On Tue, Feb 07, 2017 at 08:10:14PM -0800, Dan Williams wrote:
On Tue, Feb 7, 2017 at 7:51 PM, Xiong Zhou <xzhou(a)redhat.com>
> On Fri, Jan 20, 2017 at 12:40:07PM +0800, Xiong Zhou wrote:
>> At first, I am not sure whether this is an issue.
>> mmap a file in a DAX mountpoint, open another file
>> in a non-DAX mountpoint with O_DIRECT, write the
>> mapped area to the other file.
>> This write Success on pmem ramdisk(memmap=2G!20G like)
>> This write Fail(Bad address) on nvdimm pmem devices.
>> This write Fail(Bad address) on brd based ramdisk.
>> If we skip the O_DIRECT flag, all tests pass.
>> If we write from DAX to DAX, all tests pass.
>> If we write from non-DAX to DAX, all tests pass.
> To falloc instead of pwrite while initiating test files,
> ( Thanks Ross! :)
> the write call returned success, however the following
> read back to mmaped area FAILED the same way:
> return (Bad address) on raw-mode nvdimm device;
> return (Success) on memory-mode nvdimm device;
> return (Bad address) on brd based ramdisk.
> Also, this only happends with O_DIRECT flag on.
> This smells like an issue to me, still looking into why
> read can't get that page..
Why does it smell like an issue? Any path that calls get_user_pages()
Because the write call gets its page and succeeds, while read back fails.
__get_user_pages on the same address first pass, then fail.
and a DAX mmap range will fail without "memory" mode.
I'm of the
opinion that we should disable DAX altogether for raw-mode.