4. Eigenes OS?

Eigenes OS?

  • ?

    PHPnerd schrieb:

    So, no kernelmode div by zero handler?

    of course, kernel must handle that, but must not terminate the application if an exception occurs inside guarded code. after that, the application should ask the kernel for exception code and (if there is one), retry the operation, notify the user, abort the program or something like that.
    🙂

    0
  • P

    All right, understand. But that is something for later on to me 😛

    // PHPnerd

    0
  • E
    Mod

    Yes, exception handling has to be fine-tuned in PrettyOS. Currently, there is only a message and a big stop sign. Page faults are already analyzed deeper.

    I worked on the VFS and the RAM disk, because it is very important to bring data, specialized kernel modules or user code into the memory of the OS. The way how it works now is as follows:

    1. produce a binary image from various files (with "make_initrd.exe", source code: "make_initrd.c"). I renamed the resulting image to "file_data.dat". This is the target for incbin in process.asm.

    2a) include it in process.asm with incbin, global addresses:

    ; data for ramdisk
global _file_data_start
global _file_data_end
_file_data_start:
incbin "file_data.dat"
_file_data_end:

    2b) Linker transfers it to the memory at &file_data_start

    1. The source code line
    k_memcpy((void*)ramdisk_start, &file_data_start, (ULONG)&file_data_end - (ULONG)&file_data_start);

    in ckernel.c transfers the data with the files to the RAM disk (this is the way we exchange data with PrettyOS at the time being)

    1. The Virtual File System (VFS) with individual file headers helps to access the files in the RAM disk

    Enjoy it! 🙂

    PrettyOS: http://www.henkessoft.de/OS_Dev/Downloads/36.zip
    Screenshot: http://www.henkessoft.de/OS_Dev/Downloads/ramdisk_test.png

    Currently the framework for booting, loading asm kernel, switch to PM and C-kernel, interrupts, write/read key queue, set system timer frequency (I use 100 Hz), paging, heap, virtual file system, RAM disk, multitasking, syscalls now basically work. 🙂

    0
  • A

    Erhard Henkes schrieb:

    Hier ein prinzipielles Beispiel mit Erzeugung mehrerer Tasks und Taskswitch:...

    Aha, jetzt wird's interessanter und praxisbezogener...

    0
  • E
    Mod

    Aha, jetzt wird's interessanter und praxisbezogener...

    Die letzten beiden Wochen habe ich mich - zumindest grundlegend - um die komplexen Themen Memory-, Task- und File-Management (VFS, RAM Disk) sowie Syscalls gekümmert, um zu sehen, wie man dies am besten aufbauen und darstellen könnte. Manchmal sucht man auch nur "ewig" nach einem Fehler im Sourcecode. Hierbei behebt man dann zunächst noch alle möglichen anderen Baustellen, die man bei genauerem Hinsehen als nicht korrekt aufspürt. Bei mir liegen die "üblen" Fehler immer im Zeiger-Bereich. Kleines Beispiel:

    Diese Zeile hatte ich schnell "hingehauen":

    k_memcpy((void*)ramdisk_start, &file_data_start, &file_data_end - &file_data_start);

    Mit k_memshow(...) sah ich zufrieden, dass die "eingelinkten" Daten gefunden und sauber an den Anfang der RAM Disk kopiert wurden.
    Das Schlimme ist, dass durch die Pointer-Arithmetik zu wenige Daten kopiert wurden. Über die Blödsinnigkeit dieses Fehlers (es gibt nur wenige geniale Fehler) möchte ich hier nicht reden, so etwas passiert eben ab und zu, wenn man den cast auf die Schnelle vergisst. Dieses Wechselspiel zwischen Speicheradresse und Darstellung als unsigned long integer (ULONG) ist eine der Schwächen des Systems. Zwei kleine Casts nach ULONG, und schon funktionierte es perfekt:

    k_memcpy((void*)ramdisk_start, &file_data_start, (ULONG)&file_data_end - (ULONG)&file_data_start);

    Das Schlimme ist, dass man den Fehler an völlig anderer Stelle - nämlich den neu eingebauten Mechanismen - sucht. Auslöser ist, dass man große Speicherbereiche nicht auf einfache Weise wie bei einem Hex-Editor durchscrollen kann. Mir ist der Fehler erst aufgefallen, als ich mich mit k_memshow(...) auf die konkrete Stelle im Speicher gesetzt und dort nur Nullen gefunden habe. 😃
    Solche Erfahrungen sind in einem Tutorial wichtig und sollten auch dargestellt werden, damit andere an solche Trivialitäten denken, wenn Sie Fehler suchen. Denn das "Aufgeben" kommt oft daher, dass man Fehler nicht findet, keine geeigneten Vorbilder hat und/oder in einem Forum (http://forum.osdev.org/ ist ein Musterbeispiel) arrogant behandelt wird.

