linux-kernel - Re: [PATCH v3 00/3] MIPS: SGI-IP27 rework part2

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 7 May 2019 17:30:41 +0100
From:   Lorenzo Pieralisi <lorenzo.pieralisi@....com>
To:     Thomas Bogendoerfer <tbogendoerfer@...e.de>
Cc:     Bjorn Helgaas <helgaas@...nel.org>,
        Ralf Baechle <ralf@...ux-mips.org>,
        Paul Burton <paul.burton@...s.com>,
        James Hogan <jhogan@...nel.org>, linux-mips@...r.kernel.org,
        linux-kernel@...r.kernel.org, linux-pci@...r.kernel.org
Subject: Re: [PATCH v3 00/3] MIPS: SGI-IP27 rework part2

On Tue, May 07, 2019 at 05:48:20PM +0200, Thomas Bogendoerfer wrote:
> On Tue, 7 May 2019 16:31:17 +0100
> Lorenzo Pieralisi <lorenzo.pieralisi@....com> wrote:
> 
> > On Thu, Apr 18, 2019 at 03:57:26PM -0500, Bjorn Helgaas wrote:
> > > Hi Thomas,
> > > 
> > > On Tue, Mar 19, 2019 at 04:47:49PM +0100, Thomas Bogendoerfer wrote:
> > > > SGI IP27 (Origin/Onyx2) and SGI IP30 (Octane) have a similair
> > > > architecture and share some hardware (ioc3/bridge). To share
> > > > the software parts this patchset reworks SGI IP27 interrupt
> > > > and pci bridge code. By using features Linux gained during the
> > > > many years since SGI IP27 code was integrated this even results
> > > > in code reduction and IMHO cleaner code.
> > > > 
> > > > Tests have been done on a two module O200 (4 CPUs) and an
> > > > Origin 2000 (8 CPUs).
> > > 
> > > Thanks for doing all this work!  It seems like it basically converts
> > > some of the SGI PCI code to the structure typical of current host
> > > controller drivers and moves it to drivers/pci/controller, which all
> > > seems great to me.
> > 
> > I had a look and the code is really, really MIPS specific, actually
> > I would be interested in understanding how many platforms it supports,
> > it is not even FW configurable.
> 
> it's MIPS only and used in basically 3 different SGI platforms.
> 
> > With hard-coded resources, <asm/...> includes in driver code and MIPS
> > specific kludges even if it does reuse some APIs shared with controller
> > drivers I am not 100% certain that moving it to drivers/pci/controller
> > buys us anything, this is really arch specific code, however we slice
> > it.
> 
> hmm, I thought the idea of having one drivers/pci/controller directory
> is to have all of them in one place.

Well, drivers/pci/controller code should at least try to be arch
agnostic to belong there, there are exceptions, as I said but
they are the exception not the rule.

> > The line between what stays in arch and what goes to
> > drivers/pci/controller is thin but this code is definitely more on the
> > arch side IMHO.
> 
> what makes the xgene driver different from the xtalk-bridge driver ? Ok
> it used DT, but it's still just for a specific type of SOCs from one
> vendor, isn't it ?

Eg no <asm/...> includes required; its code can in principle be
reused on other arches, this series can't, it is arch code IMO.

> > I do not question Thomas' effort, which I appreciate, I question
> > the end result and its usefulness, this series is even increasing
> > lines of kernel code, I would like to see the benefits.
> 
> the move from arch/mips/pci to drivers/pci/controller increases lines of
> code by two lines. The whole patchset adds 155 lines, but also adds
> functionality to be able to use the driver with different platforms.

I am not convinced it will buy you much to be honest, I might be wrong.

If we move the code we have to have a compelling reason for it to
happen and at the moment I do not necessarily see one, happy to
be proven wrong if you extend it to support other platforms cleanly
(eg with firmware bindings) on top of this series.

> Anyway I can live with not moving to drivers/pci/controller if you don't
> like it there.

I can't certainly maintain it, it is too MIPS specific and I do not like
this code either but that's certainly not your fault.

- Hardcoded resources
- No firmware bindings
- Lots of MIPS specific workarounds
- arch code dependency

I do not think that as it stands it belongs in driver/pci/controller,
no.

Thanks,
Lorenzo