OSDev.org

mohammedisam wrote:

Code: Select all

    floppy_cmd_send(0x08);	//'check interrupt' command
    cyl = floppy_read_data();	//st0
    st1 = floppy_read_data();	//st1
    cyl = floppy_read_data();	//st2
    cyl = floppy_read_data();
    printf("cyl: %u ", (unsigned)cyl);

• You should move all the "seek()" stuff into a separate function.
• You should check the values in ST0 to determine if seek worked or not. If seek didn't work (e.g. the command was aborted, the "equipment check" flag is set, etc) then the "cylinder" it returns can be wrong/misleading.
• If the seek command fails, you should do a recalibrate command.
• You should remember the last cylinder you used, and if someone wants to seek to the previously used cylinder then you should do nothing (avoid doing "seek to where we are already").
• Turning the floppy motor on takes ages (something like a 300 ms delay as it gets up to speed?). Because of this you don't want to turn the floppy motor off immediately, but want to keep the motor on for 5 seconds or more afterwards. That way if more disk IO is needed "soon" the floppy motor is still on and you can skip the lengthy delay.
• Whenever the floppy motor has been turned off, assume the user may have removed the disk from the disk drive. This means checking the "disk changed" flag in the Digital Input Register (if possible); and if that flag is broken using some other method (e.g. checksum of the disk's first cylinder) to determine if the disk is the same disk as before. If the disk was changed (or your "detect if disk changed using other methods" failed) you'd have wipe out the driver's current state (including setting your "lastCylinderUsed" to unknown) and go back to detecting if there's a disk in the drive, determining the disk's format (1680 KiB floppy, 1440 KiB floppy, 720 KiB floppy or whatever), etc.
• Floppy is slow, so caching is important. I typically cache entire tracks, and never read less than an entire track into the disk cache.
• Floppy disks are relatively tiny (compared to the amount of RAM in modern computers). This means that it's probably a good idea to cache the entire disk; and to do pre-fetching. Basically, as soon as the floppy is first used (and you've determined the disk's format) start pre-fetching the entire disk into your cache in the background (whenever there's nothing else for the floppy controller to do).
• Floppy disks are also relatively unreliable. If "multiple sector reads" fail then fall back to one sector at a time, and if this happens often then stop trying to use "multiple sector reads". This increases the chance of successfully reading. For writing, I'd be tempted to write all sectors on a track, then verify that those sectors can be read correctly before moving on to the next track.

Brendan wrote: [*]You should check the values in ST0 to determine if seek worked or not. If seek didn't work (e.g. the command was aborted, the "equipment check" flag is set, etc) then the "cylinder" it returns can be wrong/misleading.

Brendan wrote: [*]If the seek command fails, you should do a recalibrate command.

int floppy_calibrate(uint32_t drive)
{
  uint32_t cyl;
  if(drive >= 4)
    return 0;
  
  for(int i = 0; i < 10; i++)
  {
    floppy_cmd_send(0x07);	//the calibrate command
    floppy_cmd_send(drive);
    floppy_wait_on_irq();
    /* tell FDC we handled the interrupt */
    floppy_cmd_send(0x08);	//'check interrupt' command
    floppy_read_data();	//st0
    cyl = floppy_read_data();

    //if we hit cylinder 0, we are done
    if(!cyl)
    {
      printf("Hit cyl 0!\n");
      return 1;
    }
  }
  return 0;
}

Brendan wrote: [*]Floppy disks are also relatively unreliable. If "multiple sector reads" fail then fall back to one sector at a time, and if this happens often then stop trying to use "multiple sector reads". This increases the chance of successfully reading. For writing, I'd be tempted to write all sectors on a track, then verify that those sectors can be read correctly before moving on to the next track.