[PATCH v5 0/2] x86: Add IMR support to Quark/Galileo

[PATCH v5 0/2] x86: Add IMR support to Quark/Galileo

[Bryan O'Donoghue]

This patchset adds support for Isolated Memory Regions to the kernel.

Quark SoC X1000 contains a set of registers called Isolated Memory Regions.
IMRs provide fine grained memory access control to various system agents
within the SoC such as CPU SMM/non-SMM mode, PCIe virtual channels, CPU
snoop cycles, eSRAM flush cycles and the RMU. In simple terms, IMRs provide
a mechanism to protect memory regions from unwarranted access by system
agents that should not have access to that memory.

IMRs support a lock bit. Once a lock bit is set for an individual IMR it is
not possible to tear down that IMR without performing a cold boot of the
system. IMRs support reporting of violations. The SoC system can be
configured to reboot immediately when an IMR violation has taken place.
Immediate reboot of the system on IMR violation is recommended and is
currently how Quark BIOS configures the system.

An example of IMRs in use is given with Arduino compatiable Galileo boards
which ship with an IMR around the ACPI runtime services memory. If a DMA
read/write cycle were to occur to this region of memory this would trigger
the IMR violation mechansim.

As part of the IMR init code all unlocked IMRs are removed to ensure the
EFI memory map and IMR memory map are consistent. This is necessary since at
various stages during the boot of Quark systems firmware and second stage
bootloader will place unlocked IMRs around various assets in memory, with
the expectation that subsequent phases of boot will tear-down unlocked/stale
IMRs before proceeding. The kernel needs to tear-down unlocked IMRs placed
around the boot params structure and compressed kernel in memory. Without
doing so DMA addresses given out by the kernel to DMA capable hardware runs
the risk of triggering an IMR fault when DMA happens to those addresses.
As a result any unlocked IMR must be torn down by the kernel early in the
boot process to sanitize the memory map. 

As an additional protection to the run-time kernel from unwarranted memory
transactions an IMR is placed around the kernel's .text and .rodata
sections. 

Changes since v4:
 - Quark added as a platform in a separate patch to IMRs
   Andy Shevchenko
 - imr_read
   Add extra reg++ to end of read for consistency
   Andy Shevchenko
 - ret
   Declarations of return variable moved to end of each declaration block
   Andy Shevchenko
 - Parameters passed to debugfs hooks
   Andy Shevchenko
 - struct imr_dev *idev = &imr_dev;
   Used globally for entry point functions
   Andy Shevchenko/Bryan O'Donoghue
 - struct imr_dev * passed to static functions
   Andy Shevchenko/Bryan O'Donoghue
 - Comment in imr_fixup_memmap
   Paried down and made more readable
   Andy Shevchenko
 - imr_fixup_memmap/imr_selftest
   phys_addr_t used for base
   Andy Shevchenko
 - phys_addr_t/size_t
   Supplemented globally in place of unsigned long where appropriate
   Bryan O'Donoghue
 - print routines to take %pa globally
   In place of pr_info("0x%08lx\n", (unsigned long)some_phs_addr_t_variable);
   Andy Shevchenko
 - debugfs hook moved to after init = true
   Andy Shevchenko

Changes since v3:
 - Remove reference to imr.o in arch/x86/kernel/Makefile
   Bryan O'Donoghue

Changes since v2:
 - Move IMR code to arch/x86/platform/intel-quark/imr.c
   Thomas Gleixner/Darren Hart
 - #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
    Andy Shevchenko
 - ret = iosf_mbi_read()
    Style made consistent in imr_write
    Andy Shevchenko
 - reg++/IMR_NUM_REGS
    Offset used for lock bit in imr_write
    Andy Shevchenko
 - debugfs s->private pointer used
    Andy Shevchenko
 - debugfs
    Conditional compilation defines removed
    Andy Shevchenko
 - debugfs
    Failure to hook debugfs treated as non-fatal
    Andy Shevchenko
 - phys_addr_t
    Updated API to use phys_addr_t in place of unsigned long for base param
    Andy Shevchenko
 - printk
    "KiB" instead of "k"
    Andy Shevchenko
 - imr_enabled
    -> static inline imr_is_enabled
    Ong, Boon Leong/Andy Shevchenko
 - imr_write
    trap final ret from iosf_mbi_write for lock bit in imr_write - bugfix
    Ong, Boon Leong
 - imr_fixup_size
    -> static inline imr_fixup_size
    Ong, Boon Leong
 - imr_address_overlap
    -> imr_address_overlap
    Ong, Boon Leong
 - imr_add_range
    End address in imr_add_range calculated after imr_fixup_size()
    Ong, Boon Leong
 - imr_del_range
    Pass i in place of reg in imr_del_range() - bugfix
    Ong, Boon Leong
 - Add test case
    imr_del_range(-1, addr, size)
    Ong, Boon Leong/Andy Shevchenko
 - Added text "aligned to 1 KiB" removed reference to "4 k"
    Ong, Boon Leong
 - imr_is_enabled
    Definition of enabled updated to be negation of disabled
    Bryan O'Donoghue
 - imr_add_range
    Add check for adding an IMR in the disabled state
    Bryan O'Donoghue
 - Add test case IMR @ invalid address, @0 with rmask/wmask=CPU, @ 0 size 0x800
    Bryan O'Donoghue
 - Add WARN() to failed IMR test in print routine
    Bryan O'Donoghue
 - Update license to Dual BSD/GPL
    Reflect licensing in Intel reference code
    Bryan O'Donoghue

Changes since v1:
 - Galileo platform code
    Removed completely. Policy to tear-down unlocked IMRs and setup IMR
    around kernel .text and .rodata as part of IMR init code.
    Darren Hart/Ong, Boon Leong
 - imr_add/imr_del
    Renamed to imr_add_range and imr_del_range respectively.
    Andy Shevchenko
 - x86_match_cpu
    Used in place of DMI strings specific to Galileo.
    Andy Shevchenko/Ong, Boon Leong
 - Expanded git log definitions of IMRs
    Addition of more descriptive text to deliniate between different IMR
    types.
    Ong, Boon Leong
 - struct imr
    Renamed to struct imr_regs
    Andy Shevchenko/Darren Hart
 - imr_read/imr_write
    Flow reworked flow of register indexing
    Andy Shevchenko
 - debugfs hooks changed
    Andy Shevchenko
 - imr_enabled
    Definition of an enabled IMR updated to include read/write mask values
    present in IMR. Address @ zero and read/write mask in conjunction will
    be the definition of a disabled IMR on X1000 to be consistent with
    firmware both old and current which also defines a disabled IMR this
    way.
    Darren Hart/Ong, Boon Leong
 - Overlapping
    Comment added to code to explain the design decision not to allow IMR
    overlaps.
    Darren Hart/Ong, Boon Leong
 - CONFIG_DEBUG_IMR_SELFTEST
    Automated IMR self test moved from removed Galileo platform code and
    added to IMR init code. Option exists in the kernel hacking section.
    Darren Hart
 - IMR self test
    Expanded to over more scenarios
    Bryan O'Donoghue
 - Remove reference to IMR_ENABLE bit
    Undocumented bit with respect to Quark X1000
    Ong, Boon Leong
 - Expanded kernel IMR to encompass .text and .rodata
    IMR protecting both .text and .rodata as in the same way as .text and
    .rodata are marked read-only in the relevant page-table entries.
    Bryan O'Donoghue
 - Overlap bounds checking
    Moved range checking of overlap into a function
    Andy Shevchenko
 
Bryan O'Donoghue (2):
  x86: Add Isolated Memory Regions for Quark X1000
  x86, quark: Add Intel Quark platform support

 arch/x86/Kconfig                       |  16 +
 arch/x86/Kconfig.debug                 |  12 +
 arch/x86/include/asm/imr.h             |  60 +++
 arch/x86/platform/Makefile             |   1 +
 arch/x86/platform/intel-quark/Makefile |   1 +
 arch/x86/platform/intel-quark/imr.c    | 716 +++++++++++++++++++++++++++++++++
 drivers/platform/x86/Kconfig           |  25 ++
 7 files changed, 831 insertions(+)
 create mode 100644 arch/x86/include/asm/imr.h
 create mode 100644 arch/x86/platform/intel-quark/Makefile
 create mode 100644 arch/x86/platform/intel-quark/imr.c

-- 
1.9.1

[PATCH v5 1/2] x86: Add Isolated Memory Regions for Quark X1000

[Bryan O'Donoghue]

Intel's Quark X1000 SoC contains a set of registers called Isolated Memory
Regions. IMRs are accessed over the IOSF mailbox interface. IMRs are areas
carved out of memory that define read/write access rights to the various
system agents within the Quark system. For a given agent in the system it is
possible to specify if that agent may read or write an area of memory
defined by an IMR with a granularity of 1 KiB.

