tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1In Linux 2.5 kernels (and later), USB device drivers have additional control
2over how DMA may be used to perform I/O operations. The APIs are detailed
3in the kernel usb programming guide (kerneldoc, from the source code).
4
5
6API OVERVIEW
7
8The big picture is that USB drivers can continue to ignore most DMA issues,
9though they still must provide DMA-ready buffers (see
10Documentation/PCI/PCI-DMA-mapping.txt). That's how they've worked through
11the 2.4 (and earlier) kernels.
12
13OR: they can now be DMA-aware.
14
15- New calls enable DMA-aware drivers, letting them allocate dma buffers and
16 manage dma mappings for existing dma-ready buffers (see below).
17
18- URBs have an additional "transfer_dma" field, as well as a transfer_flags
19 bit saying if it's valid. (Control requests also have "setup_dma" and a
20 corresponding transfer_flags bit.)
21
22- "usbcore" will map those DMA addresses, if a DMA-aware driver didn't do
23 it first and set URB_NO_TRANSFER_DMA_MAP or URB_NO_SETUP_DMA_MAP. HCDs
24 don't manage dma mappings for URBs.
25
26- There's a new "generic DMA API", parts of which are usable by USB device
27 drivers. Never use dma_set_mask() on any USB interface or device; that
28 would potentially break all devices sharing that bus.
29
30
31ELIMINATING COPIES
32
33It's good to avoid making CPUs copy data needlessly. The costs can add up,
34and effects like cache-trashing can impose subtle penalties.
35
36- If you're doing lots of small data transfers from the same buffer all
37 the time, that can really burn up resources on systems which use an
38 IOMMU to manage the DMA mappings. It can cost MUCH more to set up and
39 tear down the IOMMU mappings with each request than perform the I/O!
40
41 For those specific cases, USB has primitives to allocate less expensive
42 memory. They work like kmalloc and kfree versions that give you the right
43 kind of addresses to store in urb->transfer_buffer and urb->transfer_dma.
44 You'd also set URB_NO_TRANSFER_DMA_MAP in urb->transfer_flags:
45
46 void *usb_buffer_alloc (struct usb_device *dev, size_t size,
47 int mem_flags, dma_addr_t *dma);
48
49 void usb_buffer_free (struct usb_device *dev, size_t size,
50 void *addr, dma_addr_t dma);
51
52 Most drivers should *NOT* be using these primitives; they don't need
53 to use this type of memory ("dma-coherent"), and memory returned from
54 kmalloc() will work just fine.
55
56 For control transfers you can use the buffer primitives or not for each
57 of the transfer buffer and setup buffer independently. Set the flag bits
58 URB_NO_TRANSFER_DMA_MAP and URB_NO_SETUP_DMA_MAP to indicate which
59 buffers you have prepared. For non-control transfers URB_NO_SETUP_DMA_MAP
60 is ignored.
61
62 The memory buffer returned is "dma-coherent"; sometimes you might need to
63 force a consistent memory access ordering by using memory barriers. It's
64 not using a streaming DMA mapping, so it's good for small transfers on
65 systems where the I/O would otherwise thrash an IOMMU mapping. (See
66 Documentation/PCI/PCI-DMA-mapping.txt for definitions of "coherent" and
67 "streaming" DMA mappings.)
68
69 Asking for 1/Nth of a page (as well as asking for N pages) is reasonably
70 space-efficient.
71
72 On most systems the memory returned will be uncached, because the
73 semantics of dma-coherent memory require either bypassing CPU caches
74 or using cache hardware with bus-snooping support. While x86 hardware
75 has such bus-snooping, many other systems use software to flush cache
76 lines to prevent DMA conflicts.
77
78- Devices on some EHCI controllers could handle DMA to/from high memory.
79
80 Unfortunately, the current Linux DMA infrastructure doesn't have a sane
81 way to expose these capabilities ... and in any case, HIGHMEM is mostly a
82 design wart specific to x86_32. So your best bet is to ensure you never
83 pass a highmem buffer into a USB driver. That's easy; it's the default
84 behavior. Just don't override it; e.g. with NETIF_F_HIGHDMA.
85
86 This may force your callers to do some bounce buffering, copying from
87 high memory to "normal" DMA memory. If you can come up with a good way
88 to fix this issue (for x86_32 machines with over 1 GByte of memory),
89 feel free to submit patches.
90
91
92WORKING WITH EXISTING BUFFERS
93
94Existing buffers aren't usable for DMA without first being mapped into the
95DMA address space of the device. However, most buffers passed to your
96driver can safely be used with such DMA mapping. (See the first section
97of Documentation/PCI/PCI-DMA-mapping.txt, titled "What memory is DMA-able?")
98
99- When you're using scatterlists, you can map everything at once. On some
100 systems, this kicks in an IOMMU and turns the scatterlists into single
101 DMA transactions:
102
103 int usb_buffer_map_sg (struct usb_device *dev, unsigned pipe,
104 struct scatterlist *sg, int nents);
105
106 void usb_buffer_dmasync_sg (struct usb_device *dev, unsigned pipe,
107 struct scatterlist *sg, int n_hw_ents);
108
109 void usb_buffer_unmap_sg (struct usb_device *dev, unsigned pipe,
110 struct scatterlist *sg, int n_hw_ents);
111
112 It's probably easier to use the new usb_sg_*() calls, which do the DMA
113 mapping and apply other tweaks to make scatterlist i/o be fast.
114
115- Some drivers may prefer to work with the model that they're mapping large
116 buffers, synchronizing their safe re-use. (If there's no re-use, then let
117 usbcore do the map/unmap.) Large periodic transfers make good examples
118 here, since it's cheaper to just synchronize the buffer than to unmap it
119 each time an urb completes and then re-map it on during resubmission.
120
121 These calls all work with initialized urbs: urb->dev, urb->pipe,
122 urb->transfer_buffer, and urb->transfer_buffer_length must all be
123 valid when these calls are used (urb->setup_packet must be valid too
124 if urb is a control request):
125
126 struct urb *usb_buffer_map (struct urb *urb);
127
128 void usb_buffer_dmasync (struct urb *urb);
129
130 void usb_buffer_unmap (struct urb *urb);
131
132 The calls manage urb->transfer_dma for you, and set URB_NO_TRANSFER_DMA_MAP
133 so that usbcore won't map or unmap the buffer. The same goes for
134 urb->setup_dma and URB_NO_SETUP_DMA_MAP for control requests.
135
136Note that several of those interfaces are currently commented out, since
137they don't have current users. See the source code. Other than the dmasync
138calls (where the underlying DMA primitives have changed), most of them can
139easily be commented back in if you want to use them.