Re: + fix-ide-dma-resource-managment.patch added to -mm tree

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

Hello.

Bartlomiej Zolnierkiewicz wrote:

>>>On 3/27/06, Sergei Shtylylov <sshtylyov@ru.mvista.com> wrote:

[skipped]

>>>>>>- release the same number of DMA I/O ports that was requested by a driver

>>>>>please fix trm290.c to use ->mmio == 2 instead

>>>>   It's a _really_ old chipset (pre SFF-8038i) and has _no_ memory mapped
>>>>regs AFAIK...

>>>->mmio == 2 means _only_ that host driver is responsible
>>>for reserving/releasing resources - it doesn't mean that host driver
>>>is using MMIO (a bit confusing but this is what the current code does)

>>    So, your point is that ide_setup_dma() should _never_ do anything for the
>>(mmio == 2) case? Maybe it shouldn't even be called? I guess we surely need

> No, my point is that for ->mmio == 2 core code may not try to manage
> IO or MMIO resources, only that.

    Note that I didn't try to. :-)
    Your 1st NAK was because I moved ide_release_dma_engine() above that
(mmio == 2) check in ide_release_dma() -- which as we've now concluded is 
perfectly correct, isn't it?

> ide_setup_dma() call is still needed for calling ide_allocate_dma_engine()

    Er, looking at sgiioc4.c, I somewhat doubt it -- since that driver 
allocates DMA engine of the _different_ size all by itself...
    However, the driver cleanup function might be called just before releasing 
the engine in ide-dma.c, so this approach sounds reasonable. :-)

> and setting up ->dma_* / ->ide_dma_* fields.

    Yeah, it seems I've forgotten about them a bit... :-)

[skipped]

    Another question: ide_setup_dma() is not supposed to be called by non-PCI 
driver, yes?
It's just I know one (not in the community yet) that does... :-/

> Bartlomiej

WBR, Sergei