Quark_SecureBootPRM_330234_001.pdf section 4.5 details the concept of IMRs
quark-x1000-datasheet.pdf section 12.7.4 details the implementation of IMRs
in silicon.

eSRAM flush, CPU Snoop write-only, CPU SMM Mode, CPU non-SMM mode, RMU and
PCIe Virtual Channels (VC0 and VC1) can have individual read/write access
masks applied to them for a given memory region in Quark X1000. This
enables IMRs to treat each memory transaction type listed above on an
individual basis and to filter appropriately based on the IMR access mask
for the memory region. Quark supports eight IMRs.

Since all of the DMA capable SoC components in the X1000 are mapped to VC0
it is possible to define sections of memory as invalid for DMA write
operations originating from Ethernet, USB, SD and any other DMA capable
south-cluster component on VC0. Similarly it is possible to mark kernel
memory as non-SMM mode read/write only or to mark BIOS runtime memory as SMM
mode accessible only depending on the particular memory footprint on a given
system.

On an IMR violation Quark SoC X1000 systems are configured to reset the
system, so ensuring that the IMR memory map is consistent with the EFI
provided memory map is critical to ensure no IMR violations reset the
system.

The API for accessing IMRs is based on MTRR code but doesn't provide a /proc
or /sys interface to manipulate IMRs. Defining the size and extent of IMRs
is exclusively the domain of in-kernel code.

Quark firmware sets up a series of locked IMRs around pieces of memory that
firmware owns such as ACPI runtime data. During boot a series of unlocked
IMRs are placed around items in memory to guarantee no DMA modification of
those items can take place. Grub also places an unlocked IMR around the
kernel boot params data structure and compressed kernel image. It is
necessary for the kernel to tear down all unlocked IMRs in order to ensure
that the kernel's view of memory passed via the EFI memory map is consistent
with the IMR memory map. Without tearing down all unlocked IMRs on boot
transitory IMRs such as those used to protect the compressed kernel image
will cause IMR violations and system reboots.

The IMR init code tears down all unlocked IMRs and sets a protective IMR
around the kernel .text and .rodata as one contiguous block. This sanitizes
the IMR memory map with respect to the EFI memory map and protects the
read-only portions of the kernel from unwarranted DMA access.

Signed-off-by: Bryan O'Donoghue <pure.logic@nexus-software.ie>
---
 arch/x86/Kconfig.debug                 |  12 +
 arch/x86/include/asm/imr.h             |  60 +++
 arch/x86/platform/intel-quark/Makefile |   1 +
 arch/x86/platform/intel-quark/imr.c    | 716 +++++++++++++++++++++++++++++++++
 drivers/platform/x86/Kconfig           |  25 ++
 5 files changed, 814 insertions(+)
 create mode 100644 arch/x86/include/asm/imr.h
 create mode 100644 arch/x86/platform/intel-quark/Makefile
 create mode 100644 arch/x86/platform/intel-quark/imr.c

diff --git a/arch/x86/Kconfig.debug b/arch/x86/Kconfig.debug
index 61bd2ad..fcf5701 100644
--- a/arch/x86/Kconfig.debug
+++ b/arch/x86/Kconfig.debug
@@ -313,6 +313,18 @@ config DEBUG_NMI_SELFTEST
 
 	  If unsure, say N.
 
