by Finn Thain

[permalink] [raw]

Subject: Re: [PATCH v2 74/71] ncr5380: Enable PDMA for NCR53C400A

On Wed, 9 Dec 2015, Ondrej Zary wrote:

> On Tuesday 08 December 2015 03:05:11 Finn Thain wrote:
> >
> > On Sun, 6 Dec 2015, Ondrej Zary wrote:
> >
> > > Add I/O register mapping for NCR53C400A and enable PDMA mode to
> > > improve performance and fix non-working IRQ.
> > >
> > > Tested with HP C2502 (and user-space enabler).
> > >
> > > Signed-off-by: Ondrej Zary <[email protected]>
> > > ---
> > > drivers/scsi/g_NCR5380.c | 14 +++++++++++---
> > > 1 file changed, 11 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/scsi/g_NCR5380.c b/drivers/scsi/g_NCR5380.c
> > > index 86740fd..099fdac 100644
> > > --- a/drivers/scsi/g_NCR5380.c
> > > +++ b/drivers/scsi/g_NCR5380.c
> > > @@ -324,7 +324,7 @@ static int __init generic_NCR5380_detect(struct scsi_host_template *tpnt)
> > > #endif
> > > break;
> > > case BOARD_NCR53C400A:
> > > - flags = FLAG_NO_PSEUDO_DMA;
> > > + flags = FLAG_NO_DMA_FIXUP;
> > > ports = ncr_53c400a_ports;
> > > break;
> > > case BOARD_DTC3181E:
> > > @@ -406,11 +406,18 @@ static int __init generic_NCR5380_detect(struct scsi_host_template *tpnt)
> > > * On NCR53C400 boards, NCR5380 registers are mapped 8 past
> > > * the base address.
> > > */
> > > - if (overrides[current_override].board == BOARD_NCR53C400) {
> > > + switch (overrides[current_override].board) {
> > > + case BOARD_NCR53C400:
> > > instance->io_port += 8;
> > > hostdata->c400_ctl_status = 0;
> > > hostdata->c400_blk_cnt = 1;
> > > hostdata->c400_host_buf = 4;
> > > + break;
> > > + case BOARD_NCR53C400A:
> > > + hostdata->c400_ctl_status = 9;
> > > + hostdata->c400_blk_cnt = 10;
> > > + hostdata->c400_host_buf = 8;
> > > + break;
> > > }
> > > #else
> > > instance->base = overrides[current_override].NCR5380_map_name;
> >
> >
> > For SCSI_G_NCR5380_MEM and BOARD_NCR53C400A (or BOARD_DTC3181E), you have
> > not assigned c400_ctl_status, c400_blk_cnt and c400_host_buf. Perhaps we
> > should throw an error, something like this?
> >
> > hostdata->iomem = iomem;
> > - if (overrides[current_override].board == BOARD_NCR53C400) {
> > + switch (overrides[current_override].board) {
> > + case BOARD_NCR53C400:
> > hostdata->c400_ctl_status = 0x100;
> > hostdata->c400_blk_cnt = 0x101;
> > hostdata->c400_host_buf = 0x104;
> > + break;
> > + case BOARD_NCR53C400A:
> > + pr_err(DRV_MODULE_NAME ": unknown register offsets\n");
> > + goto out_unregister;
> > }
> > #endif
>
> We don't need to fail, just disable PDMA by setting:
> flags = FLAG_NO_PSEUDO_DMA;
>
> And BTW. this:
> if (overrides[current_override].board == BOARD_NCR53C400 ||
> overrides[current_override].board == BOARD_NCR53C400A)
> NCR5380_write(hostdata->c400_ctl_status, CSR_BASE);
>
> should then be, solving the problem of growing if condition:
> if (!(flags & FLAG_NO_PSEUDO_DMA))
> NCR5380_write(hostdata->c400_ctl_status, CSR_BASE);
>

We already resolved that problem. Julian Calaby and I agreed that a switch
statement is better than a flag here.

Moreover, if the user sets the 'ncr_53c400' or 'ncr_53c400a' module
options I presume they expect corresponding device initialization. I don't
think PIO means we should skip that.

Passing 'ncr_53c400' or 'ncr_53c400a' options to the g_NCR5380_mmio module
has to give an error if we don't have enough information about the board
to allow us to do the right thing.

(Passing the 'ncr_5380' module param may give the result you wanted.)

BTW, I fixed up this patch when I added it to my queue. I'll send you my
version to test.

Thanks.

--