Re: [linux-usb-devel] Re: SLAB vs. pci_alloc_xxx in usb-uhci patch

[David Brownell <david-b@pacbell.net>]

> > At the time, I didn't feel like creating a custom sub-allocator just
> > for USB, ...
> >
> > I'd be good to get it done "properly" at some point though.
> 
> Something like
> 
> struct pci_pool *pci_alloc_consistent_pool(int objectsize, int align)

    struct pci_pool *
    pci_create_consistent_pool (struct pci_dev *dev, int size, int align)

and similar for freeing the pool ... pci_alloc_consistent() needs the device,
presumably since some devices may need to dma into specific memory.
I'd probably want at least "flags" from kmem_cache_create().


> pci_alloc_pool_consistent(pool,..
> pci_free_pool_consistent(pool,..

These should have signatures just like pci_alloc_consistent() and
pci_free_consistent() except they take the pci_pool, not a pci_dev.
Oh, and likely GFP_ flags to control blocking.


> Where the pool allocator does page grabbing and chaining

Given an agreement on API, I suspect Johannes' patch could get
quickly generalized.  Then debugging support (like in slab.c) could
be added later.

- Dave