[PATCH] gpu: host1x: use %pa to print dma_addr_t

Randy Dunlap rdunlap at infradead.org
Mon Sep 16 17:18:54 PDT 2013

On 09/16/13 17:06, Olof Johansson wrote:
> On Mon, Sep 16, 2013 at 8:54 AM, Joe Perches <joe at perches.com> wrote:
>> On Mon, 2013-09-16 at 08:46 -0700, Olof Johansson wrote:
>>> On Mon, Sep 16, 2013 at 8:17 AM, Thierry Reding
>>> <thierry.reding at gmail.com> wrote:
>>>> On Wed, Sep 11, 2013 at 09:41:49PM -0700, Olof Johansson wrote:
>>>>> This removes two warnings where dma_addr_t variables were printed using
>>>>> %x when built with CONFIG_ARM_LPAE=y, thus having 64-bit dma_addr_t:
>>>>>   drivers/gpu/host1x/hw/cdma_hw.c:57:3: warning: format '%x' expects argument of type 'unsigned int', but argument 5 has type 'dma_addr_t'
>>>>>   drivers/gpu/host1x/hw/debug_hw.c:175:10: warning: format '%x' expects argument of type 'unsigned int', but argument 3 has type 'dma_addr_t'
>>>> Hi Olof,
>>>> I can't reproduce this. Does this perhaps depend on some other patch?
>>>> When I enable LPAE I do see similar warnings in drivers/iommu/tegra-*.c
>>>> and those can indeed be fixed using an equivalent patch.
>>> You need to enable LPAE on a platform that also selects
>>> ARCH_DMA_ADDR_T_64BIT, I don't think tegra does. If you do it with
>>> multi_v7_defconfig you'll see them.
>>> However, see discussion on another of the emails in the series; I'll
>>> have to introduce a new format specifier instead.
>> Or not.
>> I don't know whether or not the dma_addr_t really needs a
>> fixed 18 byte output length for 64 bit uses.
>> I think always using a cast for dma_addr_t addresses like:
>>         printk("dma_addr_t: %#llx\n", (u64)addr);
>> would probably work just fine.
> Sigh. Any color would do. I just want to get rid of the mostly-bogus
> warnings that makes it harder to spot real problems, I really don't
> care how they're resolved.
> None of the affected platforms today use 64-bit DMA anyway, so casting
> down to u32 is equally acceptable. I'll repost with that instead.

Casting to u64 and using %llx is preferred for this throughout the kernel,
not u32.

That way you would never have to 'fix' those when those platforms use
64-bit DMA (this is where you say that they never will :).


More information about the dri-devel mailing list