+config DEBUG_IMR_SELFTEST
+	bool "Isolated Memory Region self test"
+	default n
+	depends on INTEL_IMR
+	---help---
+	  This option enables automated sanity testing of the IMR code.
+	  Some simple tests are run to verify IMR bounds checking, alignment
+	  and overlapping. This option is really only useful if you are
+	  debugging an IMR memory map or are modifying the IMR code and want to
+	  test your changes.
+
+	  If unsure say N here.
 config X86_DEBUG_STATIC_CPU_HAS
 	bool "Debug alternatives"
 	depends on DEBUG_KERNEL
diff --git a/arch/x86/include/asm/imr.h b/arch/x86/include/asm/imr.h
new file mode 100644
index 0000000..0542dc2
--- /dev/null
+++ b/arch/x86/include/asm/imr.h
@@ -0,0 +1,60 @@
+/*
+ * imr.h: Isolated Memory Region API
+ *
+ * Copyright(c) 2013 Intel Corporation.
+ * Copyright(c) 2015 Bryan O'Donoghue <pure.logic@nexus-software.ie>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; version 2
+ * of the License.
+ */
+#ifndef _IMR_H
+#define _IMR_H
+
+#include <linux/types.h>
+
+/*
+ * IMR agent access mask bits
+ * See section 12.7.4.7 from quark-x1000-datasheet.pdf for register
+ * definitions.
+ */
+#define IMR_ESRAM_FLUSH		BIT(31)
+#define IMR_CPU_SNOOP		BIT(30)		/* Applicable only to write */
+#define IMR_RMU			BIT(29)
+#define IMR_VC1_SAI_ID3		BIT(15)
+#define IMR_VC1_SAI_ID2		BIT(14)
+#define IMR_VC1_SAI_ID1		BIT(13)
+#define IMR_VC1_SAI_ID0		BIT(12)
+#define IMR_VC0_SAI_ID3		BIT(11)
+#define IMR_VC0_SAI_ID2		BIT(10)
+#define IMR_VC0_SAI_ID1		BIT(9)
+#define IMR_VC0_SAI_ID0		BIT(8)
+#define IMR_CPU_0		BIT(1)		/* SMM mode */
+#define IMR_CPU			BIT(0)		/* Non SMM mode */
+#define IMR_ACCESS_NONE		0
+
+/*
+ * Read/Write access-all bits here include some reserved bits
+ * These are the values firmware uses and are accepted by hardware.
+ * The kernel defines read/write access-all in the same way as firmware
+ * in order to have a consistent and crisp definition across firmware,
+ * bootloader and kernel.
+ */
+#define IMR_READ_ACCESS_ALL	0xBFFFFFFF
+#define IMR_WRITE_ACCESS_ALL	0xFFFFFFFF
+
+/* Number of IMRs provided by Quark X1000 SoC */
+#define QUARK_X1000_IMR_MAX	0x08
+#define QUARK_X1000_IMR_REGBASE 0x40
+
+/* IMR alignment bits - only bits 31:10 are checked for IMR validity */
+#define IMR_ALIGN		0x400
+#define IMR_MASK		(IMR_ALIGN - 1)
+
+int imr_add_range(phys_addr_t base, size_t size,
+		  unsigned int rmask, unsigned int wmask, bool lock);
+
+int imr_remove_range(int reg, phys_addr_t base, size_t size);
+
+#endif /* _IMR_H */
diff --git a/arch/x86/platform/intel-quark/Makefile b/arch/x86/platform/intel-quark/Makefile
new file mode 100644
index 0000000..3180df5
--- /dev/null
+++ b/arch/x86/platform/intel-quark/Makefile
@@ -0,0 +1 @@
+obj-$(CONFIG_INTEL_IMR) += imr.o
diff --git a/arch/x86/platform/intel-quark/imr.c b/arch/x86/platform/intel-quark/imr.c
new file mode 100644
index 0000000..4e5be5c
--- /dev/null
+++ b/arch/x86/platform/intel-quark/imr.c
@@ -0,0 +1,716 @@
+/**
+ * imr.c
+ *
+ * Copyright(c) 2013 Intel Corporation.
+ * Copyright(c) 2015 Bryan O'Donoghue <pure.logic@nexus-software.ie>
+ *
+ * IMR registers define an isolated region of memory that can
+ * be masked to prohibit certain system agents from accessing memory.
+ * When a device behind a masked port performs an access - snooped or
+ * not, an IMR may optionally prevent that transaction from changing
+ * the state of memory or from getting correct data in response to the
+ * operation.
+ *
+ * Write data will be dropped and reads will return 0xFFFFFFFF, the
+ * system will reset and system BIOS will print out an error message to
+ * inform the user that an IMR has been violated.
+ *
+ * This code is based on the Linux MTRR code and reference code from
+ * Intel's Quark BSP EFI, Linux and grub code.
+ *
+ * See quark-x1000-datasheet.pdf for register definitions.
+ * http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/quark-x1000-datasheet.pdf
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <asm-generic/sections.h>
+#include <asm/cpu_device_id.h>
+#include <asm/imr.h>
+#include <asm/iosf_mbi.h>
+#include <linux/debugfs.h>
+#include <linux/init.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/types.h>
+
+struct imr_device {
+	struct dentry	*file;
+	bool		init;
+	struct mutex	lock;
+	int		max_imr;
+	int		reg_base;
+};
+
+static struct imr_device imr_dev;
+
+/*
+ * IMR read/write mask control registers.
+ * See quark-x1000-datasheet.pdf sections 12.7.4.5 and 12.7.4.6 for
+ * bit definitions.
+ *
+ * addr_hi
+ * 31		Lock bit
+ * 30:24	Reserved
+ * 23:2		1 KiB aligned lo address
+ * 1:0		Reserved
+ *
+ * addr_hi
+ * 31:24	Reserved
+ * 23:2		1 KiB aligned hi address
+ * 1:0		Reserved
+ */
+#define IMR_LOCK	BIT(31)
+
+struct imr_regs {
+	u32 addr_lo;
+	u32 addr_hi;
+	u32 rmask;
+	u32 wmask;
+};
+
+#define IMR_NUM_REGS	(sizeof(struct imr_regs)/sizeof(u32))
+#define IMR_SHIFT	8
+#define imr_to_phys(x)	((x) << IMR_SHIFT)
+#define phys_to_imr(x)	((x) >> IMR_SHIFT)
+
+/**
+ * imr_is_enabled - true if an IMR is enabled false otherwise
+ *
+ * Determines if an IMR is enabled based on address range and read/write
+ * mask. An IMR set with an address range set to zero and a read/write
+ * access mask set to all is considered to be disabled. An IMR in any
+ * other state - for example set to zero but without read/write access
+ * all is considered to be enabled. This definition of disabled is how
+ * firmware switches off an IMR and is maintained in kernel for
+ * consistency.
+ *
+ * @imr:	pointer to IMR descriptor
+ * @return:	true if IMR enabled false if disabled
+ */
+static inline int imr_is_enabled(struct imr_regs *imr)
+{
+	return !(imr->rmask == IMR_READ_ACCESS_ALL &&
+		 imr->wmask == IMR_WRITE_ACCESS_ALL &&
+		 imr_to_phys(imr->addr_lo) == 0 &&
+		 imr_to_phys(imr->addr_hi) == 0);
+}
+
+/**
+ * imr_read - read an IMR at a given index.
+ *
+ * Requires caller to hold imr mutex
+ *
+ * @idev:	pointer to imr_device structure
+ * @imr_id:	IMR entry to read
+ * @imr:	IMR structure representing address and access masks
+ * @return:	0 on success or error code passed from mbi_iosf on failure
+ */
+static int imr_read(struct imr_device *idev, u32 imr_id, struct imr_regs *imr)
+{
+	u32 reg = imr_id * IMR_NUM_REGS + idev->reg_base;
+	int ret;
+
+	ret = iosf_mbi_read(QRK_MBI_UNIT_MM, QRK_MBI_MM_READ,
+				reg++, &imr->addr_lo);
+	if (ret)
+		return ret;
+
+	ret = iosf_mbi_read(QRK_MBI_UNIT_MM, QRK_MBI_MM_READ,
+				reg++, &imr->addr_hi);
+	if (ret)
+		return ret;
+
+	ret = iosf_mbi_read(QRK_MBI_UNIT_MM, QRK_MBI_MM_READ,
+				reg++, &imr->rmask);
+	if (ret)
+		return ret;
+
+	ret = iosf_mbi_read(QRK_MBI_UNIT_MM, QRK_MBI_MM_READ,
+				reg++, &imr->wmask);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+/**
+ * imr_write - write an IMR at a given index.
+ *
+ * Requires caller to hold imr mutex
+ * Note lock bits need to be written independently of address bits
+ *
+ * @idev:	pointer to imr_device structure
+ * @imr_id:	IMR entry to write
+ * @imr:	IMR structure representing address and access masks
+ * @lock:	indicates if the IMR lock bit should be applied
+ * @return:	0 on success or error code passed from mbi_iosf on failure
+ */
+static int imr_write(struct imr_device *idev, u32 imr_id,
+		     struct imr_regs *imr, bool lock)
+{
+	unsigned long flags;
+	u32 reg = imr_id * IMR_NUM_REGS + idev->reg_base;
+	int ret;
+
+	local_irq_save(flags);
+
+	ret = iosf_mbi_write(QRK_MBI_UNIT_MM, QRK_MBI_MM_WRITE, reg++,
+				imr->addr_lo);
+	if (ret)
