Reply-To: <3D2AEE270290E82F@ims9a.libero.it> (added by postmaster@libero.it)
Message-Id: <DF17DE9C-0AF0-11D7-9FF6-0030658DC216@inwind.it>
X-Mailer: Apple Mail (2.548)

Dear David,

I'm debugging the linuxPPC version (yes, in the end I toke it as 
challenge and I'm doing both the ppc versions). I hope to finish today. 
After that I'm so late in my job that I will not have more time... :-/

May be I found that bug that affects  last qt version. Really I already 
found it, but I didn't wand disturb you because I believed that the 
condition in which the bug could "arise" should never happen.

ANYWAY IN BRIEF:


In meas.c, t_alloc allocates the memory block for the thread:

   t->stk = malloc (ssz);

then it rounds the pointer:

   t->stk = (void *)ROUND (((iaddr_t)t->stk), QT_STKALIGN);

Since t->stk is used by t_free to dispose the thread memory block, I 
believe this is incorrect and in facts, sometimes I get a "segmentation 
fault" during the t_free execution. Actually the same fault can be 
forced just setting:

t->stk = t->stk + QT_STKALIGN;

and not because the stack is not aligned, but because the pointer is 
not anymore a legal pointer of a malloc allocated block.


I don't know if this is the problem of qt version 2, anyway in version, 
1 that rounding was not performed (I checked to to have confirm, but I 
was sure of it). This is the typical "random" bug that make people 
crazy :-)
What is a bit strange is that pointers returned by malloc should be 
page aligned (but I'm not sure of it). Anyway, the compiler is free to 
handle the memory allocation just respecting the minimal alignment 
rules (4 bytes usually). So, we are not guarantee that t->stk will be 
aligned respecting the stack rules.

Ok, now I'm running to finish and I cannot try solutions, but, the most 
trivial solution is to save the original malloc pointer to use it with 
the free function.

Ok, now I run. I want finish.


Sincerely


Marco