    Momentan schreibe ich an dem Tutorial noch nicht weiter, weil die grundlegenden Mechanismen zwar funktionieren, ich aber mit der Visualisierung noch nicht zufrieden bin.

    Erste zaghafte Beispiele für Visualisierung sind:

    void k_memshow(void* start, size_t count)
{
    const UCHAR* end = (const UCHAR*)(start+count);
    for(; count != 0; count--) printformat("%x ",*(end-count));
}

    und die Paging-Analyse-Funktionen

    ULONG show_physical_address(ULONG virtual_address)
{
    page_t* page = get_page(virtual_address, 0, kernel_directory);
    return( (page->frame_addr)*PAGESIZE + (virtual_address&0xFFF) );
}

void analyze_physical_addresses()
{
   int i,j,k, k_old;
    for(i=0;i<(PHYSICAL_MEMORY/0x18000+1);++i)
    {
        for(j=i*0x18000; j<i*0x18000+0x18000; j+=0x1000)
        {
            if(show_physical_address(j)==0)
            {
                settextcolor(4,0);
                k_old=k; k=1;
            }
            else
            {
                if(show_physical_address(j)-j)
                {
                    settextcolor(3,0);
                    k_old=k; k=2;
                }
                else
                {
                    settextcolor(2,0);
                    k_old=k; k=3;
                }
            }
            if(k!=k_old)
                printformat("%x %x\n", j, show_physical_address(j));
        }
    }
}

    Im Tasking-Bereich ist mir bisher nur die Darstellung der PID als Farbe beim "Tippen" (Schreiben/Lesen KeyQueue) eingefallen.
    Vor dem Heap-Gebäude stehe ich noch etwas zögerlich bezüglich Darstellung der "blocks and holes". Man hat im Textmode mit 80*25 mit wenigen brauchbaren Farben auch nicht allzu viele Möglichkeiten.

    Didaktische Ideen sind daher immer willkommen. 🙂

    0
  • P

    Well done!

    I am still busy with kmalloc, almost finished it.

    How advanced is your multitasker now?

    And did you make your filesystem manager usable for programmers, who want to make a filesystem driver like Ext2, Ext3, FAT etc?

    // PHPnerd

    0
  • E
    Mod

    How advanced is your multitasker now?

    There is the function task_switch of PrettyOS: 1 -> 2 -> 3 -> ... -> N -> 1 ->...
    This could be triggered by the system timer according to allowed timeslices, but not done until now. I tried it out, but got General Protection Fault. You can't switch at each time. Thus, the question is how to avoid and switch on the right time?
    ..

    As mentioned above, I don't have a time-based scheduler. Here I found some easy code, which might be interesting: ..

    What do you think about it?
    How to prohibit general protection faults by the task switch?

    And did you make your filesystem manager usable for programmers, who want to make a filesystem driver like Ext2, Ext3, FAT etc?

    Not yet, but it should be possible by the VFS.
    ..

    I was very glad, when it worked. 🙂

    Looking at the path forward, I think the next milestones could be:
    - loading executables (file format?)
    - kernel modules
    - shell

    How do you / will you load executables? (one of the text files in the RAM disk of the last example could have been an executable file (elf, com, exe, ...) ). What's the best way to start it?

    0
  • E
    Mod

    Try to use cli(); at start of the handler.

    I did, but the GPF didn't leave. In between I was told that C might be not perfect for the task switch. 🙄 The above presented function for task_switch does not save all registers. That might be the real problem in some situations.

    0
  • E
    Mod

    Another point in the whole source code is the mixing of physical and virtual memory management in paging.h/paging.c. That has to be separated.

    0
  • P

    Sure it has to be separate, i do, the PMM, VMM and kmalloc, all apart.

