Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <YzQ6pqykLhJVeD2p@infradead.org> <mhng-96bfb7a5-6579-4ebc-9517-db2e62d2f82b@palmer-ri-x1c9>
In-Reply-To: <mhng-96bfb7a5-6579-4ebc-9517-db2e62d2f82b@palmer-ri-x1c9>
From:   Anup Patel <anup@brainfault.org>
Date:   Fri, 7 Oct 2022 11:04:28 +0530
Message-ID: <CAAhSdy15JqNAiuv84OogmoxyxwXE3EJzadfZk-wNcoUUXJ-fJQ@mail.gmail.com>
Subject: Re: [PATCH v2 1/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for
 systems with Svpbmt
To:     Palmer Dabbelt <palmer@dabbelt.com>
Cc:     Christoph Hellwig <hch@infradead.org>, apatel@ventanamicro.com,
        Paul Walmsley <paul.walmsley@sifive.com>,
        atishp@atishpatra.org, heiko@sntech.de,
        linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org,
        mchitale@ventanamicro.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Oct 7, 2022 at 9:20 AM Palmer Dabbelt <palmer@dabbelt.com> wrote:
>
> On Wed, 28 Sep 2022 05:14:30 PDT (-0700), Christoph Hellwig wrote:
> > On Thu, Sep 22, 2022 at 09:35:56AM -0700, Palmer Dabbelt wrote:
> >> Sorry I missed this, I thought it was just part of the rest of this patch
> >> set.  That said, I'm not actually sure this is a critical fix: sure it's a
> >> performance problem, and if some driver is expecting ioremap_cache() to go
> >> fast then possibly a pretty big one, but the only Svpmbt hardware that
> >
> > More importantly nothing should be using ioremap_cache at all.  It is
> > an API that has long been deprecated in favor of memremap.
>
> There's a few uses of it, I just send along a patch to make it
> arch-specific and make the users depend on it.  That should let us stop
> adding this to ports just to get the drivers building.

I agree, ioremap_cache() should not be used by drivers.

I encountered this issue with the PMEM driver which uses devm_memremap()
which in-turn calls memremap() (kernel/iomem.c). The kernel memremap()
still expects arch-specific ioremap_xyz() functions/macros to implement various
MEMREMAP_xyz flags which is why we need these RISC-V specific ioremap_xyz().

An alternative way is to implement RISC-V specific arch_memremap_wb()
but this will still look similar to ioremap_cache() added by this patch. Also,
only 32bit ARM implements arch_memremap_wb() whereas all other archs
(ARM64, x86_64, etc) implement ioremap_xyz() functions/macros.

Regards,
Anup