Fixes #13: dma_set_coherent_mask to 64 bits

[glimchb]

to be able to allocate large memory using CMA

[glimchb]

@wesleywesley @lsgunth @dmeyergio - please review and merge

[glimchb]

@wesleywesley @lsgunth @dmeyergio - please review and merge

[lsgunth]

I think the change belongs in the proper switchtec driver during PCI init. Not in the NTB code.

I'd also like to see this go upstream and for that we will need a better commit message and a sign-off tag.

Thanks,

Logan

[glimchb]

@lsgunth do you mean inside switchtec_init_pci() function in switchtec.c ?
the basic management driver have no idea regarding DMA usage... so why it have to be there?

[lsgunth]

The DMA mask is a property of the PCI device not really of the usage of that device and we don't need other potential client drivers competing over what/when to set the mask.

The latest changes for switchtec kernel also have started doing DMA requests too.

switchtec_init_pci() seems like a good spot.

Logan

[glimchb]

@lsgunth done, please review

[lsgunth]

Looks good to me.

[glimchb]

@wesleywesley please merge

[wesleywesley]

One question:
Should we use
int dma_set_mask_and_coherent(struct device *dev, u64 mask);
instead of
int dma_set_mask_and_coherent(struct device *dev, u64 mask);
because the latter is only apply to consistent allocations, but the former API is for both streaming and coherent?

Regard,
Wesley

[lsgunth]

Yes, dma_set_mask_and_coherent() is probably a better choice.

Logan

[wesleywesley]

@lsgunth
Thx for comments.
Will draft a pull request for this.

Regard,
Wesley