+		goto done;
+
+	ret = iosf_mbi_write(QRK_MBI_UNIT_MM, QRK_MBI_MM_WRITE,
+				reg++, imr->addr_hi);
+	if (ret)
+		goto done;
+
+	ret = iosf_mbi_write(QRK_MBI_UNIT_MM, QRK_MBI_MM_WRITE,
+				reg++, imr->rmask);
+	if (ret)
+		goto done;
+
+	ret = iosf_mbi_write(QRK_MBI_UNIT_MM, QRK_MBI_MM_WRITE,
+				reg++, imr->wmask);
+	if (ret)
+		goto done;
+
+	/* Lock bit must be set separately to addr_lo address bits */
+	if (lock) {
+		imr->addr_lo |= IMR_LOCK;
+		ret = iosf_mbi_write(QRK_MBI_UNIT_MM, QRK_MBI_MM_WRITE,
+					reg - IMR_NUM_REGS, imr->addr_lo);
+		if (ret)
+			goto done;
+	}
+
+	local_irq_restore(flags);
+	return 0;
+done:
+	/*
+	 * If writing to the IOSF failed then we're in an unknown state,
+	 * likely a very bad state. An IMR in an invalid state will almost
+	 * certainly lead to a memory access violation.
+	 */
+	local_irq_restore(flags);
+	WARN(ret, "IOSF-MBI write fail range 0x%08x-0x%08x unreliable\n",
+	     imr_to_phys(imr->addr_lo), imr_to_phys(imr->addr_hi) + IMR_MASK);
+
+	return ret;
+}
+
+/**
+ * imr_dbgfs_state_show
+ * Print state of IMR registers
+ *
+ * @s:		pointer to seq_file for output
+ * @unused:	unused parameter
+ * @return:	0 on success or error code passed from mbi_iosf on failure
+ */
+static int imr_dbgfs_state_show(struct seq_file *s, void *unused)
+{
+	phys_addr_t base;
+	phys_addr_t end;
+	int i;
+	struct imr_device *idev = s->private;
+	struct imr_regs imr;
+	size_t size;
+	int ret = -ENODEV;
+
+	mutex_lock(&idev->lock);
+
+	for (i = 0; i < idev->max_imr; i++) {
+
+		ret = imr_read(idev, i, &imr);
+		if (ret)
+			break;
+
+		/*
+		 * Remember to add IMR_ALIGN bytes to size to indicate the
+		 * inherent IMR_ALIGN size bytes contained in the masked away
+		 * lower ten bits
+		 */
+		if (imr_is_enabled(&imr)) {
+			base = imr_to_phys(imr.addr_lo);
+			end = imr_to_phys(imr.addr_hi) + IMR_MASK;
+		} else {
+			base = 0;
+			end = 0;
+		}
+		size = end - base;
+		seq_printf(s, "imr%02i: base=%pa, end=%pa, size=0x%08x "
+			   "rmask=0x%08x, wmask=0x%08x, %s, %s\n", i,
+			   &base, &end, size, imr.rmask, imr.wmask,
+			   imr_is_enabled(&imr) ? "enabled " : "disabled",
+			   imr.addr_lo & IMR_LOCK ? "locked" : "unlocked");
+	}
+
+	mutex_unlock(&idev->lock);
+
+	return ret;
+}
+
+/**
+ * imr_state_open - debugfs open callback
+ *
+ * @inode:	pointer to struct inode
+ * @file:	pointer to struct file
+ * @return:	result of single open
+ */
+static int imr_state_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, imr_dbgfs_state_show, inode->i_private);
+}
+
+static const struct file_operations imr_state_ops = {
+	.open		= imr_state_open,
+	.read		= seq_read,
+	.llseek		= seq_lseek,
+	.release	= single_release,
+};
+
+/**
+ * imr_debugfs_register - register debugfs hooks
+ *
+ * @idev:	pointer to imr_device structure
+ * @return:	0 on success - errno on failure
+ */
+static int imr_debugfs_register(struct imr_device *idev)
+{
+	idev->file = debugfs_create_file("imr_state", S_IFREG | S_IRUGO, NULL,
+					 idev, &imr_state_ops);
+	if (!idev->file)
+		return -ENOMEM;
+
+	return 0;
+}
+
+/**
+ * imr_debugfs_unregister - unregister debugfs hooks
+ *
+ * @idev:	pointer to imr_device structure
+ * @return:
+ */
+static void imr_debugfs_unregister(struct imr_device *idev)
+{
+	debugfs_remove(idev->file);
+}
+
+/**
+ * imr_check_range - checks passed address range for an IMR is aligned
+ *
+ * @base:	base address of intended IMR
+ * @size:	size of intended IMR
+ * @return:	zero on valid range -EINVAL on unaligned base/size
+ */
+static int imr_check_range(phys_addr_t base, size_t size)
+{
+	if ((base & IMR_MASK) || (size & IMR_MASK)) {
+		pr_warn("base %pa size 0x%08x must align to 1KiB\n",
+			&base, size);
+		return -EINVAL;
+	}
+	return 0;
+}
+
+/**
+ * imr_fixup_size - account for the IMR_ALIGN bytes that addr_hi appends
+ *
+ * IMR addr_hi has a built in offset of plus IMR_ALIGN (0x400) bytes from the
+ * value in the register. We need to subtract IMR_ALIGN bytes from input sizes
+ * as a result.
+ *
+ * @size:	input size bytes
+ * @return:	reduced size
+ */
+static inline size_t imr_fixup_size(size_t size)
+{
+	return size - IMR_ALIGN;
+}
+
+/**
+ * imr_address_overlap - detects an address overlap
+ *
+ * @addr:	address to check against an existing IMR
+ * @imr:	imr being checked
+ * @return:	true for overlap false for no overlap
+ */
+static inline int imr_address_overlap(phys_addr_t addr, struct imr_regs *imr)
+{
+	return addr >= imr_to_phys(imr->addr_lo) && addr <= imr_to_phys(imr->addr_hi);
+}
+
+/**
+ * imr_add_range - add an Isolated Memory Region
+ *
+ * @base:	physical base address of region aligned to 1KiB
+ * @size:	physical size of region in bytes must be aligned to 1KiB
+ * @read_mask:	read access mask
+ * @write_mask:	write access mask
+ * @lock:	indicates whether or not to permanently lock this region
+ * @return:	index of the IMR allocated or negative value indicating error
+ */
+int imr_add_range(phys_addr_t base, size_t size,
+		  unsigned int rmask, unsigned int wmask, bool lock)
+{
+	phys_addr_t end;
+	unsigned int i;
+	struct imr_device *idev = &imr_dev;
+	struct imr_regs imr;
+	int reg;
+	int ret;
+
+	if (idev->init == false)
+		return -ENODEV;
+
+	ret = imr_check_range(base, size);
+	if (ret)
+		return ret;
+
+	if (size == 0)
+		return -EINVAL;
+
+	/* Tweak the size value */
+	size = imr_fixup_size(size);
+	end = base + size;
+
+	/*
+	 * Check for reserved IMR value common to firmware, kernel and grub
+	 * indicating a disabled IMR
+	 */
+	imr.addr_lo = phys_to_imr(base);
+	imr.addr_hi = phys_to_imr(end);
+	imr.rmask = rmask;
+	imr.wmask = wmask;
+	if (!imr_is_enabled(&imr))
+		return -EINVAL;
+
+	mutex_lock(&idev->lock);
+
+	/*
+	 * Find a free IMR while checking for an existing overlapping range.
+	 * Note there's no restriction in silicon to prevent IMR overlaps.
+	 * For the sake of simplicity and ease in defining/debugging an IMR
+	 * memory map we exclude IMR overlaps.
+	 */
+	reg = -1;
+	for (i = 0; i < idev->max_imr; i++) {
+		ret = imr_read(idev, i, &imr);
+		if (ret)
+			goto done;
+
+		/* Find overlap @ base or end of requested range */
+		if (imr_is_enabled(&imr)) {
+			if (imr_address_overlap(base, &imr)) {
+				ret = -EINVAL;
+				goto done;
+			}
+			if (imr_address_overlap(end, &imr)) {
+				ret = -EINVAL;
+				goto done;
+			}
+		} else {
+			reg = i;
+		}
+	}
+
+	/* Error out if we have no free IMR entries */
+	if (reg == -1) {
+		ret = -ENODEV;
+		goto done;
+	}
+
+	pr_debug("IMR %d phys %pa-%pa rmask 0x%08x wmask 0x%08x\n",
+		reg, &base, &end, rmask, wmask);
+
+	/* Allocate IMR */
+	imr.addr_lo = phys_to_imr(base);
+	imr.addr_hi = phys_to_imr(end);
+	imr.rmask = rmask;
+	imr.wmask = wmask;
+
+	ret = imr_write(idev, reg, &imr, lock);
+
+done:
+	mutex_unlock(&idev->lock);
+	return ret == 0 ? reg : ret;
+}
+EXPORT_SYMBOL(imr_add_range);
+
+/**
+ * imr_remove_range - delete an Isolated Memory Region
+ *
+ * This function allows you to delete an IMR by its index specified by reg or
+ * by address range specified by base and size respectively. If you specify an
+ * index on its own the base and size parameters are ignored.
+ * imr_remove_range(0, size, base); delete IMR at index 0 base/size ignored
+ * imr_remove_range(-1, base, size); delete IMR from base to base+size
+ *
+ * @reg:	imr index to remove
+ * @base:	physical base address of region aligned to 1 KiB
+ * @size:	physical size of region in bytes aligned to 1 KiB
+ * @return:	-EINVAL on invalid range or out or range id
