ide: fix ATAPI CD in Win9x (UNIT ATTENTION root cause) without breaking OAKCDROM (MiSTer-devel#1224)

PR MiSTer-devel#1221 fixed the Win9x protected-mode "Code 10" on the ATAPI CD's IDE
controller by globally stripping the DSC/SERVICE status bit (0x10) for CD
drives. That approach had two problems:

1. It broke real-mode CD detection. OAKCDROM.SYS (the DOS / Win9x-setup ATAPI
   driver) keys off an explicit status byte after the ATAPI DEVICE RESET to
   detect the drive; the blanket strip clobbered it -> "No drives found".
2. The Code 10 it "fixed" was never a DSC problem. The strip only masked the
   real bug.

Real root cause of the Code 10: a successful ATAPI INQUIRY was completing with
a spurious UNIT ATTENTION. INQUIRY / REQUEST SENSE are exempt from the
media-change gate so they deliver their data even with a pending media change,
but the WAIT_PKT_RD completion called cdrom_reply() with the default
unit_attention=true, stamping CHECK CONDITION / err=0x60 onto the finished
transfer. Win9x's ESDI_506 reads "INQUIRY returned its data AND failed" as a
dead controller and refuses to start the channel -> Device Manager Code 10.
Bochs and 86Box both return GOOD for INQUIRY.

Fixes:
- ide_set_regs: replace the global DSC strip with "auto-assert DSC for non-CD
  drives only", letting each ATAPI command publish its own (DSC-free) status.
  Satisfies both OAK and Win9x without any blanket bit-4 policy.
- ATAPI DEVICE RESET (cmd 0x08): present status 0x00 (no DRDY/DSC), matching
  real ATAPI hardware, which routes OAKCDROM to its signature-detect path.
- WAIT_PKT_RD completion: pass unit_attention=false so a successful data phase
  reports GOOD. The pending media-change UA is still delivered on the first
  non-exempt command (TEST UNIT READY -> cdrom_nodisk).
- Post ATAPI diagnostic code 01h in the error register after a bus reset; a
  stale ABRT was otherwise read by ESDI_506 as a failed power-on self-test.

Tested on DE10-Nano: Win98 SE enumerates "MiSTer CDROM" with both Standard
IDE/ESDI controllers clean (no Code 10), and real-mode OAKCDROM.SYS detects the
CD (MSCDEX assigns drive D:).

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

Author: kblood

ide.cpp

@@ -160,14 +160,11 @@ void ide_get_regs(ide_config *ide)
 
 void ide_set_regs(ide_config *ide)
 {
-	if (!(ide->regs.status & (ATA_STATUS_BSY | ATA_STATUS_ERR))) ide->regs.status |= ATA_STATUS_DSC;
-
-	// For ATAPI (packet) devices, status bit 4 is the SERVICE bit, not DSC. We never run
-	// overlapped commands, so SERVICE must read 0. Force it low for CD drives, matching 86Box
-	// ide_status() which masks DSC out of every ATAPI status read (hdc_ide.c). Win9x ESDI_506
-	// otherwise reads the permanent bit-4 as "overlapped-command SERVICE pending", rejects the
-	// channel, and the device fails to start (Device Manager Code 10).
-	if (ide->drive[ide->regs.drv].cd) ide->regs.status &= ~ATA_STATUS_DSC;
+	// Auto-assert DSC for non-CD drives only; let each ATAPI command set its own status.
+	if (!ide->drive[ide->regs.drv].cd && !(ide->regs.status & (ATA_STATUS_BSY | ATA_STATUS_ERR)))
+	{
+		ide->regs.status |= ATA_STATUS_DSC;
+	}
 
 	uint8_t data[12] =
 	{
@@ -1016,7 +1013,8 @@ void ide_io(int num, int req)
 		else if (ide->state == IDE_STATE_WAIT_PKT_RD)
 		{
 			if (ide->regs.pkt_cnt) cdrom_read(ide);
-			else cdrom_reply(ide, 0);
+			// A successful data-phase completion must report GOOD, not UNIT ATTENTION.
+			else cdrom_reply(ide, 0, 0, 0, false);
 		}
 		else if (ide->state == IDE_STATE_WAIT_PKT_MODE)
 		{
@@ -1049,7 +1047,8 @@ void ide_io(int num, int req)
 
 		ide_get_regs(ide);
 		ide->regs.head = 0;
-		ide->regs.error = 0;
+		// ATAPI posts diagnostic code 01h after a bus reset; an ATA HDD posts 0.
+		ide->regs.error = ide->drive[ide->regs.drv].cd ? 1 : 0;
 		ide->regs.sector = 1;
 		ide->regs.sector_count = 1;
 		ide->regs.cylinder = (!ide->drive[ide->regs.drv].present) ? 0xFFFF : ide->drive[ide->regs.drv].cd ? 0xEB14 : 0x0000;

ide_cdrom.cpp

@@ -1579,11 +1579,14 @@ int cdrom_handle_cmd(ide_config *ide)
 		ide->regs.sector_count = 1;
 		ide->regs.cylinder = 0xEB14;
 		ide->regs.head = 0;
+		// ATAPI diagnostic code 01h after DEVICE RESET.
+		ide->regs.error = 1;
 		ide->regs.io_size = 0;
-		ide->regs.status = ATA_STATUS_RDY | ATA_STATUS_DSC;
+		// Present status 0x00 (no DRDY/DSC), matching real ATAPI HW; routes OAKCDROM to signature detect.
+		ide->regs.status = 0;
 		ide_set_regs(ide);
 		break;
-	
+
 		case 0xEF: // set features
 		switch(ide->regs.features)
 		{