mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: "Arnd Bergmann" <arnd@arndb.de>
To: "Julian Vetter" <jvetter@kalrayinc.com>,
	"Catalin Marinas" <catalin.marinas@arm.com>,
	"Will Deacon" <will@kernel.org>, guoren <guoren@kernel.org>,
	"Huacai Chen" <chenhuacai@kernel.org>,
	"WANG Xuerui" <kernel@xen0n.name>,
	"Andrew Morton" <akpm@linux-foundation.org>
Cc: linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org,
	"linux-csky@vger.kernel.org" <linux-csky@vger.kernel.org>,
	loongarch@lists.linux.dev,
	"Yann Sionneau" <ysionneau@kalrayinc.com>
Subject: Re: [PATCH 1/4] Consolidate __memcpy_{to,from}io and __memset_io into a single lib
Date: Mon, 09 Sep 2024 10:35:58 +0000	[thread overview]
Message-ID: <3c56dbee-241c-4742-a586-254e10b98cbb@app.fastmail.com> (raw)
In-Reply-To: <20240906114151.519028-2-jvetter@kalrayinc.com>

On Fri, Sep 6, 2024, at 11:41, Julian Vetter wrote:
> Various architectures have almost the same implementations for
> __memcpy_{to,from}io and __memset_io functions. So, consolidate them and
> introduce a CONFIG_GENERIC_IO flag to build the given lib/io.c.
>
> Reviewed-by: Yann Sionneau <ysionneau@kalrayinc.com>
> Signed-off-by: Julian Vetter <jvetter@kalrayinc.com>

Thanks for working on this. The implementation looks correct
to me and we should be able to use this on most architectures,
but since this is a shared file, it would be good to make it
as polished as possible.

>  lib/Kconfig  |   3 ++
>  lib/Makefile |   2 +
>  lib/io.c     | 101 +++++++++++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 106 insertions(+)
>  create mode 100644 lib/io.c

I feel the name is a little too geneal, maybe use io_copy.c and
CONFIG_GENERIC_IO_COPY for the name?

Alternatively, this could be part of lib/iomap_copy.c,
which already has some related helper functions. In that
case, I think we would rely on the per-architecture "#define
__memcpy_fromio __memcpy_fromio" etc macros instead of a
Kconfig symbol.

> +void __memcpy_fromio(void *to, const volatile void __iomem *from, size_t count)
> +{
> +	while (count && !IS_ALIGNED((unsigned long)from, NATIVE_STORE_SIZE)) {
> +		*(u8 *)to = __raw_readb(from);
> +		from++;
> +		to++;
> +		count--;
> +	}

There is a corner case that this bit actually misses (same in the
existing implementation): If the __iomem buffer is aligned, but
the 'to' buffer is not, this may cause an alignment fault on
architectures without native unaligned stores. At the moment
I think both csky and some configurations of loongarch64 can
run into this. I think the best way to deal with this is to
use get_unaligned()/put_unaligned() in place of the pointer
dereference. This has no effect on architectures that have
native unaligned access but otherwise uses byte access on the
kernel buffer, which I think is a good tradeoff.

> +	while (count >= NATIVE_STORE_SIZE) {
> +		*(uintptr_t *)to = __raw_read_native(from);
> +		from += NATIVE_STORE_SIZE;
> +		to += NATIVE_STORE_SIZE;
> +		count -= NATIVE_STORE_SIZE;
> +	}

The use of __raw_read_native() somehow bothers me, it seems
a little more complicated than open-coding the two
possibles paths, or even using the aligned-access
helpers like

      if (IS_ENABLED(CONFIG_64BIT))
             __iowrite64_copy(to, from, count & WORD_MASK)
      else
             __iowrite32_copy(to, from, count & WORD_MASK)

Those helpers do require the kernel buffer to be naturally
aligned though.

> +void __memset_io(volatile void __iomem *dst, int c, size_t count)
> +{
> +	uintptr_t qc = (u8)c;
> +
> +	qc |= qc << 8;
> +	qc |= qc << 16;
> +#if IS_ENABLED(CONFIG_64BIT)
> +	qc |= qc << 32;
> +#endif

If you use IS_ENABLED() here, please do it like

       if (IS_ENABLED(CONFIG_64BIT)
              qc |= qc << 32;

      Arnd

  reply	other threads:[~2024-09-09  8:36 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-09-06 11:41 [PATCH 0/4] Consolidate IO memcpy functions Julian Vetter
2024-09-06 11:41 ` [PATCH 1/4] Consolidate __memcpy_{to,from}io and __memset_io into a single lib Julian Vetter
2024-09-09 10:35   ` Arnd Bergmann [this message]
2024-09-06 11:41 ` [PATCH 2/4] Activate GENERIC_IO for the arm64 architecture Julian Vetter
2024-09-06 11:41 ` [PATCH 3/4] Activate GENERIC_IO for the csky architecture Julian Vetter
2024-09-09  5:38   ` Guo Ren
2024-09-06 11:41 ` [PATCH 4/4] Activate GENERIC_IO for the loongarch architecture Julian Vetter

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=3c56dbee-241c-4742-a586-254e10b98cbb@app.fastmail.com \
    --to=arnd@arndb.de \
    --cc=akpm@linux-foundation.org \
    --cc=catalin.marinas@arm.com \
    --cc=chenhuacai@kernel.org \
    --cc=guoren@kernel.org \
    --cc=jvetter@kalrayinc.com \
    --cc=kernel@xen0n.name \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-csky@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=loongarch@lists.linux.dev \
    --cc=will@kernel.org \
    --cc=ysionneau@kalrayinc.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®