Re: [PATCH] [31/31] HWPOISON: Add a madvise() injector for soft page offlining

[Michael Kerrisk]

On Sun, Jun 27, 2010 at 1:30 AM, Wu Fengguang <fengguang.wu@xxxxxxxxx> wrote:
> On Sat, Jun 26, 2010 at 09:18:52PM +0800, Michael Kerrisk wrote:
>> Hi Fengguang,
>>
>> On Sun, Jun 20, 2010 at 9:14 AM, Wu Fengguang <fengguang.wu@xxxxxxxxx> wrote:
>> > On Sun, Jun 20, 2010 at 02:19:35PM +0800, Michael Kerrisk wrote:
>> >> Hi Andi,
>> >> On Sat, Jun 19, 2010 at 9:52 PM, Andi Kleen <andi@xxxxxxxxxxxxxx> wrote:
>> >> >> .TP
>> >> >> .BR MADV_SOFT_OFFLINE " (Since Linux 2.6.33)
>> >> >> Soft offline the pages in the range specified by
>> >> >> .I addr
>> >> >> and
>> >> >> .IR length .
>> >> >> This memory of each page in the specified range is copied to a new page,
>> >> >
>> >> > Actually there are some cases where it's also dropped if it's cached page.
>> >> >
>> >> > Perhaps better would be something more fuzzy like
>> >> >
>> >> > "the contents are preserved"
>> >>
>> >> The problem to me is that this gets so fuzzy that it's hard to
>> >> understand the meaning (I imagine many readers will ask: "What does it
>> >> mean that the contents are preserved"?). Would you be able to come up
>> >> with a wording that is a little miore detailed?
>> >
>> > That is, MADV_SOFT_OFFLINE won't lose data.
>> >
>> > If a process writes "1" to some virtual address and then called
>> > madvice(MADV_SOFT_OFFLINE) on that virtual address, it can continue
>> > to read "1" from that virtual address.
>> >
>> > MADV_SOFT_OFFLINE "transparently" replaces the underlying physical page
>> > frame with a new one that contains the same data "1". The original page
>> > frame is offlined, and the new page frame may be installed lazily.
>>
>> Thanks. That helps me come up with a description that is I think a bit clearer:
>>
>>        MADV_SOFT_OFFLINE (Since Linux 2.6.33)
>>               Soft offline the pages in the range specified by
>>               addr and length.  The memory of each page in the
>>               specified  range  is  preserved (i.e., when next
>>               accessed, the same content will be visible,  but
>>               in  a new physical page frame), and the original
>>               page is offlined  (i.e.,  no  longer  used,  and
>>               taken  out  of  normal  memory management).  The
>>               effect of  the  MADV_SOFT_OFFLINE  operation  is
>>               invisible  to  (i.e., does not change the seman-
>>               tics of) the calling process. ...
>>
>> The actual patch for man-pages-3.26 is below.
>
> Thanks. The change looks good to me.

Thanks for checking it.

> Note that the other perceivable change may be a little access delay.
> The kernel could choose to simply drop the in-memory data when there
> is another copy in disk. When accessed again, the content for the new
> physical page will be populated from disk IO.

Yes, I'd suposed as much, but decided that was a detail that probably
didm\t need to be mentioned in tha man page.

Thanks,

Michael


>>
>> --- a/man2/madvise.2
>> +++ b/man2/madvise.2
>> @@ -163,12 +163,14 @@ Soft offline the pages in the range specified by
>>  .I addr
>>  and
>>  .IR length .
>> -The memory of each page in the specified range is copied to a new page,
>> +The memory of each page in the specified range is preserved
>> +(i.e., when next accessed, the same content will be visible,
>> +but in a new physical page frame),
>>  and the original page is offlined
>>  (i.e., no longer used, and taken out of normal memory management).
>>  The effect of the
>>  .B MADV_SOFT_OFFLINE
>> -operation is normally invisible to (i.e., does not change the semantics of)
>> +operation is invisible to (i.e., does not change the semantics of)
>>  the calling process.
>>  This feature is intended for testing of memory error-handling code;
>>  it is only available if the kernel was configured with
>



-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Author of "The Linux Programming Interface" http://blog.man7.org/
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/