Received-SPF: pass (google.com: domain of linux-kernel+bounces-198713-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
From: =?UTF-8?q?Ilpo=20J=C3=A4rvinen?= <ilpo.jarvinen@linux.intel.com>
Date: Mon, 3 Jun 2024 10:28:27 +0300 (EEST)
To: Chia-I Wu <olvaffe@gmail.com>
cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>, 
    amd-gfx@lists.freedesktop.org, dri-devel@lists.freedesktop.org, 
    christian.koenig@amd.com, alexander.deucher@amd.com, 
    Greg Kroah-Hartman <gregkh@linuxfoundation.org>, 
    Alison Schofield <alison.schofield@intel.com>, 
    Dave Jiang <dave.jiang@intel.com>, Baoquan He <bhe@redhat.com>, 
    LKML <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] kernel/resource: optimize find_next_iomem_res
In-Reply-To: <CAPaKu7SsD+X7KAO=3vEYU_7YGM_f+7k1fdC9nEK=-NaJw8oYaA@mail.gmail.com>
Message-ID: <492bba9e-2e22-e6dd-e2e3-c218ec413c3e@linux.intel.com>
References: <20240531053704.2009827-1-olvaffe@gmail.com> <ZlmQ3_wcL3cgp4Hb@smile.fi.intel.com> <CAPaKu7SsD+X7KAO=3vEYU_7YGM_f+7k1fdC9nEK=-NaJw8oYaA@mail.gmail.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="8323328-2104200125-1717399707=:1529"

  This message is in MIME format.  The first part should be readable text,
  while the remaining parts are likely unreadable without MIME-aware tools.

--8323328-2104200125-1717399707=:1529
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: QUOTED-PRINTABLE

On Fri, 31 May 2024, Chia-I Wu wrote:
> On Fri, May 31, 2024 at 1:57=E2=80=AFAM Andy Shevchenko <andriy.shevchenk=
o@linux.intel.com>
> wrote:
>       On Thu, May 30, 2024 at 10:36:57PM -0700, Chia-I Wu wrote:
>       > We can skip children resources when the parent resource does not =
cover
>       > the range.
>       >
>       > This should help vmf_insert_* users on x86, such as several DRM
>       drivers.
>       > On my AMD Ryzen 5 7520C, when streaming data from cpu memory into
>       amdgpu
>       > bo, the throughput goes from 5.1GB/s to 6.6GB/s.=C2=A0 perf repor=
t says
>       >
>       >=C2=A0 =C2=A034.69%--__do_fault
>       >=C2=A0 =C2=A034.60%--amdgpu_gem_fault
>       >=C2=A0 =C2=A034.00%--ttm_bo_vm_fault_reserved
>       >=C2=A0 =C2=A032.95%--vmf_insert_pfn_prot
>       >=C2=A0 =C2=A025.89%--track_pfn_insert
>       >=C2=A0 =C2=A024.35%--lookup_memtype
>       >=C2=A0 =C2=A021.77%--pat_pagerange_is_ram
>       >=C2=A0 =C2=A020.80%--walk_system_ram_range
>       >=C2=A0 =C2=A017.42%--find_next_iomem_res
>       >
>       > before this change, and
>       >
>       >=C2=A0 =C2=A026.67%--__do_fault
>       >=C2=A0 =C2=A026.57%--amdgpu_gem_fault
>       >=C2=A0 =C2=A025.83%--ttm_bo_vm_fault_reserved
>       >=C2=A0 =C2=A024.40%--vmf_insert_pfn_prot
>       >=C2=A0 =C2=A014.30%--track_pfn_insert
>       >=C2=A0 =C2=A012.20%--lookup_memtype
>       >=C2=A0 =C2=A09.34%--pat_pagerange_is_ram
>       >=C2=A0 =C2=A08.22%--walk_system_ram_range
>       >=C2=A0 =C2=A05.09%--find_next_iomem_res
>       >
>       > after.
>=20
>       Is there any documentation that explicitly says that the children
>       resources
>       must not overlap parent's one? Do we have some test cases? (Either =
way
>       they
>       needs to be added / expanded).
>=20
> I think it's the opposite.=C2=A0 The assumption here is that a child=C2=
=A0is always a subset of
> its parent.=C2=A0 Thus, if the range to be checked is not covered by a pa=
rent, we can skip
> the children.
>
> That's guaranteed by=C2=A0__request_resource.=C2=A0 I am less sure about=
=C2=A0__insert_resource but
> it appears to be the case too.=C2=A0 FWIW, resource_is_exclusive has the =
same assumption
> already.

Yes, the children resources are contained within the parent resource (at=20
least in PCI but given the code, I'd expect that to be general state of=20
affairs).

> It looks like I need to do some refactoring to add tests.
>=20
>=20
>       P.S> I'm not so sure about this change. It needs a thoroughly testi=
ng,
>       esp.
>       in PCI case. Cc'ing to Ilpo.
>=20
> What's special about PCI?

--=20
 i.

--8323328-2104200125-1717399707=:1529--