1 Devres - Managed Device Resource
2 ================================
4 Tejun Heo <teheo@suse.de>
6 First draft 10 January 2007
9 1. Intro : Huh? Devres?
10 2. Devres : Devres in a nutshell
11 3. Devres Group : Group devres'es and release them together
12 4. Details : Life time rules, calling context, ...
13 5. Overhead : How much do we have to pay for this?
14 6. List of managed interfaces : Currently implemented managed interfaces
20 devres came up while trying to convert libata to use iomap. Each
21 iomapped address should be kept and unmapped on driver detach. For
22 example, a plain SFF ATA controller (that is, good old PCI IDE) in
23 native mode makes use of 5 PCI BARs and all of them should be
26 As with many other device drivers, libata low level drivers have
27 sufficient bugs in ->remove and ->probe failure path. Well, yes,
28 that's probably because libata low level driver developers are lazy
29 bunch, but aren't all low level driver developers? After spending a
30 day fiddling with braindamaged hardware with no document or
31 braindamaged document, if it's finally working, well, it's working.
33 For one reason or another, low level drivers don't receive as much
34 attention or testing as core code, and bugs on driver detach or
35 initialization failure don't happen often enough to be noticeable.
36 Init failure path is worse because it's much less travelled while
37 needs to handle multiple entry points.
39 So, many low level drivers end up leaking resources on driver detach
40 and having half broken failure path implementation in ->probe() which
41 would leak resources or even cause oops when failure occurs. iomap
42 adds more to this mix. So do msi and msix.
48 devres is basically linked list of arbitrarily sized memory areas
49 associated with a struct device. Each devres entry is associated with
50 a release function. A devres can be released in several ways. No
51 matter what, all devres entries are released on driver detach. On
52 release, the associated release function is invoked and then the
53 devres entry is freed.
55 Managed interface is created for resources commonly used by device
56 drivers using devres. For example, coherent DMA memory is acquired
57 using dma_alloc_coherent(). The managed version is called
58 dmam_alloc_coherent(). It is identical to dma_alloc_coherent() except
59 for the DMA memory allocated using it is managed and will be
60 automatically released on driver detach. Implementation looks like
66 dma_addr_t dma_handle;
69 static void dmam_coherent_release(struct device *dev, void *res)
71 struct dma_devres *this = res;
73 dma_free_coherent(dev, this->size, this->vaddr, this->dma_handle);
76 dmam_alloc_coherent(dev, size, dma_handle, gfp)
78 struct dma_devres *dr;
81 dr = devres_alloc(dmam_coherent_release, sizeof(*dr), gfp);
84 /* alloc DMA memory as usual */
85 vaddr = dma_alloc_coherent(...);
88 /* record size, vaddr, dma_handle in dr */
97 If a driver uses dmam_alloc_coherent(), the area is guaranteed to be
98 freed whether initialization fails half-way or the device gets
99 detached. If most resources are acquired using managed interface, a
100 driver can have much simpler init and exit code. Init path basically
101 looks like the following.
107 d = devm_kzalloc(dev, sizeof(*d), GFP_KERNEL);
111 d->ring = dmam_alloc_coherent(...);
119 return register_to_upper_layer(d);
126 unregister_from_upper_layer(d);
127 shutdown_my_hardware();
130 As shown above, low level drivers can be simplified a lot by using
131 devres. Complexity is shifted from less maintained low level drivers
132 to better maintained higher layer. Also, as init failure path is
133 shared with exit path, both can get more testing.
139 Devres entries can be grouped using devres group. When a group is
140 released, all contained normal devres entries and properly nested
141 groups are released. One usage is to rollback series of acquired
142 resources on failure. For example,
144 if (!devres_open_group(dev, NULL, GFP_KERNEL))
156 devres_remove_group(dev, NULL);
160 devres_release_group(dev, NULL);
163 As resource acquisition failure usually means probe failure, constructs
164 like above are usually useful in midlayer driver (e.g. libata core
165 layer) where interface function shouldn't have side effect on failure.
166 For LLDs, just returning error code suffices in most cases.
168 Each group is identified by void *id. It can either be explicitly
169 specified by @id argument to devres_open_group() or automatically
170 created by passing NULL as @id as in the above example. In both
171 cases, devres_open_group() returns the group's id. The returned id
172 can be passed to other devres functions to select the target group.
173 If NULL is given to those functions, the latest open group is
176 For example, you can do something like the following.
178 int my_midlayer_create_something()
180 if (!devres_open_group(dev, my_midlayer_create_something, GFP_KERNEL))
185 devres_close_group(dev, my_midlayer_create_something);
189 void my_midlayer_destroy_something()
191 devres_release_group(dev, my_midlayer_create_something);
198 Lifetime of a devres entry begins on devres allocation and finishes
199 when it is released or destroyed (removed and freed) - no reference
202 devres core guarantees atomicity to all basic devres operations and
203 has support for single-instance devres types (atomic
204 lookup-and-add-if-not-found). Other than that, synchronizing
205 concurrent accesses to allocated devres data is caller's
206 responsibility. This is usually non-issue because bus ops and
207 resource allocations already do the job.
209 For an example of single-instance devres type, read pcim_iomap_table()
212 All devres interface functions can be called without context if the
213 right gfp mask is given.
219 Each devres bookkeeping info is allocated together with requested data
220 area. With debug option turned off, bookkeeping info occupies 16
221 bytes on 32bit machines and 24 bytes on 64bit (three pointers rounded
222 up to ull alignment). If singly linked list is used, it can be
223 reduced to two pointers (8 bytes on 32bit, 16 bytes on 64bit).
225 Each devres group occupies 8 pointers. It can be reduced to 6 if
226 singly linked list is used.
228 Memory space overhead on ahci controller with two ports is between 300
229 and 400 bytes on 32bit machine after naive conversion (we can
230 certainly invest a bit more effort into libata core layer).
233 6. List of managed interfaces
234 -----------------------------
241 devm_request_region()
242 devm_request_mem_region()
243 devm_release_region()
244 devm_release_mem_region()
251 dmam_alloc_coherent()
253 dmam_alloc_noncoherent()
254 dmam_free_noncoherent()
255 dmam_declare_coherent_memory()
260 pcim_enable_device() : after success, all PCI ops become managed
261 pcim_pin_device() : keep PCI device enabled after release
267 devm_ioremap_nocache()
269 devm_request_and_ioremap() : checks resource, requests region, ioremaps
272 pcim_iomap_table() : array of mapped addresses indexed by BAR
273 pcim_iomap_regions() : do request_region() and iomap() on multiple BARs