Re: [PATCH] cdrom access patch

[Bean]

On Sun, Mar 9, 2008 at 11:33 PM, Bean <address@hidden> wrote:
> On Fri, Mar 7, 2008 at 11:37 PM, Vesa Jääskeläinen <address@hidden> wrote:
>  >
>  > Bean wrote:
>  >  > Hi,
>  >  >
>  >  > This patch use int 13 func 4B01 to detect cdrom. It also change the
>  >  > name of the cd device to (hdN), where N is the bios drive number.
>  >
>  >  Hi Bean,
>  >
>  >  Have you tried to use EDD's functions to query for type of device?
>  >  (references are to EDD3 spec)
>  >
>  >  8.24 Get Device Parameters (FN 48h)
>  >  26 DWord Pointer to the Device Parameter Table Extension (DPTE)
>  >
>  >  8.24.3 Device Parameter Table Extension (DPTE)
>  >  10-11 Word BIOS selected hardware specific option flags
>  >
>  >  Bits 5 and 6; Removable device, ATAPI device.
>  >
>  >  This FN 48h can also be used to filter out PCI cards from ATA driver in
>  >  case there is BIOS support for the device already. So use BIOS if there
>  >  is BIOS handler for it, otherwise fall back to ATA driver.
>  >
>  >  I would prefer that CD's would be counted like (cd0) for first CD-ROM
>  >  device, (cd1) for second CD-ROM device, ... and so on. I think it is OK
>  >  to first allocate BIOS devices for CD's and after that use ATA driver
>  >  for rest of them. Ata driver can of course be accessed with ata name too.
>
>  Hi,
>
>  after some consideration, i think the best way to detect cdrom is to
>  check the number of bytes per sector. The handling of cdrom assume
>  sector size of 4096, cdrom that has other block size would not work.
>  On the other hand, the device don't even needs to be cdrom, as long as
>  it has 2048 sector size, the current handling works. Therefore,
>  checking sector size should be sufficient.
>
>  As for the device name, there are some drawbacks by using (cdN), in
>  init.c, we need to get the root device from bios drive number.
>  However, as Christian Franke discovers, the drive number can be:
>
>  0x82 (old Toshiba Laptop with unknown BIOS)
>  0x9f (PC with Phoenix-Award BIOS, VMware)
>  0xe0 (VirtualBox)
>  0xef (PC with AMI BIOS, VirtualPC)
>
>  It' quite tricky to determine whether a drive is cdrom or harddisk.
>
>  And in the multiboot loader, we need to do the oppisite, map the root
>  device to bios drive number. As names like (cd0) doesn't suggests the
>  number at all, we need to ask biosdisk for the number. Therefore,
>  multiboot loader will be depended on biosdisk, which is not a good
>  idea.

After some testing, i think the new version has some compatible issue,
i remember Christian Franke reports that some bios doesn't support int
42 for cdrom, so grub_biosdisk_check_int13_extensions will fail. Would
someone check if func 4b01 (the first patch cd.diff) works in such
situation ?

-- 
Bean