I'm currently writing my own operating system. It is a non-preemptive OS and my threads are working fine with the scheduler. I have one edge case though. If I don't call my yield() function from C but do asm("call yield") the calculations later on fails but only with -O2 optimization.
Like many have said, it is always our fault
I have tried all approaches i can think of but now I'm desperate.
So if anybody has some tips on what might be happening or what i should investigate please share.
I guess this gets shutdown for being offtopic but any tips is greatly appreciated.
When the compiler generates a call to a function, it preserves the contents of any registers which may be modified by the called function ("caller-save registers") before making the call.
Since you've buried this function call within an inline assembler block, the compiler doesn't know it needs to save and restore registers around the call.
Simple solution: Don't do that, then. If you want to call a function, use C function call syntax.
Complicated solution: Declare which registers this function call will clobber using extended inline assembler syntax.
Thanks for all the comments. It helped allot. I finally figured it out. My context switch did not handle the state of the FPU. After going back to dissable once again I observed that the normal generated C code helped me save the FPU state.
After implementing FSAVE/FNSAVE and FRSTOR in the context switch the -O2 optimized code also worked as expected.
#melpomene: As you asked:
Have you looked at the generated code?
In retrospect, obviously not hard enough.
Thanks for all suggestions, and I hope this thread can help others to remember the x87 FPU instructions now :-P
Related
I am new to LLVM backend. What I am trying to do is inserting several register loading and storing between different instructions. So I create a machine function pass to do this job. When I disable the optimisation via -O0, everything works fine.
However, when I enable the optimisation, I found that the code has been optimised in a wrong way. For example, some labels of -O0 code has been kept.
For example it still keeps jl label1; while label1 is not exist in -O3 code.
I am trying to figure out a way to bypass it. My thought is that maybe we can decide the execution order of Machine Function Pass and I could then run my Pass at the very end. Or maybe there is other ways bypass it?
I have been searching it for a while, but I didn't find anything useful.
Thanks for your kind helps !
A while ago during documenting legacy code I found out there is a tool for displaying call graph (call stack) of any standard program. Absurdly I wasn't aware of this tool for years :D
It gives fancy list/hierarchy of calls of the program, though it is not a call graph in a full sense, it is very helpful in some cases.
The problem is this tool is linked only to SE93 so it can be used only for transactions.
I tried to search but didn't find any similar tool for reports or function modules. Yes, I can create a tcode for report, but for function module this approach doesn't work.
If I put FM call inside report and build a graph using this tool, it wraps this call as a single unit and does not analyze deeper. And that's it
Anybody knows a workaround how we can build graph for smth besides transaction?
The cynic in me thinks RS_CALL_HIERARCHY was left to rot. Sandra is right, it definitely used to work. Once OO came to abap, interfaces and dynamic/generic code became possible. So a call heirarchy based on static code analysis was pushing proverbial up hill.
IMO the best way to solve this is a FULL trace and then to extract the data from the trace.
There are even external tool that do that.
This is of course, still limited as running a trace on every execution path can be very time consuming. Did I hear someone say small Classes please ?
Trans SAT.
Make sure teh profile you use isnt aggregating, and measure the blocks you are interested.
Now wade you way through the trace.
https://help.sap.com/doc/saphelp_ewm93/9.3/en-US/4e/c3e66b6e391014adc9fffe4e204223/content.htm?no_cache=true
Have fun :)
The call hierarchy displays also works for programs and function modules.
In my S/4HANA system, for VA01, it displays:
Clicking the hierarchy of function module CJWI_INIT displays:
I get exactly the same result by calling the function module RS_CALL_HIERARCHY this way:
The parameter OBJECT_TYPE may have these values:
P : program
FF : function module
The "call graph" is not maintained anymore since at least Basis 4.6, and it doesn't work for classes and methods.
But the tool is buggy: in some cases, a function module containing a PERFORM at the first line, it may not be displayed, whatever the call graph is launched from SE93 or directly from RS_CALL_HIERARCHY.
I'm trying to write a pam backdoor scanner, which may call fopen function in pam_sm_authenticate(normal file will not call fopen in this function) to store username and password, but I can't use external command such as "nm, readelf" or something like that, so the only way seems to scan pam_sm_authenticate function and find all call instructions and caculate the address to check if it is calling fopen, but it is too troublesome and i'm not very familiar with ELF file(I even dont know how to find offset of pam_sm_authenticate, I'm useing dlopen and dlsym to get the address..), so I wonder if there is a better or easy way to detect it? Thankyou.
TL;DR: building a robust "pam backdoor scanner" is theoretically impossible, so you should give up now and think about other ways to solve your problem.
Your question is very confusing, but I think what you are asking is: "can I determine programmatically whether pam_sm_authenticate calls fopen".
That is the wrong question to ask, for several reasons:
if pam_sm_authenticate calls foo, and foo calls fopen, then you still have a problem, so you really should be scanning pam_sm_authenticate and every function it calls (recursively),
the fopen is far from the only way to write files: you could also use open, or system (as in system("echo $secret > /tmp/backdoor"), or direct sys_open syscall, or a multitude of other hacks.
finally, the pam_sm_authenticate can use just-in-time compilation techniques to build arbitrary code (including code calling fopen) at runtime, and answering whether it does by examining its code is equivalent to solving the halting problem (i.e. impossible).
First, thank you for taking pity on me and reading this issue. I CANNOT for the life of me figure out what extension I might have installed that is causing this issue, but it is EXTREMELY cumbersome.
Whenever I begin to type code (VB I think it also occurs in C#), for example "For Each" once I hit the F it forces a set of parentheses. Which would look like F(), but because I keep typing it looks like F(or). This only occurs when coding inside code blocks like a function or a sub, but when I'm creating the function it does not occur. I've disabled any and all power tools and the like, or at least I'm 90% sure I've done this for all of them, and yet it still occurs.
I'm usually pretty proficient at digging about the net and finding the answer, but for this one I'm at a loss. There is just too many keywords involved, so all I see is non-related topics, or how to make the parentheses occur, not get rid of them.
If anyone can provide some steps to resolve this, I'm happy and eager to try them. It's just such a hassle to live with for right now.
If you think it is a Visual Studio extension, then start by disabling all of them and adding them back one at a time.
You can also run VS with the command line switched to disable features.
Devenv switches
The simple answer to the cause is the Codealike VS Extension. I logged a bug with them and hopefully they'll fix it soon
I have written a fortran routine that uses some legacy fortran 77 code for finite elements. However, with a particular mesh, when the -O optimization flag is turned on, an important do-loop iterator is somehow being modified, even though fortran supposedly prohibits this. I have compiled this code using gfortran4.5 with the -fcheck=do run-time checking enabled and it has verifies what I've noted above. A runtime error occurs, only when optimizations are turned on and points directly to the do-iterator.
Using gdb on optimized code seems (while it seems erratic - lines bouncing back and forth) seems to clearly indicate that the do-iterator somehow gets set back to zero, and essentially this causes a nice infinite loop.
Any suggestions as to how to hunt down and fix whatever is causing this bug would be greatly appreciated, as I'd like to make sure the whole project can be consistently compiled with the same flags.
You say that you use fcheck=do; why not go all the way and use fcheck=all? What you're seeing sounds like a typical case of memory corruption due to an array bounds violation, which fcheck=all can in some cases catch. Where the array bounds checking doesn't work that well is with implicit interfaces and incorrect bounds being passed; a solution here is to put your procedures into modules, allowing the compiler to check interfaces.
And, like Jonathan Dursi said, consider using a tool like valgrind.