On Thu, 26 Jul 2012 17:35:13 +0800
Xiao Guangrong [off-list ref] wrote:

quoted

Is this patch really safe for all architectures?

IS_ERR_VALUE() casts -MAX_ERRNO to unsigned long and then does comparison.
Isn't it possible to conflict with valid pfns?

See IS_ERR_VALUE():

#define IS_ERR_VALUE(x) unlikely((x) >= (unsigned long)-MAX_ERRNO)

The minimal value of the error code is:
0xffff f001 on 32-bit and 0x ffff ffff ffff f001 on 64-bit,
it is fair larger that a valid pfn (for the pfn, the most top of 12 bits
are always 0).

Note, PAE is a special case, but only 64G physical memory is valid,
0xffff f001 is also suitable for that.

Ah, I see.  I misread the type pfn_t and was confused.
Thank you!

	Takuya