+ *		-ENODEV if reg is valid but no IMR exists or is locked
+ *		0 on success
+ */
+int imr_remove_range(int reg, phys_addr_t base, size_t size)
+{
+	phys_addr_t end;
+	bool found = false;
+	unsigned int i;
+	struct imr_device *idev = &imr_dev;
+	struct imr_regs imr;
+	int ret = 0;
+
+	if (idev->init == false)
+		return -ENODEV;
+
+	if (imr_check_range(base, size))
+		return -EINVAL;
+
+	if (reg >= idev->max_imr)
+		return -EINVAL;
+
+	/* Tweak the size value */
+	size = imr_fixup_size(size);
+	end = base + size;
+
+	mutex_lock(&idev->lock);
+
+	if (reg >= 0) {
+		/* If a specific IMR is given try to use it */
+		ret = imr_read(idev, reg, &imr);
+		if (ret)
+			goto done;
+
+		if (!imr_is_enabled(&imr) || imr.addr_lo & IMR_LOCK) {
+			ret = -ENODEV;
+			goto done;
+		}
+		found = true;
+	} else {
+		/* Search for match based on address range */
+		for (i = 0; i < idev->max_imr; i++) {
+			ret = imr_read(idev, i, &imr);
+			if (ret)
+				goto done;
+
+			if (!imr_is_enabled(&imr) || imr.addr_lo & IMR_LOCK)
+				continue;
+
+			if ((imr_to_phys(imr.addr_lo) == base) &&
+			    (imr_to_phys(imr.addr_hi) == end)) {
+				found = true;
+				reg = i;
+				break;
+			}
+		}
+	}
+
+	if (found == false) {
+		ret = -ENODEV;
+		goto done;
+	}
+
+	/* Tear down the IMR */
+	imr.addr_lo = 0;
+	imr.addr_hi = 0;
+	imr.rmask = IMR_READ_ACCESS_ALL;
+	imr.wmask = IMR_WRITE_ACCESS_ALL;
+
+	ret = imr_write(idev, reg, &imr, false);
+
+done:
+	mutex_unlock(&idev->lock);
+	return ret;
+}
+EXPORT_SYMBOL(imr_remove_range);
+
+#ifdef CONFIG_DEBUG_IMR_SELFTEST
+#define SELFTEST KBUILD_MODNAME ": self_test "
+
+/**
+ * imr_self_test_result - Print result string for self test
+ *
+ * @res:	result code - true if test passed false otherwise
+ * @fmt:	format string
+ * ...		variadic argument list
+ */
+static void __init imr_self_test_result(int res, const char *fmt, ...)
+{
+	va_list vlist;
+
+	va_start(vlist, fmt);
+	if (res)
+		printk(KERN_INFO SELFTEST "pass ");
+	else
+		printk(KERN_ERR SELFTEST "fail ");
+	vprintk(fmt, vlist);
+	va_end(vlist);
+	WARN(res == 0, "test failed");
+}
+#undef SELFTEST
+
+/**
+ * imr_self_test
+ *
+ * Verify IMR self_test with some simple tests to verify overlap,
+ * zero sized allocations and 1 KiB sized areas.
+ *
+ */
+static void __init imr_self_test(void)
+{
+	phys_addr_t base  = virt_to_phys(&_text);
+	size_t size = virt_to_phys(&__end_rodata) - base;
+	const char *fmt_over = "overlapped IMR @ (0x%08lx - 0x%08lx)\n";
+	int idx;
+
+	/* Test zero zero */
+	idx = imr_add_range(0, 0, 0, 0, false);
+	imr_self_test_result(idx < 0, "zero sized IMR\n");
+
+	/* Test exact overlap */
+	idx = imr_add_range(base, size, IMR_CPU, IMR_CPU, false);
+	imr_self_test_result(idx < 0, fmt_over, __va(base), __va(base + size));
+
+	/* Test overlap with base inside of existing */
+	base += size - IMR_ALIGN;
+	idx = imr_add_range(base, size, IMR_CPU, IMR_CPU, false);
+	imr_self_test_result(idx < 0, fmt_over, __va(base), __va(base + size));
+
+	/* Test overlap with end inside of existing */
+	base -= size + IMR_ALIGN * 2;
+	idx = imr_add_range(base, size, IMR_CPU, IMR_CPU, false);
+	imr_self_test_result(idx < 0, fmt_over, __va(base), __va(base + size));
+
+	/* Test that a 1 KiB IMR @ zero with read/write all will bomb out */
+	idx = imr_add_range(0, IMR_ALIGN, IMR_READ_ACCESS_ALL,
+			    IMR_WRITE_ACCESS_ALL, false);
+	imr_self_test_result(idx < 0, "1KiB IMR @ 0x00000000 - access-all\n");
+
+	/* Test that a 1 KiB IMR @ zero with CPU only will work */
+	idx = imr_add_range(0, IMR_ALIGN, IMR_CPU, IMR_CPU, false);
+	imr_self_test_result(idx >= 0, "1KiB IMR @ 0x00000000 - cpu access\n");
+	if (idx >= 0) {
+		/* For an index removal address doesn't matter */
+		idx = imr_remove_range(idx, 0, 0);
+		imr_self_test_result(idx == 0, "index-teardown - cpu access\n");
+	}
+
+	/* Test 2 KiB works - remove based on index */
+	size = IMR_ALIGN * 2;
+	idx = imr_add_range(0, size, IMR_READ_ACCESS_ALL,
+			    IMR_WRITE_ACCESS_ALL, false);
+	imr_self_test_result(idx >= 0, "2KiB IMR @ 0x00000000\n");
+	if (idx >= 0) {
+		/* For an index removal address doesn't matter */
+		idx = imr_remove_range(idx, 0, 0);
+		imr_self_test_result(idx == 0, "index-teardown 2KiB\n");
+	}
+
+	/* Test 1 KiB - remove based on address range */
+	idx = imr_add_range(0, size, IMR_READ_ACCESS_ALL,
+			    IMR_WRITE_ACCESS_ALL, false);
+	imr_self_test_result(idx >= 0, "2KiB IMR @ 0x00000000\n");
+	if (idx >= 0) {
+		idx = imr_remove_range(-1, 0, size);
+		imr_self_test_result(idx == 0, "addr-teardown 2KiB\n");
+	}
+}
+#endif /* CONFIG_DEBUG_IMR_SELFTEST */
+
+/**
+ * imr_fixup_memmap - Tear down IMRs used during bootup.
+ *
+ * BIOS and Grub both setup IMRs around compressed kernel, initrd memory
+ * that need to be removed before the kernel hands out one of the IMR
+ * encased addresses to a downstream DMA agent such as the SD or Ethernet.
+ * IMRs on Galileo are setup to immediately reset the system on violation.
+ * As a result if you're running a root filesystem from SD - you'll need
+ * the boot-time IMRs torn down or you'll find seemingly random resets when
+ * using your filesystem.
+ *
+ * @idev:	pointer to imr_device structure
+ * @return:
+ */
+static void __init imr_fixup_memmap(struct imr_device *idev)
+{
+	phys_addr_t base = virt_to_phys(&_text);
+	size_t size = virt_to_phys(&__end_rodata) - base;
+	int i;
+	int ret;
+
+	/* Tear down all existing unlocked IMRs */
+	for (i = 0; i < idev->max_imr; i++)
+		imr_remove_range(i, 0, 0);
+
+	/*
+	 * Setup a locked IMR around the physical extent of the kernel
+	 * from the beginning of the .text secton to the end of the
+	 * .rodata section as one physically contiguous block.
+	 */
+	ret = imr_add_range(base, size, IMR_CPU, IMR_CPU, true);
+	if (ret < 0)
+		pr_err("unable to setup IMR for kernel: (%p - %p)\n",
+			&_text, &__end_rodata);
+	else
+		pr_info("protecting kernel .text - .rodata: %d KiB (%p - %p)\n",
+			size / 1024, &_text, &__end_rodata);
+
+}
+
+static const struct x86_cpu_id imr_ids[] __initconst = {
+	{ X86_VENDOR_INTEL, 5, 9 },	/* Intel Quark SoC X1000 */
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, imr_ids);
+
+/**
+ * imr_probe - entry point for IMR driver
+ *
+ * return: -ENODEV for no IMR support 0 if good to go
+ */
+static int __init imr_init(void)
+{
+	struct imr_device *idev = &imr_dev;
+	int ret;
+
+	if (!x86_match_cpu(imr_ids) || !iosf_mbi_available())
+		return -ENODEV;
+
+	idev->max_imr = QUARK_X1000_IMR_MAX;
+	idev->reg_base = QUARK_X1000_IMR_REGBASE;
+	idev->init = true;
+
+	mutex_init(&idev->lock);
+	ret = imr_debugfs_register(idev);
+	if (ret != 0)
+		pr_warn("debugfs register failed!\n");
+	imr_fixup_memmap(idev);
+
+#ifdef CONFIG_DEBUG_IMR_SELFTEST
+	/* Run optional self test */
+	imr_self_test();
+#endif
+	return 0;
+}
+
+/**
+ * imr_exit - exit point for IMR code
+ *
+ * Deregisters debugfs, leave IMR state as-is.
+ *
+ * return:
+ */
+static void __exit imr_exit(void)
+{
+	imr_debugfs_unregister(&imr_dev);
+}
+
+module_init(imr_init);
+module_exit(imr_exit);
+
+MODULE_AUTHOR("Bryan O'Donoghue <pure.logic@nexus-software.ie>");
+MODULE_DESCRIPTION("Intel Isolated Memory Region driver");
+MODULE_LICENSE("Dual BSD/GPL");
diff --git a/drivers/platform/x86/Kconfig b/drivers/platform/x86/Kconfig
index 638e7970..9752761 100644
--- a/drivers/platform/x86/Kconfig
+++ b/drivers/platform/x86/Kconfig
@@ -735,6 +735,31 @@ config INTEL_IPS
 	  functionality.  If in doubt, say Y here; it will only load on
 	  supported platforms.
 
