Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 9 Aug 2010 19:56:49 +0530
From:      "Jayachandran C." <c.jayachandran@gmail.com>
To:        Attilio Rao <attilio@freebsd.org>
Cc:        src-committers@freebsd.org, John Baldwin <jhb@freebsd.org>, "Jayachandran C." <jchandra@freebsd.org>, svn-src-all@freebsd.org, Joe Landers <jlanders@vmware.com>, Randall Stewart <rrs@freebsd.org>, sbruno@freebsd.org, svn-src-head@freebsd.org
Subject:   Re: svn commit: r208165 - in head/sys: kern mips/conf mips/include  mips/mips mips/rmi mips/rmi/dev/xlr
Message-ID:  <AANLkTin9tn%2BOFOkQSYSewtgVQrDNoa2gE2SqPxA73cih@mail.gmail.com>
In-Reply-To: <AANLkTim_pTQgquRRuz4G=jC5G0Amc%2ByF6VeDcg-cpTCX@mail.gmail.com>
References:  <201005161943.o4GJhnTo096839@svn.freebsd.org> <AANLkTikAarFgbxgGu-8XG7gh6VidPoVGwva54NN4rcRF@mail.gmail.com> <AANLkTi=jkUiXStPV3Gq206GgmctM9%2B1ofbNWuh5CSeWw@mail.gmail.com> <AANLkTim_pTQgquRRuz4G=jC5G0Amc%2ByF6VeDcg-cpTCX@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Aug 9, 2010 at 5:35 PM, Attilio Rao <attilio@freebsd.org> wrote:
> 2010/8/9 Jayachandran C. <c.jayachandran@gmail.com>:
>> On Mon, Aug 9, 2010 at 5:31 AM, Attilio Rao <attilio@freebsd.org> wrote:
>>> 2010/5/16 Randall Stewart <rrs@freebsd.org>:
>>>> Author: rrs
>>>> Date: Sun May 16 19:43:48 2010
>>>> New Revision: 208165
>>>> URL: http://svn.freebsd.org/changeset/base/208165
>>>>
>>>> Log:
>>>> =A0This pushes all of JC's patches that I have in place. I
>>>> =A0am now able to run 32 cores ok.. but I still will hang
>>>> =A0on buildworld with a NFS problem. I suspect I am missing
>>>> =A0a patch for the netlogic rge driver.
>>>>
>>>> =A0JC check and see if I am missing anything except your
>>>> =A0core-mask changes
>>>
>>>
>>>> Modified: head/sys/kern/subr_smp.c
>>>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D
>>>> --- head/sys/kern/subr_smp.c =A0 =A0Sun May 16 19:25:56 2010 =A0 =A0 =
=A0 =A0(r208164)
>>>> +++ head/sys/kern/subr_smp.c =A0 =A0Sun May 16 19:43:48 2010 =A0 =A0 =
=A0 =A0(r208165)
>>>> @@ -503,7 +503,10 @@ smp_topo_none(void)
>>>> =A0 =A0 =A0 =A0top =3D &group[0];
>>>> =A0 =A0 =A0 =A0top->cg_parent =3D NULL;
>>>> =A0 =A0 =A0 =A0top->cg_child =3D NULL;
>>>> - =A0 =A0 =A0 top->cg_mask =3D (1 << mp_ncpus) - 1;
>>>> + =A0 =A0 =A0 if (mp_ncpus =3D=3D sizeof(top->cg_mask) * 8)
>>>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 top->cg_mask =3D -1;
>>>> + =A0 =A0 =A0 else
>>>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 top->cg_mask =3D (1 << mp_ncpus) - 1;
>>>> =A0 =A0 =A0 =A0top->cg_count =3D mp_ncpus;
>>>> =A0 =A0 =A0 =A0top->cg_children =3D 0;
>>>> =A0 =A0 =A0 =A0top->cg_level =3D CG_SHARE_NONE;
>>>>
>>>
>>> ... and this is why I particulary hate big commits with complete lack
>>> of technical details.
>>>
>>> This particulary chunk was supposed to fix a nasty and completely MI
>>> bug that some users have already met (kern/148698). The complete lack
>>> of details didn't help in identify the issue neither that it was a
>>> valuable fix.
>>>
>>> The fix is, however, improper (there is no clear relationship between
>>> the multiplication and why that happens) thus I would rather use what
>>> Joe has reported in the PR.
>>
>>
>> I was not aware of the PR when I sent this changes to rrs@, and this
>> went as a part of MIPS SMP support for XLR which has 32 CPUs
>>
>> But I'm not sure that the current change is correct, cg_mask is of
>> type cpumask_t and I don't think it is guaranteed to be 32 bit (as it
>> is a machine dependent type).
>
> Actually the 32 bits limit is well aware and acknowledged in cpumask_t.
> While it is true that it should be an 'opaque' type, in reality it is
> not. The maximum limit of 32 CPUs is a reality due to cpumask_t on all
> our architectures.
> That is why we are going to use cpuset_t for dealing with CPUs numbers
> (which is really opaque, at same extent, and is not bound to any
> size).

In my opinion, your changes added another pitfall for the person who
tries to make the cpumask_t 64 bit to support more cpus, which really
could have been avoided.

But if cpumask_t is going to be changed to cpuset_t soon, I guess we
are fine  :)

JC.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?AANLkTin9tn%2BOFOkQSYSewtgVQrDNoa2gE2SqPxA73cih>