Re: [PATCH 02/20] PCI: fix pci_remap_iospace() remap attribute

From: Bjorn Helgaas <helgaas@kernel.org>
To: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Cc: "Luis R. Rodriguez" <mcgrof@kernel.org>,
	Liviu Dudau <Liviu.Dudau@arm.com>,
	"Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
	Andy Lutomirski <luto@amacapital.net>,
	linux-pci@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
	Wenrui Li <wenrui.li@rock-chips.com>,
	Gabriele Paoloni <gabriele.paoloni@huawei.com>,
	Catalin Marinas <catalin.marinas@arm.com>,
	Shawn Lin <shawn.lin@rock-chips.com>,
	Will Deacon <will.deacon@arm.com>,
	Michal Simek <michal.simek@xilinx.com>,
	Thierry Reding <thierry.reding@gmail.com>,
	Tanmay Inamdar <tinamdar@apm.com>,
	linux-arch@vger.kernel.org,
	Pratyush Anand <pratyush.anand@gmail.com>,
	Russell King <linux@armlinux.org.uk>,
	Jon Mason <jonmason@broadcom.com>,
	Murali Karicheri <m-karicheri2@ti.com>,
	Arnd Bergmann <arnd@arndb.de>,
	Bharat Kumar Gogada <bharat.kumar.gogada@xilinx.com>,
	Ray Jui <rjui@broadcom.com>, John Garry <john.garry@huawei.com>,
	Joao Pinto <Joao.Pinto@synopsys.com>,
	Bjorn Helgaas <bhelgaas@google.com>,
	Mingkai Hu <mingkai.hu@freescale.com>,
	Thomas Petazzoni <thomas.petazzoni@free-electrons.com>,
	Jingoo Han <jingoohan1@gmail.com>,
	linux-kernel@vger.kernel.org,
	Stanimir Varbanov <svarbanov@mm-sol.com>,
	Minghuan Lian <minghuan.Lian@freescale.com>,
	Zhou Wang <wangzhou1@hisilicon.com>,
	Roy Zang <tie-fei.zang@freescale.com>
Subject: Re: [PATCH 02/20] PCI: fix pci_remap_iospace() remap attribute
Date: Mon, 20 Mar 2017 11:38:43 -0500	[thread overview]
Message-ID: <20170320163843.GD24572@bhelgaas-glaptop.roam.corp.google.com> (raw)
In-Reply-To: <20170320162627.GE7632@red-moon>

On Mon, Mar 20, 2017 at 04:26:27PM +0000, Lorenzo Pieralisi wrote:
> On Mon, Mar 20, 2017 at 11:06:36AM -0500, Bjorn Helgaas wrote:
> 
> [...]
> 
> > > > > @@ -3375,7 +3375,7 @@ int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
> > > > >  		return -EINVAL;
> > > > >  
> > > > >  	return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
> > > > > -				  pgprot_device(PAGE_KERNEL));
> > > > > +				  pgprot_noncached(PAGE_KERNEL));
> > > > 
> > > > ...
> > > > I do find this puzzling because I naively expected pgprot_noncached()
> > > > to match up with ioremap_nocache(), and apparently it doesn't.
> > > > 
> > > > For example, ARM64 ioremap_nocache() uses PROT_DEVICE_nGnRE, which
> > > > doesn't match the MT_DEVICE_nGnRnE in pgprot_noncached().
> > > > 
> > > > The point of these patches is to use non-posted mappings.  Apparently
> > > > you can do that with pgprot_noncached() here, but ioremap_nocache()
> > > > isn't enough for the config space mappings?
> > > 
> > > This is for iospace it seems, so the other patch I think was for
> > > config space.
> > 
> > I understand that 02/20 is for PCI I/O port space and 03/20 is for PCI
> > config space.  I'm confused because I thought we wanted the same
> > non-posted mapping for both, but looks like they're different.
> > 
> > Patch 02/20 uses ioremap_page_range(..., pgprot_noncached(PAGE_KERNEL))
> > to map PCI I/O port space:
> > 
> >   #define pgprot_noncached(prot) \
> >           __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
> > 
> > Patch 03/20 uses ioremap_nocache() to map PCI config space:
> > 
> >   #define ioremap_nocache(addr, size)     __ioremap((addr), (size), __pgprot(PROT_DEVICE_nGnRE))
> > 
> > So the I/O port mapping uses MT_DEVICE_nGnRnE, while the config space
> > mapping uses PROT_DEVICE_nGnRE, which looks different.
> 
> On ARM64 (PATCH 4) and ARM (PATCH 5) we override pci_remap_cfgspace()
> with implementations that provide non-posted writes bus attributes,
> PATCH 3 is just there to provide a "safe" (well, I need input on that)
> fall-back.

Ah, thanks, that makes sense to me finally!  After patch 4,
pci_remap_cfgspace() uses PROT_DEVICE_nGnRnE, which does look like the
MT_DEVICE_nGnRnE used in the ioremap_page_range() path.

I wonder if we can get rid of pgprot_device() altogether?  Patch 2
removed the use in pci_remap_iospace(), and patch 18 removes the use
in tegra_pcie_bus_alloc().

Bjorn