    What is o_0:

    if(!(queue_empty(&roundrobin_prozesse))&&p->prozess_timetorun<1)

    the double && looks very bad to me

    I checked task switch code with my old, its the same (diff is what I added, like threads and states). I put a cli() just after very first statement. (if(!task)).

    And where exactly (line) does the #GP trigger? (just try putting chars like ! on as much as lines as possible, and count numbeer of printed when running)

    // PHPnerd

    0
  • E
    Mod

    just try putting chars like ! on as much as lines as possible, and count numbeer of printed when running

    Interesting trick. This I try at once:

    !!!!#!!!!#!!!!#General Protection Fault >>> Exception. System Halted! <<<

    Hence, I know that the GPF happens in the asm-code:
    ..
    because the last ! did not show up.

    0
  • P

    You really do? Dont be so sure. What does the ASM do? Yes, it switches task, it chnages page directory, and the instrcution pointer after that code, we continue the task, and not the code of the scheduler.

    Or am i wrong, and is the ! showed at the other switchs?

    // PHP

    0
  • ?

    Erhard Henkes schrieb:

    In between I was told that C might be not perfect for the task switch. The above presented function for task_switch does not save all registers. That might be the real problem in some situations.

    that's true. even if most c-compilers have support for interrupt routines, you can't access cpu registers directly in C (only few compilers for embedded systems support that). you have to code this very low level stuff in assembly. all real life RTOS (although portable among many cpus), with preemptive schedulers, use asm for the heart of the task switcher.
    🙂

    0
  • E
    Mod

    Well, you think, the last ! cannot show up. OK, then I have to try a situation where it works correctly.

    You are right! If it works, then !!!!#!!!!#!!!!## is presented at the screen. Thus, the last # is given out twice instead of the ! 😕

    Now I am in a dead-end-street. 😞

    0
  • P

    Are you using Bochs? I hope so, do following for me:

    put cli, hlt, for(;;); before the big assembly part. Shutdown, and open the bochsout file. Put the register info in here. Btw: remove any calls to the scheduler and call it the way it will fault. So it will give us the right values.

    // PHPnerd

    0
  • E
    Mod 

    /// TEST
      cli();
      asm volatile("hlt;");
      for(;;);
    /// TEST

    asm volatile("         \
    //...

    01368000000p[WGUI ] >>PANIC<< POWER button turned off.
01368000000i[CPU0 ] CPU is in protected mode (halted)
01368000000i[CPU0 ] CS.d_b = 32 bit
01368000000i[CPU0 ] SS.d_b = 32 bit
01368000000i[CPU0 ] EFER   = 0x00000000
01368000000i[CPU0 ] | RAX=000000000000008f  RBX=000000000000d502
01368000000i[CPU0 ] | RCX=00000000000b8000  RDX=00000000000003d5
01368000000i[CPU0 ] | RSP=000000000018ff1c  RBP=000000000018ff44
01368000000i[CPU0 ] | RSI=000000000018ffd8  RDI=000000000018fff0
01368000000i[CPU0 ] |  R8=0000000000000000   R9=0000000000000000
01368000000i[CPU0 ] | R10=0000000000000000  R11=0000000000000000
01368000000i[CPU0 ] | R12=0000000000000000  R13=0000000000000000
01368000000i[CPU0 ] | R14=0000000000000000  R15=0000000000000000
01368000000i[CPU0 ] | IOPL=0 id vip vif ac vm rf nt of df if tf sf zf af pf cf
01368000000i[CPU0 ] | SEG selector     base    limit G D
01368000000i[CPU0 ] | SEG sltr(index|ti|rpl)     base    limit G D
01368000000i[CPU0 ] |  CS:0008( 0001| 0|  0) 00000000 000fffff 1 1
01368000000i[CPU0 ] |  DS:0010( 0002| 0|  0) 00000000 000fffff 1 1
01368000000i[CPU0 ] |  SS:0010( 0002| 0|  0) 00000000 000fffff 1 1
01368000000i[CPU0 ] |  ES:0010( 0002| 0|  0) 00000000 000fffff 1 1
01368000000i[CPU0 ] |  FS:0010( 0002| 0|  0) 00000000 000fffff 1 1
01368000000i[CPU0 ] |  GS:0010( 0002| 0|  0) 00000000 000fffff 1 1
01368000000i[CPU0 ] |  MSR_FS_BASE:0000000000000000
01368000000i[CPU0 ] |  MSR_GS_BASE:0000000000000000
01368000000i[CPU0 ] | RIP=000000000000d3f9 (000000000000d3f9)
01368000000i[CPU0 ] | CR0=0xe0000011 CR1=0x0 CR2=0x0000000000000000
01368000000i[CPU0 ] | CR3=0x00306000 CR4=0x00000000
01368000000i[CPU0 ] >> add esp, 0x00000010 : 83C410
01368000000i[CMOS ] Last time is 1240781579 (Sun Apr 26 23:32:59 2009)
01368000000i[     ] restoring default signal behavior
01368000000i[CTRL ] quit_sim called with exit code 1

