lkml.org 
[lkml]   [2012]   [Sep]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC PATCH 0/3] target: try satisfying memory requests with higher-order allocations
Il 06/09/2012 03:58, Nicholas A. Bellinger ha scritto:
>> This patch series fixes this problem by using higher-order allocations
>> to build the data scatterlist. The problem is that iscsi assumes that the
>> scatterlist consists of single pages, which is not true anymore. So
>> patch 2 has to introduce some relatively complicated changes to
>> iscsi_map_iovec and iscsi_unmap_iovec.
>
> So enabling multi-page per SGL support is a feature that has been
> dormant within target core for a long time. It's about time that we
> start taking advantage of it again. ;)

Yeah, I noticed some preparation for it in tcm_fc/tfc_io.c, though too
late (they look a lot like my iscsi changes, it would have saved me some
time!).

While this is obviously not to be taken lightly, I disagree with making
this a per-fabric choice. With a properly organized (and bisectable)
series, it should be relatively easy to review and to get right. I
looked a bit more closely now and there are no changes needed to other
targets (actually there is a change needed in tcm_qla2xxx, but the code
is currently disabled).

There are however changes to transport_kmap_data_sg needed and a few
other places.

I definitely agree with your other comments, including making max_order
a DEF_DEV_ATTRIB. In addition, the default max_order should be capped
based on queue_max_sectors(q) if applicable, to avoid hitting this scenario:

/*
* XXX: if the length the device accepts is shorter than the
* length of the S/G list entry this will cause and
* endless loop. Better hope no driver uses huge pages.
*/

Paolo

>> While doing this, I noticed something strange in iscsit_do_crypto_hash_sg.
>> Patch 1 adds a warning about it.
>>
>
> Mmmmm, looks like a separate bug with DataDigest enabled.
>
>> The approach may be completely wrong and it needs more testing anyway.
>> Please review!
>>
>
> Adding my comments inline.
>
> Thanks Paolo!
>
> --nab
>
>> Paolo
>>
>> Paolo Bonzini (3):
>> tcm_iscsi: warn on incorrect precondition for iscsit_do_crypto_hash_sg
>> tcm_iscsi: support multiple sizes in the scatterlist
>> target: try satisfying memory requests with contiguous blocks
>>
>> drivers/target/iscsi/iscsi_target.c | 106 +++++++++++++++++++++++++-----
>> drivers/target/iscsi/iscsi_target_core.h | 2 +-
>> drivers/target/target_core_transport.c | 58 ++++++++++++++---
>> 3 files changed, 138 insertions(+), 28 deletions(-)
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe target-devel" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>
>



\
 
 \ /
  Last update: 2012-09-06 12:23    [W:0.060 / U:0.720 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site