Re: [PATCH -next] net: hisilicon: Never build on SPARC

Arnd,

On Wed, Oct 21, 2015 at 02:53:20PM -0700, Guenter Roeck wrote:

On Wed, Oct 21, 2015 at 09:11:53PM +0200, Arnd Bergmann wrote:

quoted

On Wednesday 21 October 2015 10:03:05 Guenter Roeck wrote:

quoted

quoted

Something like this?

static inline u64 of_translate_address(struct device_node *np, const __be32 *addr)
{
#if defined(CONFIG_SPARC) || defined(CONFIG_M68K)
      int pna = of_n_addr_cells(np);
      u64 ret = be32_to_cpu(addr[pna - 1]);

      if (pna > 1)
              ret += (u64)be32_to_cpu(addr[pna - 2]) << 32;

      return ret;

That suggests that sparc would need a translation after all, which
seems to contradict what David said earlier.

No, not a translation: the value is used without any offset that
factors in the location of the bus, the above is just the shortest
possible way to read the 64-bit number from a big-endian property
of variable length.

Out of my realm .. David would have to comment on that.

quoted

quoted

Anyway, if it gets that complicated, I think we should stick with
just returning OF_BAD_ADDR. The above really suggests the need for
an architecture specific solution.

Probably no harm in this really: the far more common
of_address_to_resource() and of_iomap() helpers are equally
broken on SPARC and we just return a runtime error for those
as well without CONFIG_OF_ADDRESS rather than breaking the build.

Agreed. Given this, returning OF_BAD_ADDR sounds like a better choice.

Arnd,

do you know if a fix for this problem is pending in some branch ?
Mainline sparc builds are now affected.

Thanks,
Guenter