+config INTEL_IMR
+	bool "Intel Isolated Memory Region support"
+	default n
+	depends on X86_INTEL_QUARK && IOSF_MBI
+	---help---
+	  This option provides a means to manipulate Isolated Memory Regions.
+	  IMRs are a set of registers that define read and write access masks
+	  to prohibit certain system agents from accessing memory with 1 KiB
+	  granularity.
+
+	  IMRs make it possible to control read/write access to an address
+	  by hardware agents inside the SoC. Read and write masks can be
+	  defined for:
+		- eSRAM flush
+		- Dirty CPU snoop (write only)
+		- RMU access
+		- PCI Virtual Channel 0/Virtual Channel 1
+		- SMM mode
+		- Non SMM mode
+
+	  Quark contains a set of eight IMR registers and makes use of those
+	  registers during its bootup process.
+
+	  If you are running on a Galileo/Quark say Y here.
+
 config IBM_RTL
 	tristate "Device driver to enable PRTL support"
 	depends on X86 && PCI
-- 
1.9.1

[PATCH v5 2/2] x86, quark: Add Intel Quark platform support

[Bryan O'Donoghue]

Add Intel Quark platform support. Quark needs to pull down all unlocked IMRs
to ensure agreement with the EFI memory map post boot.

This patch adds an entry in Kconfig for Quark as a platform and makes IMR
support mandatory if selected.

