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;
Date:   Thu, 27 Jul 2023 18:21:24 +0800
From:   Baoquan He <bhe@redhat.com>
To:     Arnd Bergmann <arnd@arndb.de>
Cc:     Arnd Bergmann <arnd@kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] iomem: remove __weak ioremap_cache helper
Message-ID: <ZMJFJIZrf5tRANLm@MiWiFi-R3L-srv>
References: <20230726145432.1617809-1-arnd@kernel.org>
 <ZMG4kw3k1cCGTL2L@MiWiFi-R3L-srv>
 <95ce8aa7-46fc-405c-acec-e94e8f93d186@app.fastmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <95ce8aa7-46fc-405c-acec-e94e8f93d186@app.fastmail.com>
Precedence: bulk

On 07/27/23 at 10:18am, Arnd Bergmann wrote:
> On Thu, Jul 27, 2023, at 02:21, Baoquan He wrote:
> > On 07/26/23 at 04:54pm, Arnd Bergmann wrote:
> >> From: Arnd Bergmann <arnd@arndb.de>
> >> 
> >> No portable code calls into this function any more, and on
> >> architectures that don't use or define their own, it causes
> >> a warning:
> >> 
> >> kernel/iomem.c:10:22: warning: no previous prototype for 'ioremap_cache' [-Wmissing-prototypes]
> >>    10 | __weak void __iomem *ioremap_cache(resource_size_t offset, unsigned long size)
> >> 
> >> Fold it into the only caller that uses it on architectures
> >> without the #define.
> >> 
> >> Note that the fallback to ioremap is probably still wrong on
> >> those architectures, but this is what it's always done there.
> >
> > Do we need to add a definition of ioremap_cache in asm-generic/io.h like
> > ioremap_wc|wt?
> >
> > #ifndef ioremap_cache
> > #define ioremap_cahce ioremap
> > #endif
> 
> No, we really want to eliminate ioremap_cache() from the API,
> not make it more visible.

Got it now, thanks a lot for explanation.

> 
> > Unless it's for sure that drivers calling ioremap_cache are only built in
> > on those architecures defining it. Or we just want to see the breakage
> > on those ARCH-es so that they will add their own definition.
> 
> Right now, it's only possible to call ioremap_cache() on the couple
> of architectures that explicitly declare this in their asm/io.h header
> (arm, arm64, ia64, loongarch, mips, powerpc, sh, x86, and xtensa).
> 
> I think we could also go ahead and remove it from sh and xtensa.

Agree, if no actual need of ioremap_cache() on sh and xtensa.

> 
> > Not sure if I missed anything when understanding this.
> >
> > drivers/acpi/apei/bert.c:       boot_error_region = 
> > ioremap_cache(bert_tab->address, region_len);
> > drivers/acpi/apei/einj.c:       trigger_tab = 
> > ioremap_cache(trigger_paddr, sizeof(*trigger_tab));
> > drivers/acpi/apei/einj.c:       trigger_tab = 
> > ioremap_cache(trigger_paddr, table_size);
> > drivers/acpi/apei/erst.c:       erst_erange.vaddr = 
> > ioremap_cache(erst_erange.base,
> > include/acpi/acpi_io.h:       return ioremap_cache(phys, size);
> 
> acpi is used on x86, arm64 ia64, and loongarch
> 
> > drivers/firmware/meson/meson_sm.c:      return 
> > ioremap_cache(sm_phy_base, size);
> 
> arm/arm64 specific
> 
> > drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c:                
> > adev->mman.aper_base_kaddr = ioremap_cache(adev->gmc.aper_base,
> > drivers/gpu/drm/hyperv/hyperv_drm_drv.c:        hv->vram = 
> > ioremap_cache(hv->mem->start, hv->fb_size);
> > drivers/gpu/drm/ttm/ttm_bo_util.c:                      map->virtual = 
> > ioremap_cache(res, size);
> > drivers/gpu/drm/ttm/ttm_bo_util.c:                      vaddr_iomem = 
> > ioremap_cache(mem->bus.offset,
> 
> All guardeded with an x86 check
> 
> > drivers/hv/hv.c:                /* Mask out vTOM bit. ioremap_cache() 
> > maps decrypted */
> > drivers/hv/hv.c:                        = (void *)ioremap_cache(base, 
> > HV_HYP_PAGE_SIZE);
> > drivers/hv/hv.c:                /* Mask out vTOM bit. ioremap_cache() 
> > maps decrypted */
> > drivers/hv/hv.c:                        = (void *)ioremap_cache(base, 
> > HV_HYP_PAGE_SIZE);
> > drivers/video/fbdev/hyperv_fb.c:        fb_virt = 
> > ioremap_cache(par->mem->start, screen_fb_size);
> 
> x86 and arm64 specific
> 
> > drivers/mtd/devices/bcm47xxsflash.c:            b47s->window = 
> > ioremap_cache(res->start, resource_size(res));
> > drivers/mtd/maps/pxa2xx-flash.c:        info->map.cached = 
> > ioremap_cache(info->map.phys, info->map.size);
> > drivers/soc/fsl/qbman/qman_ccsr.c:      void __iomem *tmpp = 
> > ioremap_cache(addr, sz);
> 
> arm/arm64 specific.
> 
> There is also one more caller in mips that gets used by dmi:
> 
> arch/mips/include/asm/dmi.h:#define dmi_early_remap(x, l)		ioremap_cache(x, l)
> arch/mips/include/asm/dmi.h:#define dmi_remap(x, l)			ioremap_cache(x, l)
> 
>       Arnd
>