    This I did not understand:

    remove any calls to the scheduler and call it the way it will fault. So it will give us the right values.

    The timer_handler:

    void timer_handler(struct regs* r)
{
    ++timer_ticks;
    if (eticks)
        --eticks;

    //TEST
    static ULONG c = 0;
    ++c;
    if(c>200)
    {
      task_switch();
      settextcolor(getpid(),0);
      c=0;
    }
    //TEST
}
    0
  • P

    You see that task_switch() will never be called? you declare a variable, does 1 time incrmeent, and checks for 200?

    ///////////

    Ok, good.

    Try the same, but than in the assbly part itself, each time one line, until the GPF occurs. then we know which opcode makes the fault. (Maybe it is just the jump).

    If that fails, try getting all values of the registers when GPF triggers (print out all registers etc from the Registers given at the handler and post them, and maybe a disassembly of your schedule() (so we can exactly see when it trigger according to RIP's. We cant put cli hlt etc into the assembly part

    // PHPnerd

    0
  • E
    Mod

    PHPnerd schrieb:

    You see that task_switch() will never be called? you declare a variable, does 1 time incrmeent, and checks for 200?

    c is declared static, therefore it works:
    ..

    0
  • P

    What is the info from the #GP?

    // PHP

    0
  • E
    Mod

    Currently, I do not get GPF again. I have done the following:

    1. I used the clobber list:
    asm volatile("..." : : "r"(eip), "r"(esp), "r"(ebp), "r"(current_directory->physicalAddr) : "ebx", "edx");

    ebx and edx were accepted, edi and esi gave failure ("can't find a register in class `GENERAL_REGS' while reloading `asm'").
    2) I accelerated the task switch:

    //TEST
    static ULONG c = 0;
    ++c;
    if(c>0)
    {
      printformat("task_switch\n");
      task_switch();
      settextcolor(getpid(),0);
      c=0;
    }
    //TEST

    output at bochs:

    !task_switch
    !task_switch
    !task_switch
    !HEAP start: 40081000h end: 40100000h max: 4FFFF000h kernel mode: 0 read-only: 0

    ask_switch
    !
    hole 40081000h hole-size: 0007F000h
    after create_heap: placement_address: 0030C020h allocated frames: 795
    ask_switch
    !
    task_switch
    !tasking install
    !tasking install
    after moving stack, ESP: 0018FFD0h ddress: 0030C020h
    After k_mallocing kernel_stack), placement_address: 0030C020h kernel_stack: 4010
    1000h
    VFS & RAM Disk install
    ask_switch
    !!!!#!!!!#!!!!##
    placement_address after ram disk install: 40100000h
    ramdisk_start: 40081000h file_data_start: 0000A2F0h file_data_end: 0000B662h
    ask_switch
    !Page Fault (page not present) at 00D2ECF0h - EIP: 0000D2C9h
    Page Fault >>> Exception. System Halted! <<<

    As you can see, there is one failure free task switch.
    But the output looks disturbed.

    What is the info from the #GP?

    Ihave no specific info from GPF. You mentioned a failure analysis routine. Do you have a link for that code or a table with the error flag meanings?

    I post the current version with the problematic task switch in the time handler:
    http://www.henkessoft.de/OS_Dev/Downloads/37.zip

    I use gcc.exe (version 3.1) and ld.exe (version 2.13) due to linking aout-format of NASM
    http://www.osdever.net/downloads/compilers/DJGPP-Installer-nocpp.exe

    0
Anmelden zum Antworten