Signed-off-by: Bryan O'Donoghue <pure.logic@nexus-software.ie>
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Suggested-by: Andy Shevchenko <andy.shevchenko@gmail.com>
---
 arch/x86/Kconfig           | 16 ++++++++++++++++
 arch/x86/platform/Makefile |  1 +
 2 files changed, 17 insertions(+)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 0dc9d01..36d50cd 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -485,6 +485,22 @@ config X86_INTEL_MID
 	  Intel MID platforms are based on an Intel processor and chipset which
 	  consume less power than most of the x86 derivatives.
 
+config X86_INTEL_QUARK
+	bool "Intel Quark platform support"
+	depends on X86_32
+	depends on X86_EXTENDED_PLATFORM
+	depends on X86_PLATFORM_DEVICES
+	depends on X86_TSC
+	depends on PCI
+	depends on PCI_GOANY
+	depends on X86_IO_APIC
+	select IOSF_MBI
+	select INTEL_IMR
+	---help---
+	  Select to include support for Quark X1000 SoC.
+	  Say Y here if you have a Quark based system such as the Arduino
+	  compatible Intel Galileo.
+
 config X86_INTEL_LPSS
 	bool "Intel Low Power Subsystem Support"
 	depends on ACPI
diff --git a/arch/x86/platform/Makefile b/arch/x86/platform/Makefile
index 85afde1..a62e0be 100644
--- a/arch/x86/platform/Makefile
+++ b/arch/x86/platform/Makefile
@@ -5,6 +5,7 @@ obj-y	+= geode/
 obj-y	+= goldfish/
 obj-y	+= iris/
 obj-y	+= intel-mid/
+obj-y	+= intel-quark/
 obj-y	+= olpc/
 obj-y	+= scx200/
 obj-y	+= sfi/
-- 
1.9.1

Re: [PATCH v5 2/2] x86, quark: Add Intel Quark platform support

[Andy Shevchenko]

On Wed, Jan 28, 2015 at 5:00 PM, Bryan O'Donoghue
<pure.logic@nexus-software.ie> wrote:
> Add Intel Quark platform support. Quark needs to pull down all unlocked IMRs
> to ensure agreement with the EFI memory map post boot.
>
> This patch adds an entry in Kconfig for Quark as a platform and makes IMR
> support mandatory if selected.

Looks good for me.

>
> Signed-off-by: Bryan O'Donoghue <pure.logic@nexus-software.ie>
> Suggested-by: Thomas Gleixner <tglx@linutronix.de>
> Suggested-by: Andy Shevchenko <andy.shevchenko@gmail.com>
> ---
>  arch/x86/Kconfig           | 16 ++++++++++++++++
>  arch/x86/platform/Makefile |  1 +
>  2 files changed, 17 insertions(+)
>
> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> index 0dc9d01..36d50cd 100644
> --- a/arch/x86/Kconfig
> +++ b/arch/x86/Kconfig
> @@ -485,6 +485,22 @@ config X86_INTEL_MID
>           Intel MID platforms are based on an Intel processor and chipset which
>           consume less power than most of the x86 derivatives.
>
> +config X86_INTEL_QUARK
> +       bool "Intel Quark platform support"
> +       depends on X86_32
> +       depends on X86_EXTENDED_PLATFORM
> +       depends on X86_PLATFORM_DEVICES
> +       depends on X86_TSC
> +       depends on PCI
> +       depends on PCI_GOANY
> +       depends on X86_IO_APIC
> +       select IOSF_MBI
> +       select INTEL_IMR
> +       ---help---
> +         Select to include support for Quark X1000 SoC.
> +         Say Y here if you have a Quark based system such as the Arduino
> +         compatible Intel Galileo.
> +
>  config X86_INTEL_LPSS
>         bool "Intel Low Power Subsystem Support"
>         depends on ACPI
> diff --git a/arch/x86/platform/Makefile b/arch/x86/platform/Makefile
> index 85afde1..a62e0be 100644
> --- a/arch/x86/platform/Makefile
> +++ b/arch/x86/platform/Makefile
> @@ -5,6 +5,7 @@ obj-y   += geode/
>  obj-y  += goldfish/
>  obj-y  += iris/
>  obj-y  += intel-mid/
> +obj-y  += intel-quark/
>  obj-y  += olpc/
>  obj-y  += scx200/
>  obj-y  += sfi/
> --
> 1.9.1
>



-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH v5 1/2] x86: Add Isolated Memory Regions for Quark X1000

[Andy Shevchenko]

On Wed, Jan 28, 2015 at 5:00 PM, Bryan O'Donoghue
<pure.logic@nexus-software.ie> wrote:
> Intel's Quark X1000 SoC contains a set of registers called Isolated Memory
> Regions. IMRs are accessed over the IOSF mailbox interface. IMRs are areas
> carved out of memory that define read/write access rights to the various
> system agents within the Quark system. For a given agent in the system it is
> possible to specify if that agent may read or write an area of memory
> defined by an IMR with a granularity of 1 KiB.
>
> Quark_SecureBootPRM_330234_001.pdf section 4.5 details the concept of IMRs
> quark-x1000-datasheet.pdf section 12.7.4 details the implementation of IMRs
> in silicon.
>
> eSRAM flush, CPU Snoop write-only, CPU SMM Mode, CPU non-SMM mode, RMU and
> PCIe Virtual Channels (VC0 and VC1) can have individual read/write access
> masks applied to them for a given memory region in Quark X1000. This
> enables IMRs to treat each memory transaction type listed above on an
> individual basis and to filter appropriately based on the IMR access mask
> for the memory region. Quark supports eight IMRs.
>
> Since all of the DMA capable SoC components in the X1000 are mapped to VC0
> it is possible to define sections of memory as invalid for DMA write
> operations originating from Ethernet, USB, SD and any other DMA capable
> south-cluster component on VC0. Similarly it is possible to mark kernel
> memory as non-SMM mode read/write only or to mark BIOS runtime memory as SMM
> mode accessible only depending on the particular memory footprint on a given
> system.
>
> On an IMR violation Quark SoC X1000 systems are configured to reset the
> system, so ensuring that the IMR memory map is consistent with the EFI
> provided memory map is critical to ensure no IMR violations reset the
> system.
>
> The API for accessing IMRs is based on MTRR code but doesn't provide a /proc
> or /sys interface to manipulate IMRs. Defining the size and extent of IMRs
> is exclusively the domain of in-kernel code.
>
> Quark firmware sets up a series of locked IMRs around pieces of memory that
> firmware owns such as ACPI runtime data. During boot a series of unlocked
> IMRs are placed around items in memory to guarantee no DMA modification of
> those items can take place. Grub also places an unlocked IMR around the
> kernel boot params data structure and compressed kernel image. It is
> necessary for the kernel to tear down all unlocked IMRs in order to ensure
> that the kernel's view of memory passed via the EFI memory map is consistent
> with the IMR memory map. Without tearing down all unlocked IMRs on boot
> transitory IMRs such as those used to protect the compressed kernel image
> will cause IMR violations and system reboots.
>
> The IMR init code tears down all unlocked IMRs and sets a protective IMR
> around the kernel .text and .rodata as one contiguous block. This sanitizes
> the IMR memory map with respect to the EFI memory map and protects the
> read-only portions of the kernel from unwarranted DMA access.
>
> Signed-off-by: Bryan O'Donoghue <pure.logic@nexus-software.ie>

Few generic comments:
 - missed dots at the end of several sentences
 - %zu is suitable for size_t in printk()

One optional (though I suppose not bad) idea is to split test to
separate module which might go as a separate patch in the series.

