probing of ATA slaves with no master broken (was: Re: ATAPI CDstill not detected, verbose boot logs available)

On Wed, 3 Dec 2003, Soren Schmidt wrote:

> Could you try this simple patch and see if that helps?
>
> Index: ata-lowlevel.c
> ===================================================================
> RCS file: /home/ncvs/src/sys/dev/ata/ata-lowlevel.c,v
> retrieving revision 1.23
> diff -u -r1.23 ata-lowlevel.c
> --- ata-lowlevel.c      2 Nov 2003 22:04:53 -0000       1.23
> +++ ata-lowlevel.c      3 Dec 2003 07:50:44 -0000
> _at__at_ -575,7 +575,7 _at__at_
>                 }
>             }
>         }
> -       if (stat1 & ATA_S_BUSY) {
> +       if (!((mask == 0x03) && (stat0 & ATA_S_BUSY)) && (stat1 & ATA_S_BUSY)) {
>             ATA_IDX_OUTB(ch, ATA_DRIVE, ATA_D_IBM | ATA_SLAVE);
>             DELAY(10);

I finally got around to testing this although not on the machine that
it should fix problems for (one with a slow master and a slave; the
patch should help in this case).  It breaks another machine with an
ATA slave with no master: it causes a 31 second delay and fails to
find the slave:

Dec 10 22:31:42 gamplex kernel: atapci1: <HighPoint HPT366 UDMA66 controller> port 0xbc00-0xbcff,0xb800-0xb803,0xb400-0xb407 irq 11 at device 19.0 on pci0
Dec 10 22:31:42 gamplex kernel: ithread_add_handler: atapci1: id -1
Dec 10 22:31:42 gamplex kernel: atapci1: [MPSAFE]
Dec 10 22:31:42 gamplex kernel: ata2: reset tp1 mask=03 ostat0=7f ostat1=50
Dec 10 22:31:42 gamplex kernel: ata2: master _at_1.061986: 7f 7f 7f 7f; slave _at_1.062015: 00 01 14 eb
Dec 10 22:31:42 gamplex kernel: ata2-master: stat=0x7f err=0x7f lsb=0x7f msb=0x7f
Dec 10 22:31:42 gamplex kernel: ata2-slave:  stat=0x00 err=0x01 lsb=0x14 msb=0xeb
Dec 10 22:31:42 gamplex kernel: ata2: reset tp2 mask=03 stat0=7f stat1=00 devices=0x8<ATAPI_SLAVE>
Dec 10 22:31:42 gamplex kernel: ata2: at 0xb400 on atapci1
Dec 10 22:31:42 gamplex kernel: ata2: [MPSAFE]

This is because stat0 is always frobbed back to ATA_S_BUSY, so the slave is
never probed.  From ata-lowlevel.c:

% 	    if (!(stat0 & ATA_S_BUSY)) {
% 		if ((err & 0x7f) == ATA_E_ILI) {
% 		    if (lsb == ATAPI_MAGIC_LSB && msb == ATAPI_MAGIC_MSB) {
% 			ch->devices |= ATA_ATAPI_MASTER;
% 		    }
% 		    else if (stat0 & ATA_S_READY) {
% 			ch->devices |= ATA_ATA_MASTER;
% 		    }
% 		}
% 		else if ((stat0 & 0x4f) && err == lsb && err == msb) {
% 		    stat0 |= ATA_S_BUSY;
  		    ^^^^^^^^^^^^^^^^^^^^
The master's real-stat0/err/lsb/msb are always 0x7f with my hardware, so
this is always reached.

% 		}
% 	    }
% 	}
% 	if (!((mask == 0x03) && (stat0 & ATA_S_BUSY)) && (stat1 & ATA_S_BUSY)) {
  	                        ^^^^^^^^^^^^^^^^^^^^
The above makes stat0 always unreal here, so the slave is never probed.

Bruce