linux-kernel - Re: [BISECTED] Regression: Solidrun Clearfog Base won't boot since "PCI: mvebu: Only remap I/O space if configured"

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 13 Sep 2018 09:45:15 +0200
From:   Thomas Petazzoni <thomas.petazzoni@...tlin.com>
To:     Russell King - ARM Linux <linux@...linux.org.uk>
Cc:     Baruch Siach <baruch@...s.co.il>,
        Jan Kundrát <jan.kundrat@...net.cz>,
        Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
        Jason Cooper <jason@...edaemon.net>, linux-pci@...r.kernel.org,
        linux-kernel@...r.kernel.org, Bjorn Helgaas <bhelgaas@...gle.com>,
        linux-arm-kernel@...ts.infradead.org
Subject: Re: [BISECTED] Regression: Solidrun Clearfog Base won't boot since
 "PCI: mvebu: Only remap I/O space if configured"

Russell, Baruch, Jan,

On Thu, 13 Sep 2018 00:10:51 +0100, Russell King - ARM Linux wrote:

> It's probably the latter - the region is probably already mapped, that
> being the PCI IO region.
> 
> The original driver was setup to call pci_ioremap_io() as the very
> last thing - and as the driver is non-removable, we were guaranteed
> to never tear down this mapping (which is sensible, it's published
> to userspace.)
> 
> However, the current code calls pci_ioremap_io() much earlier, in
> a path where probe failures can occur.  This breaks pci_ioremap_io()'s
> requirements - it must not be called more than once.  So:
> 
> ee1604381a37 ("PCI: mvebu: Only remap I/O space if configured")
> 
> is basically incorrect - pci_ioremap_io() needs to move back to a
> place where it is only called in a path which will never fail.
> However, looking at the generic host bits, I'm not sure such a place
> exists in the new effort to make stuff more generic.

What about something like the below. I tested it, including the error
case by forcing an -EPROBE_DEFER. The new pci_unmap_io() is modeled
after pci_unmap_iospace(). Actually, I would prefer to use
pci_remap_iospace() and pci_unmap_iospace() but for now this API
doesn't allow overloading the memory type used for the mapping.

diff --git a/arch/arm/include/asm/io.h b/arch/arm/include/asm/io.h
index 2cfbc531f63b..cee0e2be5ac7 100644
--- a/arch/arm/include/asm/io.h
+++ b/arch/arm/include/asm/io.h
@@ -185,6 +185,7 @@ static inline void pci_ioremap_set_mem_type(int mem_type) {}
 #endif
 
 extern int pci_ioremap_io(unsigned int offset, phys_addr_t phys_addr);
+extern void pci_unmap_io(unsigned int offset);
 
 /*
  * PCI configuration space mapping function.
diff --git a/arch/arm/mm/ioremap.c b/arch/arm/mm/ioremap.c
index fc91205ff46c..f3a4df44bb27 100644
--- a/arch/arm/mm/ioremap.c
+++ b/arch/arm/mm/ioremap.c
@@ -482,6 +482,14 @@ int pci_ioremap_io(unsigned int offset, phys_addr_t phys_addr)
 }
 EXPORT_SYMBOL_GPL(pci_ioremap_io);
 
+void pci_unmap_io(unsigned int offset)
+{
+	BUG_ON(offset + SZ_64K > IO_SPACE_LIMIT);
+
+	unmap_kernel_range(PCI_IO_VIRT_BASE + offset, SZ_64K);
+}
+EXPORT_SYMBOL_GPL(pci_unmap_io);
+
 void __iomem *pci_remap_cfgspace(resource_size_t res_cookie, size_t size)
 {
 	return arch_ioremap_caller(res_cookie, size, MT_UNCACHED,
diff --git a/drivers/pci/controller/pci-mvebu.c b/drivers/pci/controller/pci-mvebu.c
index 50eb0729385b..772cff19c2ce 100644
--- a/drivers/pci/controller/pci-mvebu.c
+++ b/drivers/pci/controller/pci-mvebu.c
@@ -1145,7 +1145,6 @@ static int mvebu_pcie_parse_request_resources(struct mvebu_pcie *pcie)
 {
 	struct device *dev = &pcie->pdev->dev;
 	struct device_node *np = dev->of_node;
-	unsigned int i;
 	int ret;
 
 	INIT_LIST_HEAD(&pcie->resources);
@@ -1179,15 +1178,34 @@ static int mvebu_pcie_parse_request_resources(struct mvebu_pcie *pcie)
 					 resource_size(&pcie->io) - 1);
 		pcie->realio.name = "PCI I/O";
 
-		for (i = 0; i < resource_size(&pcie->realio); i += SZ_64K)
-			pci_ioremap_io(i, pcie->io.start + i);
-
 		pci_add_resource(&pcie->resources, &pcie->realio);
 	}
 
 	return devm_request_pci_bus_resources(dev, &pcie->resources);
 }
 
+static void mvebu_pcie_map_io(struct mvebu_pcie *pcie)
+{
+	int i;
+
+	if (resource_size(&pcie->io) == 0)
+		return;
+
+	for (i = 0; i < resource_size(&pcie->realio); i += SZ_64K)
+		pci_ioremap_io(i, pcie->io.start + i);
+}
+
+static void mvebu_pcie_unmap_io(struct mvebu_pcie *pcie)
+{
+	int i;
+
+	if (resource_size(&pcie->io) == 0)
+		return;
+
+	for (i = 0; i < resource_size(&pcie->realio); i += SZ_64K)
+		pci_unmap_io(i);
+}
+
 static int mvebu_pcie_probe(struct platform_device *pdev)
 {
 	struct device *dev = &pdev->dev;
@@ -1258,6 +1276,8 @@ static int mvebu_pcie_probe(struct platform_device *pdev)
 
 	pcie->nports = i;
 
+	mvebu_pcie_map_io(pcie);
+
 	list_splice_init(&pcie->resources, &bridge->windows);
 	bridge->dev.parent = dev;
 	bridge->sysdata = pcie;
@@ -1268,7 +1288,13 @@ static int mvebu_pcie_probe(struct platform_device *pdev)
 	bridge->align_resource = mvebu_pcie_align_resource;
 	bridge->msi = pcie->msi;
 
-	return pci_host_probe(bridge);
+	ret = pci_host_probe(bridge);
+	if (ret) {
+		mvebu_pcie_unmap_io(pcie);
+		return ret;
+	}
+
+	return 0;
 }
 
 static const struct of_device_id mvebu_pcie_of_match_table[] = {


-- 
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com