Re: [PATCH 2/2] ide: clear bmdma status in ide_intr() (revised)

[Sergei Shtylyov <sshtylyov@ru.mvista.com>]

Albert Lee wrote:
> patch 2/2 (revised):
> - Do the dma status clearing in ide_intr() and add a new hwif->ide_dma_clear_irq such that LLDD can override it.
> - Fix drive->waiting_for_dma to work with CDB-intr devices.
> 
> Signed-off-by: Albert Lee <albertcc@tw.ibm.com>
> ---
> hwif->dma is not reliable: ide_intr() races with dma_start(). Sometimes we got hwif->dma == 0
> in ide_intr() even though it is actually a DMA interrupt. So, drive->waiting_for_dma is used instead.
> 
> Also revised per Sergei's comments and let ide_dma_clear_irq return "void".
> The "if (hwif->dma_status)" check in __ide_dma_clear_irq() is kept as is since I think
> it would be safer for the old ISA/VESA IDE devices that has no BMDMA registers.

    ISA/VESA drivers shouldn't have this method defined at all since they're 
PIO only. But well, it won't hurt...

> For your review, thanks.

> diff -Nrup 01_remove_from_ide_cd/drivers/ide/ide-cd.c 02_add_to_ide_intr/drivers/ide/ide-cd.c
> --- 01_remove_from_ide_cd/drivers/ide/ide-cd.c	2007-01-24 11:00:03.000000000 +0800
> +++ 02_add_to_ide_intr/drivers/ide/ide-cd.c	2007-01-25 16:52:20.000000000 +0800
> @@ -923,6 +923,10 @@ static ide_startstop_t cdrom_start_packe
>  		HWIF(drive)->OUTB(drive->ctl, IDE_CONTROL_REG);
>   
>  	if (CDROM_CONFIG_FLAGS (drive)->drq_interrupt) {
> +		/* waiting for CDB interrupt, not DMA yet. */
> +		if (info->dma)
> +			drive->waiting_for_dma = 0;
> +
>  		/* packet command */
>  		ide_execute_command(drive, WIN_PACKETCMD, handler, ATAPI_WAIT_PC, cdrom_timer_expiry);
>  		return ide_started;
> @@ -965,6 +969,10 @@ static ide_startstop_t cdrom_transfer_pa
>  		/* Check for errors. */
>  		if (cdrom_decode_status(drive, DRQ_STAT, NULL))
>  			return ide_stopped;
> +
> +		/* Ok, next interrupt will be dma interrupt. */
> +		if (info->dma)
> +			drive->waiting_for_dma = 1;
>  	} else {
>  		/* Otherwise, we must wait for DRQ to get set. */
>  		if (ide_wait_stat(&startstop, drive, DRQ_STAT,

    Erm... shouldn't we set drive->waiting_for_dma in hwif->dma_start() then?
Why it's set in hwif->dma_setup() at all I wonder?

MBR, Sergei