Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 09 Jan 2012 21:13:42 +0200
From:      Alexander Motin <mav@FreeBSD.org>
To:        Maksim Yevmenkin <emax@freebsd.org>
Cc:        svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org
Subject:   Re: svn commit: r228939 - head/sys/dev/mps
Message-ID:  <4F0B3C66.6020701@FreeBSD.org>
In-Reply-To: <CAFPOs6rF1iSDMPdaXCbC_QXsEmqF5iqN3ZNcJkg5AFj15kmGCA@mail.gmail.com>
References:  <201112282249.pBSMnTZu028304@svn.freebsd.org> <CAFPOs6oaMNsHP9VFtL-Szs=vDQj-Ss=JKCECqsUNTMXi7UwNHw@mail.gmail.com> <4F0B38B9.1020302@FreeBSD.org> <CAFPOs6rF1iSDMPdaXCbC_QXsEmqF5iqN3ZNcJkg5AFj15kmGCA@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On 09.01.2012 21:01, Maksim Yevmenkin wrote:
> 2012/1/9 Alexander Motin<mav@freebsd.org>:
>> On 09.01.2012 20:54, Maksim Yevmenkin wrote:
>>>
>>> On Wed, Dec 28, 2011 at 2:49 PM, Alexander Motin<mav@freebsd.org>    wrote:
>>>>
>>>> Author: mav
>>>> Date: Wed Dec 28 22:49:28 2011
>>>> New Revision: 228939
>>>> URL: http://svn.freebsd.org/changeset/base/228939
>>>>
>>>> Log:
>>>>   Set maximum I/O size for mps(4) to MAXPHYS. Looking into the code, I see
>>>>   no reason why it should be limited to 64K of DFLTPHYS. DMA data tag is
>>>> any
>>>>   way set to allow MAXPHYS, S/G lists (chain elements) are sufficient and
>>>>   overflows are also handled. On my tests even 1MB I/Os are working fine.
>>>>
>>>>   Reviewed by:  ken@
>>>>
>>>> Modified:
>>>>   head/sys/dev/mps/mps_sas.c
>>>>
>>>> Modified: head/sys/dev/mps/mps_sas.c
>>>>
>>>> ==============================================================================
>>>> --- head/sys/dev/mps/mps_sas.c  Wed Dec 28 22:18:53 2011        (r228938)
>>>> +++ head/sys/dev/mps/mps_sas.c  Wed Dec 28 22:49:28 2011        (r228939)
>>>> @@ -937,6 +937,7 @@ mpssas_action(struct cam_sim *sim, union
>>>>                 cpi->transport_version = 0;
>>>>                 cpi->protocol = PROTO_SCSI;
>>>>                 cpi->protocol_version = SCSI_REV_SPC;
>>>> +               cpi->maxio = MAXPHYS;
>>>>                 cpi->ccb_h.status = CAM_REQ_CMP;
>>>>                 break;
>>>>         }
>>>
>>>
>>> sorry for the late reply, but can we make in into tunable please? i
>>> have in local tree
>>>
>>> --- mps_sas.c.orig      2011-11-17 02:05:04.000000000 -0800
>>> +++ mps_sas.c   2011-12-28 16:05:10.000000000 -0800
>>> @@ -121,6 +121,11 @@
>>>
>>>   MALLOC_DEFINE(M_MPSSAS, "MPSSAS", "MPS SAS memory");
>>>
>>> +int mps_maxio = MAXPHYS;
>>> +TUNABLE_INT("hw.mps.maxio",&mps_maxio);
>>> +SYSCTL_INT(_hw_mps, OID_AUTO, maxio, CTLFLAG_RD,&mps_maxio, 0,
>>>
>>> +       "CAM maxio override\n");
>>> +
>>>   static __inline int mpssas_set_lun(uint8_t *lun, u_int ccblun);
>>>   static struct mpssas_target * mpssas_alloc_target(struct mpssas_softc *,
>>>       struct mpssas_target *);
>>> @@ -938,6 +943,7 @@
>>>                 cpi->protocol = PROTO_SCSI;
>>>                 cpi->protocol_version = SCSI_REV_SPC;
>>>                 cpi->ccb_h.status = CAM_REQ_CMP;
>>> +               cpi->maxio = mps_maxio;
>>>                 break;
>>>         }
>>>         case XPT_GET_TRAN_SETTINGS:
>>
>>
>> We can. but could you explain why? Have you found any problems this change?
>
> not really. i've had this patch in the local tree for a while now. we
> are experimenting with various MAXPHYS/maxio settings and having this
> tunable is very handy. basically, we can set MAXPHYS to some larger
> value and tweak maxio (for testing purposes) without
> recompiling/installing new kernel.

I don't really think that it is perfect place for such tunable. It is a 
bit strange IMHO to have different maxio for different types of HBAs 
except physical limitations. I would prefer it to be configurable on 
above layers, for example, file systems, if needed. But if you need it 
here for something, I won't object against adding it.

Have you found any benefits of having maxio below MAXPHYS while 
experimenting? May be those results could be used to improve FS behavior 
somehow to make tuning not needed?

-- 
Alexander Motin



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4F0B3C66.6020701>