On 3/27/06, Sergei Shtylylov <sshtylyov@xxxxxxxxxxxxx> wrote: > Hello. > > Sergei Shtylylov wrote: > > > Bartlomiej Zolnierkiewicz wrote: > > [skipped] > > >>> Resolve several IDE DMA resource management issues: > > >>> - release DMA engine for memory mapped DMA as well > > >> NAK - this won't work, see siimage.c how these resources were > >> really reserved in the first place [ request_mem_region() ] > > NB: it's reserved this way only when available -- this driver can work > with both I/O and memory mapped regs... > > > Oops, looks like I missed that missing ide_setup_dma() call in that > > driver... > > No, I've just forgotten that it's called implicitly from setup-pci.c since > init_dma() "method" in not defined in that driver. So, looks like you're wrong > here -- it should work currently BUT the DMA engine is not released since > (mmio == 2) check and return precedes call to ide_release_dma_engine() in > ide_release_dma(). please take a look at siimage.c:init_iops_siimage(): if (pci_get_drvdata(dev) == NULL) return; init_mmio_iops_siimage(hwif); and siimage.c:init_mmio_iops_siimage(): hwif->mmio = 2; so ->mmio == 2 IFF MMIO is used * for ->mmio == 0 current code seem to release resources properly * for ->mmio == 2 you need to release the region that was reserved originally (for siimage it is one big fat mem region) and not standard DMA I/O regions, therefore this change is wrong > And as I said, there's no provision to release any driver-specific > resources on a call to ide_release_dma()... That's why asked you to take look at libata-core.c:ata_host_stop() etc. and to add any needed core code... :-) > [skipped] > > >> hwif->mmio == 2 means that host driver is responsible > >> for reserving/releasing resources (this is _preferred_ way > >> of handling resources) > > > Hm, didn't know that... > > ... and this was hard to guess from the current code. ;-) > > >> 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... > > The only thing that can be done is not to call ide_setup_dma() from it... Exactly and make trm290.c to handle resources itself... > >>> - release the channel's secondary DMA I/O ports for real > > >> NAK, looking and the code - secondary DMA base is only used > >> by siimage.c and sgiioc4.c (both use ->mmio == 2, for siimage ->dma_base2 > >> is unused, for sgiioc4 it is used for some weird DMA transfer ending > >> stuff) > > > Yes, I have sgioc4.c patch that deals with this (almost) ready... > > ... and for siimage, this seems a no-brainer. :-) Yes, you may want to fix it... :-) Thanks, Bartlomiej - : send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
