Re: [PATCH] sh: pgtable-3level: Fix cast to pointer from integer of different size

From: Rob Landley <rob@landley.net>
To: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>,
	Rich Felker <dalias@libc.org>,
	"Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Jacopo Mondi <jacopo+renesas@jmondi.org>,
	Daniel Palmer <daniel@0x0f.com>,
	Linux-sh list <linux-sh@vger.kernel.org>,
	Linux MM <linux-mm@kvack.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] sh: pgtable-3level: Fix cast to pointer from integer of different size
Date: Sat, 18 Sep 2021 00:04:50 -0500	[thread overview]
Message-ID: <f83c84d1-27d7-978f-4a3a-0c1f5a97c24f@landley.net> (raw)
In-Reply-To: <CAMuHMdUYUbjPSitt3wTi-YY5CEKwDJNJKPp9Pbfewm7B0CXP2A@mail.gmail.com>

On 9/17/21 4:21 AM, Geert Uytterhoeven wrote:
> Hi Rob,
> 
> On Fri, Sep 17, 2021 at 3:12 AM Rob Landley <rob@landley.net> wrote:
>> On 9/15/21 8:50 AM, Geert Uytterhoeven wrote:
>> >     arch/sh/include/asm/pgtable-3level.h:37:9: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
>> ...
>> > The truncation to "unsigned long" has been there since forever, so
>> > probably it still works fine ;-)
>>
>> 1) Linux is LP64 so sizeof(long) and sizeof(pointer) always match, so it's not
>> truncating.
>>
>> 2) The sh5 only ever shipped evaluation units, it never had a production run,
>> and we haven't implemented j64 yet, so all superh targets are currently 32 bit.
>> (I.E. it's complaining about _expanding_ the pointer, which shouldn't be a
>> problem as long as endianness is respected.)
> 
> The build error is for 32-bit.

I know. I'm just saying your fix doesn't seem wrong. (Personally I would have
just cast once to (void *) and let C not being C++ do the right thing, but eh...)

Not sure I understand the point of the original code though. Under what
conditions do you want 64 bit page tables on a 32 bit system? (Some PAE variant?)

> If CONFIG_X2TLB=y, pgd_t.pgd is "unsigned long long", i.e. 64-bit, so
> casting it to a pointer needs an intermediate cast to "unsigned long".
> See arch/sh/include/asm/page.h:

I can't find a user of it. None of the existing defconfigs select this option:

  $ grep -r X2TLB arch/sh/configs

And it's weird in other ways, like 8k page size. (Why...?)

Google finds sh-x2 mentioned in Renesas' sh4a debugger docs, but I've never
played with the renesas "a" variants. The stuff I'm familiar with is all
following up on what Hitachi did, not Renesas.

*shrug* You fix looks good enough to me.

Acked-by: Rob Landley <rob@landley.net>

Rob