On Mon, Jul 9, 2018 at 4:04 PM Mark Rutland mark.rutland@arm.com wrote:
On Mon, Jul 09, 2018 at 03:54:09PM +0200, Peter Zijlstra wrote:
On Mon, Jul 09, 2018 at 04:45:50PM +0300, Alexey Brodkin wrote:
diff --git a/drivers/base/devres.c b/drivers/base/devres.c index f98a097e73f2..d65327cb83c9 100644 --- a/drivers/base/devres.c +++ b/drivers/base/devres.c @@ -24,8 +24,12 @@ struct devres_node {
struct devres { struct devres_node node;
- /* -- 3 pointers */
- unsigned long long data[]; /* guarantee ull alignment */
- /*
- data[] must be 64 bit aligned even on 32 bit architectures
- because it might be accessed by instructions that require
- aligned memory arguments such as atomic64_t.
- */
- u8 __aligned(8) data[];
};
From a quick reading in Documentation/driver-model/devres.txt this devres muck is supposed to be device memory, right?
It's for associating resources (e.g. memory allocations) with a struct device.
e.g. you do:
devm_kmalloc(dev, size, GFP_KERNEL);... and that allocates sizeof(struct devres) + size, putting some accounting data into that devres, and returning a pointer to the remaining size bytes.
The data[] thing is a hack to ensure that the structure is padded to 64-bit alignment, in case you'd done:
struct foo { atomic64_t counter; }
struct foo *f = devm_kmalloc(dev, sizeof(*f), GFP_KERNEL);
So the big issue is that the minimum alignment of a buffer allocated with devm_kmalloc() and friends is different (lower) than when allocated with kmalloc().
On 32-bit, it's only aligned to 4 bytes. Ugh. I wouldn't be surprised if some callers assume it to be cacheline-aligned...
Which means blind conversions to the devm_*() versions can be dangerous.
Gr{oetje,eeting}s,
Geert