stack pointer values using uxw - show strange behaviour

[Adarsh Thampan]

Hi ,

I am trying to print the stack pointer values in each frame of my stack. I am using the attached code , basically i step through the frame one by one and print the SP and IP . Although the IP is printed correctly , I get the same SP value for all frames.

I have attached the source and the output.

Thanks,
Adarsh

[Dennis Handly]

>I get the same SP value for all frames.

There probably isn't anything wrong with that. You need to print out tuples:
PC, SP and BSP

In your case BSP (the RSE stack pointer) should change.

[Adarsh Thampan]

In HPPARISC , we get the current frame SP and previous SP using U_get_previous_frame API .

In Itanium also we need the current frame SP and previous frame SP.

You are syaing that the SP value is same fro all frames in Itanium .

Can you explain why this difference >

[Dennis Handly]

>In Itanium also we need the current frame SP and previous frame SP.

On Integrity you have contexts not frames.
Conceptually you have the tuple PC, SP and BSP as an unique identifier for each context. And you use uwx_step(3X) to find the previous context.

>You are saying that the SP value is same for all frames in Integrity.

No, I'm saying your trivial test case doesn't need user stack frames for some of the functions. It only needs RSE stack frames for those different contexts. That's why you need both SP and BSP.

[Suprateeka R Hegde]

unwind library on IPF is different. you have to use uwx_step(3X) to get to the previous frame. are you using this?

[Suprateeka R Hegde]

sorry, missed to look at your code snippet.

[Adarsh Thampan]

Dennis,

My code does an exception tracing , and as a part of the tracing I need to print the SP and framesize of all the frames within the current stack( one that caused the exception).

Now for AIX,Linux we just print the SP and calculate the frame size ( current SP - previous SP) .

I tried to do the same in HP Itanium , but since the SP values are all the same it seems my logic is wrong, BTW I have note read about the Itanium architecture.

Now it looks you are suggesting to use a combination of the RSE as well ( I assume it is the register stack) and the memory stack.

So how do I know when to use RSE with BSP and when to use memory stack with SP.

Thanks,
Adarsh.

[Suprateeka R Hegde]

the logic of current SP - previous SP to calculate the frame size may not hold good always. for example there may be non-contiguous pieces.

you have to read the itanium runtime architecture guide which provides a wealth of information. get it here: http://www.intel.com/design/itanium/downloads/245358.htm

it clearly tells when the memory stack is used and when RSE is used, etc.

[Dennis Handly]

>My code does an exception tracing

You mean stack tracing on a signal or some some software condition?

You do know about U_STACK_TRACE(3) and _UNW_STACK_TRACE(3)?

>as a part of the tracing I need to print the SP and framesize of all the frames within the current stack (one that caused the exception).

You must print the SP and BSP of all contexts.

>but since the SP values are all the same it seems my logic is wrong

There is nothing wrong with the logic. The user framesize IS 0. You could print the RSE framesize.

>Now it looks you are suggesting to use a combination of the RSE as well

Yes. You should print both.

>how do I know when to use RSE with BSP and when to use memory stack with SP.

You always print SP and BSP and if you want, print the memory stack framesize and the RSE stack framesize.

[Adarsh Thampan]

Hi ,

Is there direct way to know if the frame resides on teh register stack or memory stack ?

Thanks,
Adarsh.

[Dennis Handly · Accepted solution]

>Is there direct way to know if the frame resides on the register stack or memory stack?

You are asking the wrong question.
Every function (context) has both a memory stack and a register stack. With the memory stack possibly being of size 0.

[Adarsh Thampan]

So how am I going to find my framesize and SP value for each frame in the stack for HP itanium.

In Linux and Solaris I just had a memory stack to worry about and I could just has to use the SP and framesize value.

I am sure there is a way since the gdb on Itanoium does display the backtrace , along with the stackpointer values for each frame in the stack . I just need to do the same.

[Suprateeka R Hegde]

as dennis suggested above, the test case you have, cannot show the difference in stack pointer. even gdb does not show that. what gdb shows in 'bt' command is the IP. if you want to see the difference in SP, add alloca(3C) to each of your calls or have more than 8 parameters being passed to the function calls, etc. you can see the difference.

difference in stack pointer is seen when the memory stack size is non zero. or else you can see only BSP changing.

looks like gdb uses other mechanisms to get the frame size. it is not a simple arithmetic based on register values between contexts.

[Dennis Handly]

>how am I going to find my framesize and SP value for each frame?

Your program already knows how to find the SP value. In your other thread I told you how to compute the framesize:
http://forums.itrc.hp.com/service/forums/questionanswer.do?threadId=1407123

>along with the SP values for each frame in the stack.

Yes, "info frame" will do that for the current frame.

>Suprateeka: cannot show the difference in stack pointer.

Sure you can. The difference is 0.

>add alloca(3C) to each of your calls or have more than 8 parameters being passed to the function calls, etc.

Why work that hard? If you want a non-zero sized frame, have some local variables, that you use:
char buf[1024 * 1024];

>looks like gdb uses other mechanisms to get the frame size. it is not a simple arithmetic based on register values between contexts.

Sure it is that simple. The "other" method is only needed if you want to distinguish between the fixed size frame and an any alloca(3) parts.

[Adarsh Thampan]

Hi ,

I have attached teh sample program that prints out framesize usng both SP and BSP for memory stack and register stack coresspondingly.

I have observed the following:-

1> Using SP the framesize is mostly 0, even if I use some local variables.

2> Using BSP I seem to get some correct values of framesize.


Can I always use BSP to get the framesize for any function in my code. In case not , please explain why ?


I am unclear about the fact that under what circumstances does a function use

1. only register stack
2. only memory stack ( I am not sure whether this is possible in Itanium).
3. Both register and memory stack.


P.S - My program need is to use 32 bit binaries only.

[Suprateeka R Hegde]

> 1> Using SP the framesize is mostly 0, even if I use some local variables.

you are just declaring the buffers. you have to use it in your code or the compiler may optimize it away.

> I am unclear about the fact that under what circumstances does a function use
1. only register stack
2. only memory stack ( I am not sure whether this is possible in Itanium).
3. Both register and memory stack.

as far as i know, its always both register and memory stack. and for trivial functions, the memory stack is of zero size.

[Suprateeka R Hegde]

> Can I always use BSP to get the framesize for any function in my code. In case not , please explain why ?

no. it is because a function may also have memory stack size which is not reflected by bsp values. you have to consider both sp and bsp.

[Dennis Handly]

>that prints out framesize usng both SP and BSP
>framesize=psp-sp;

This should be reversed.

>%d for function %s \n",

You should not print a space before the newline.

>if I use some local variables.

As I cautioned you, you have to really use them, not just define them.

>2> Using BSP I seem to get some correct values of framesize.

You wouldn't know if this was correct until you looked at the machine code.

>Can I always use BSP to get the framesize for any function in my code.

Of course not. There are two stack "frames" and you must have two sizes.

>I am unclear about the fact that under what circumstances does a function use
>1. only register stack
>2. only memory stack
>3. Both register and memory stack.

You should assume it always uses both. As I said, they can have zero size.

>Suprateeka: you have to consider both sp and bsp.

Right. There is no need to special case a zero size. Zero is a valid size.

[Dennis Handly]

Attached is your original test case modified to print framesize and other variables on the same line.