After addressing those comments:
Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>

> ---
>  arch/x86/Kconfig.debug                 |  12 +
>  arch/x86/include/asm/imr.h             |  60 +++
>  arch/x86/platform/intel-quark/Makefile |   1 +
>  arch/x86/platform/intel-quark/imr.c    | 716 +++++++++++++++++++++++++++++++++
>  drivers/platform/x86/Kconfig           |  25 ++
>  5 files changed, 814 insertions(+)
>  create mode 100644 arch/x86/include/asm/imr.h
>  create mode 100644 arch/x86/platform/intel-quark/Makefile
>  create mode 100644 arch/x86/platform/intel-quark/imr.c
>
> diff --git a/arch/x86/Kconfig.debug b/arch/x86/Kconfig.debug
> index 61bd2ad..fcf5701 100644
> --- a/arch/x86/Kconfig.debug
> +++ b/arch/x86/Kconfig.debug
> @@ -313,6 +313,18 @@ config DEBUG_NMI_SELFTEST
>
>           If unsure, say N.
>
> +config DEBUG_IMR_SELFTEST
> +       bool "Isolated Memory Region self test"
> +       default n
> +       depends on INTEL_IMR
> +       ---help---
> +         This option enables automated sanity testing of the IMR code.
> +         Some simple tests are run to verify IMR bounds checking, alignment
> +         and overlapping. This option is really only useful if you are
> +         debugging an IMR memory map or are modifying the IMR code and want to
> +         test your changes.
> +
> +         If unsure say N here.
>  config X86_DEBUG_STATIC_CPU_HAS
>         bool "Debug alternatives"
>         depends on DEBUG_KERNEL
> diff --git a/arch/x86/include/asm/imr.h b/arch/x86/include/asm/imr.h
> new file mode 100644
> index 0000000..0542dc2
> --- /dev/null
> +++ b/arch/x86/include/asm/imr.h
> @@ -0,0 +1,60 @@
> +/*
> + * imr.h: Isolated Memory Region API
> + *
> + * Copyright(c) 2013 Intel Corporation.
> + * Copyright(c) 2015 Bryan O'Donoghue <pure.logic@nexus-software.ie>
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; version 2
> + * of the License.
> + */
> +#ifndef _IMR_H
> +#define _IMR_H
> +
> +#include <linux/types.h>
> +
> +/*
> + * IMR agent access mask bits
> + * See section 12.7.4.7 from quark-x1000-datasheet.pdf for register
> + * definitions.
> + */
> +#define IMR_ESRAM_FLUSH                BIT(31)
> +#define IMR_CPU_SNOOP          BIT(30)         /* Applicable only to write */
> +#define IMR_RMU                        BIT(29)
> +#define IMR_VC1_SAI_ID3                BIT(15)
> +#define IMR_VC1_SAI_ID2                BIT(14)
> +#define IMR_VC1_SAI_ID1                BIT(13)
> +#define IMR_VC1_SAI_ID0                BIT(12)
> +#define IMR_VC0_SAI_ID3                BIT(11)
> +#define IMR_VC0_SAI_ID2                BIT(10)
> +#define IMR_VC0_SAI_ID1                BIT(9)
> +#define IMR_VC0_SAI_ID0                BIT(8)
> +#define IMR_CPU_0              BIT(1)          /* SMM mode */
> +#define IMR_CPU                        BIT(0)          /* Non SMM mode */
> +#define IMR_ACCESS_NONE                0
> +
> +/*
> + * Read/Write access-all bits here include some reserved bits
> + * These are the values firmware uses and are accepted by hardware.
> + * The kernel defines read/write access-all in the same way as firmware
> + * in order to have a consistent and crisp definition across firmware,
> + * bootloader and kernel.
> + */
> +#define IMR_READ_ACCESS_ALL    0xBFFFFFFF
> +#define IMR_WRITE_ACCESS_ALL   0xFFFFFFFF
> +
> +/* Number of IMRs provided by Quark X1000 SoC */
> +#define QUARK_X1000_IMR_MAX    0x08
> +#define QUARK_X1000_IMR_REGBASE 0x40
> +
> +/* IMR alignment bits - only bits 31:10 are checked for IMR validity */
> +#define IMR_ALIGN              0x400
> +#define IMR_MASK               (IMR_ALIGN - 1)
> +
> +int imr_add_range(phys_addr_t base, size_t size,
> +                 unsigned int rmask, unsigned int wmask, bool lock);
> +
> +int imr_remove_range(int reg, phys_addr_t base, size_t size);
> +
> +#endif /* _IMR_H */
> diff --git a/arch/x86/platform/intel-quark/Makefile b/arch/x86/platform/intel-quark/Makefile
> new file mode 100644
> index 0000000..3180df5
> --- /dev/null
> +++ b/arch/x86/platform/intel-quark/Makefile
> @@ -0,0 +1 @@
> +obj-$(CONFIG_INTEL_IMR) += imr.o
> diff --git a/arch/x86/platform/intel-quark/imr.c b/arch/x86/platform/intel-quark/imr.c
> new file mode 100644
> index 0000000..4e5be5c
> --- /dev/null
> +++ b/arch/x86/platform/intel-quark/imr.c
> @@ -0,0 +1,716 @@
> +/**
> + * imr.c
> + *
> + * Copyright(c) 2013 Intel Corporation.
> + * Copyright(c) 2015 Bryan O'Donoghue <pure.logic@nexus-software.ie>
> + *
> + * IMR registers define an isolated region of memory that can
> + * be masked to prohibit certain system agents from accessing memory.
> + * When a device behind a masked port performs an access - snooped or
> + * not, an IMR may optionally prevent that transaction from changing
> + * the state of memory or from getting correct data in response to the
> + * operation.
> + *
> + * Write data will be dropped and reads will return 0xFFFFFFFF, the
> + * system will reset and system BIOS will print out an error message to
> + * inform the user that an IMR has been violated.
> + *
> + * This code is based on the Linux MTRR code and reference code from
> + * Intel's Quark BSP EFI, Linux and grub code.
> + *
> + * See quark-x1000-datasheet.pdf for register definitions.
> + * http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/quark-x1000-datasheet.pdf
> + */
> +
> +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
> +
> +#include <asm-generic/sections.h>
> +#include <asm/cpu_device_id.h>
> +#include <asm/imr.h>
> +#include <asm/iosf_mbi.h>
> +#include <linux/debugfs.h>
> +#include <linux/init.h>
> +#include <linux/mm.h>
> +#include <linux/module.h>
> +#include <linux/platform_device.h>
> +#include <linux/types.h>
> +
> +struct imr_device {
> +       struct dentry   *file;
> +       bool            init;
> +       struct mutex    lock;
> +       int             max_imr;
> +       int             reg_base;
> +};
> +
> +static struct imr_device imr_dev;
> +
> +/*
> + * IMR read/write mask control registers.
> + * See quark-x1000-datasheet.pdf sections 12.7.4.5 and 12.7.4.6 for
> + * bit definitions.
> + *
> + * addr_hi
> + * 31          Lock bit
> + * 30:24       Reserved
> + * 23:2                1 KiB aligned lo address
> + * 1:0         Reserved
> + *
> + * addr_hi
> + * 31:24       Reserved
> + * 23:2                1 KiB aligned hi address
> + * 1:0         Reserved
> + */
> +#define IMR_LOCK       BIT(31)
> +
> +struct imr_regs {
> +       u32 addr_lo;
> +       u32 addr_hi;
> +       u32 rmask;
> +       u32 wmask;
> +};
> +
> +#define IMR_NUM_REGS   (sizeof(struct imr_regs)/sizeof(u32))
> +#define IMR_SHIFT      8
> +#define imr_to_phys(x) ((x) << IMR_SHIFT)
> +#define phys_to_imr(x) ((x) >> IMR_SHIFT)
> +
> +/**
> + * imr_is_enabled - true if an IMR is enabled false otherwise
> + *
> + * Determines if an IMR is enabled based on address range and read/write
> + * mask. An IMR set with an address range set to zero and a read/write
> + * access mask set to all is considered to be disabled. An IMR in any
> + * other state - for example set to zero but without read/write access
> + * all is considered to be enabled. This definition of disabled is how
> + * firmware switches off an IMR and is maintained in kernel for
> + * consistency.
> + *
> + * @imr:       pointer to IMR descriptor
> + * @return:    true if IMR enabled false if disabled
> + */
> +static inline int imr_is_enabled(struct imr_regs *imr)
> +{
> +       return !(imr->rmask == IMR_READ_ACCESS_ALL &&
> +                imr->wmask == IMR_WRITE_ACCESS_ALL &&
> +                imr_to_phys(imr->addr_lo) == 0 &&
> +                imr_to_phys(imr->addr_hi) == 0);
> +}
> +
> +/**
> + * imr_read - read an IMR at a given index.
> + *
> + * Requires caller to hold imr mutex
> + *
> + * @idev:      pointer to imr_device structure
> + * @imr_id:    IMR entry to read
> + * @imr:       IMR structure representing address and access masks
> + * @return:    0 on success or error code passed from mbi_iosf on failure
> + */
> +static int imr_read(struct imr_device *idev, u32 imr_id, struct imr_regs *imr)
> +{
> +       u32 reg = imr_id * IMR_NUM_REGS + idev->reg_base;
> +       int ret;
> +
> +       ret = iosf_mbi_read(QRK_MBI_UNIT_MM, QRK_MBI_MM_READ,
> +                               reg++, &imr->addr_lo);
> +       if (ret)
> +               return ret;
> +
> +       ret = iosf_mbi_read(QRK_MBI_UNIT_MM, QRK_MBI_MM_READ,
> +                               reg++, &imr->addr_hi);
> +       if (ret)
> +               return ret;
> +
> +       ret = iosf_mbi_read(QRK_MBI_UNIT_MM, QRK_MBI_MM_READ,
> +                               reg++, &imr->rmask);
> +       if (ret)
> +               return ret;
> +
> +       ret = iosf_mbi_read(QRK_MBI_UNIT_MM, QRK_MBI_MM_READ,
> +                               reg++, &imr->wmask);
> +       if (ret)
> +               return ret;
> +
> +       return 0;
> +}
> +
> +/**
> + * imr_write - write an IMR at a given index.
> + *
> + * Requires caller to hold imr mutex
> + * Note lock bits need to be written independently of address bits
> + *
> + * @idev:      pointer to imr_device structure
> + * @imr_id:    IMR entry to write
> + * @imr:       IMR structure representing address and access masks
> + * @lock:      indicates if the IMR lock bit should be applied
> + * @return:    0 on success or error code passed from mbi_iosf on failure
> + */
> +static int imr_write(struct imr_device *idev, u32 imr_id,
> +                    struct imr_regs *imr, bool lock)
> +{
> +       unsigned long flags;
> +       u32 reg = imr_id * IMR_NUM_REGS + idev->reg_base;
> +       int ret;
> +
> +       local_irq_save(flags);
> +
> +       ret = iosf_mbi_write(QRK_MBI_UNIT_MM, QRK_MBI_MM_WRITE, reg++,
> +                               imr->addr_lo);
> +       if (ret)
> +               goto done;
> +
> +       ret = iosf_mbi_write(QRK_MBI_UNIT_MM, QRK_MBI_MM_WRITE,
> +                               reg++, imr->addr_hi);
> +       if (ret)
> +               goto done;
> +
> +       ret = iosf_mbi_write(QRK_MBI_UNIT_MM, QRK_MBI_MM_WRITE,
> +                               reg++, imr->rmask);
> +       if (ret)
> +               goto done;
> +
> +       ret = iosf_mbi_write(QRK_MBI_UNIT_MM, QRK_MBI_MM_WRITE,
> +                               reg++, imr->wmask);
> +       if (ret)
> +               goto done;
> +
> +       /* Lock bit must be set separately to addr_lo address bits */
> +       if (lock) {
> +               imr->addr_lo |= IMR_LOCK;
> +               ret = iosf_mbi_write(QRK_MBI_UNIT_MM, QRK_MBI_MM_WRITE,
> +                                       reg - IMR_NUM_REGS, imr->addr_lo);
> +               if (ret)
> +                       goto done;
> +       }
> +
> +       local_irq_restore(flags);
> +       return 0;
> +done:
> +       /*
> +        * If writing to the IOSF failed then we're in an unknown state,
> +        * likely a very bad state. An IMR in an invalid state will almost
> +        * certainly lead to a memory access violation.
> +        */
> +       local_irq_restore(flags);
> +       WARN(ret, "IOSF-MBI write fail range 0x%08x-0x%08x unreliable\n",
> +            imr_to_phys(imr->addr_lo), imr_to_phys(imr->addr_hi) + IMR_MASK);
> +
> +       return ret;
> +}
> +
> +/**
> + * imr_dbgfs_state_show
> + * Print state of IMR registers
> + *
> + * @s:         pointer to seq_file for output
> + * @unused:    unused parameter
> + * @return:    0 on success or error code passed from mbi_iosf on failure
> + */
> +static int imr_dbgfs_state_show(struct seq_file *s, void *unused)
> +{
> +       phys_addr_t base;
> +       phys_addr_t end;
> +       int i;
> +       struct imr_device *idev = s->private;
> +       struct imr_regs imr;
> +       size_t size;
> +       int ret = -ENODEV;
> +
> +       mutex_lock(&idev->lock);
> +
> +       for (i = 0; i < idev->max_imr; i++) {
> +
> +               ret = imr_read(idev, i, &imr);
> +               if (ret)
> +                       break;
> +
> +               /*
> +                * Remember to add IMR_ALIGN bytes to size to indicate the
> +                * inherent IMR_ALIGN size bytes contained in the masked away
> +                * lower ten bits
> +                */
> +               if (imr_is_enabled(&imr)) {
> +                       base = imr_to_phys(imr.addr_lo);
> +                       end = imr_to_phys(imr.addr_hi) + IMR_MASK;
> +               } else {
> +                       base = 0;
> +                       end = 0;
> +               }
> +               size = end - base;
> +               seq_printf(s, "imr%02i: base=%pa, end=%pa, size=0x%08x "
> +                          "rmask=0x%08x, wmask=0x%08x, %s, %s\n", i,
> +                          &base, &end, size, imr.rmask, imr.wmask,
> +                          imr_is_enabled(&imr) ? "enabled " : "disabled",
> +                          imr.addr_lo & IMR_LOCK ? "locked" : "unlocked");
> +       }
> +
> +       mutex_unlock(&idev->lock);
> +
> +       return ret;
> +}
> +
> +/**
> + * imr_state_open - debugfs open callback
> + *
> + * @inode:     pointer to struct inode
> + * @file:      pointer to struct file
> + * @return:    result of single open
> + */
> +static int imr_state_open(struct inode *inode, struct file *file)
> +{
> +       return single_open(file, imr_dbgfs_state_show, inode->i_private);
> +}
> +
> +static const struct file_operations imr_state_ops = {
> +       .open           = imr_state_open,
> +       .read           = seq_read,
> +       .llseek         = seq_lseek,
> +       .release        = single_release,
> +};
> +
> +/**
> + * imr_debugfs_register - register debugfs hooks
> + *
> + * @idev:      pointer to imr_device structure
> + * @return:    0 on success - errno on failure
> + */
> +static int imr_debugfs_register(struct imr_device *idev)
> +{
> +       idev->file = debugfs_create_file("imr_state", S_IFREG | S_IRUGO, NULL,
> +                                        idev, &imr_state_ops);
> +       if (!idev->file)
> +               return -ENOMEM;
> +
> +       return 0;
> +}
> +
> +/**
> + * imr_debugfs_unregister - unregister debugfs hooks
> + *
> + * @idev:      pointer to imr_device structure
> + * @return:
> + */
> +static void imr_debugfs_unregister(struct imr_device *idev)
> +{
> +       debugfs_remove(idev->file);
> +}
> +
> +/**
> + * imr_check_range - checks passed address range for an IMR is aligned
> + *
> + * @base:      base address of intended IMR
> + * @size:      size of intended IMR
> + * @return:    zero on valid range -EINVAL on unaligned base/size
> + */
> +static int imr_check_range(phys_addr_t base, size_t size)
> +{
> +       if ((base & IMR_MASK) || (size & IMR_MASK)) {
> +               pr_warn("base %pa size 0x%08x must align to 1KiB\n",
> +                       &base, size);
> +               return -EINVAL;
> +       }
> +       return 0;
> +}
> +
> +/**
> + * imr_fixup_size - account for the IMR_ALIGN bytes that addr_hi appends
> + *
> + * IMR addr_hi has a built in offset of plus IMR_ALIGN (0x400) bytes from the
> + * value in the register. We need to subtract IMR_ALIGN bytes from input sizes
> + * as a result.
> + *
> + * @size:      input size bytes
> + * @return:    reduced size
> + */
> +static inline size_t imr_fixup_size(size_t size)
> +{
> +       return size - IMR_ALIGN;
> +}
> +
> +/**
> + * imr_address_overlap - detects an address overlap
> + *
> + * @addr:      address to check against an existing IMR
> + * @imr:       imr being checked
> + * @return:    true for overlap false for no overlap
> + */
> +static inline int imr_address_overlap(phys_addr_t addr, struct imr_regs *imr)
> +{
> +       return addr >= imr_to_phys(imr->addr_lo) && addr <= imr_to_phys(imr->addr_hi);
> +}
> +
> +/**
> + * imr_add_range - add an Isolated Memory Region
> + *
> + * @base:      physical base address of region aligned to 1KiB
> + * @size:      physical size of region in bytes must be aligned to 1KiB
> + * @read_mask: read access mask
> + * @write_mask:        write access mask
> + * @lock:      indicates whether or not to permanently lock this region
> + * @return:    index of the IMR allocated or negative value indicating error
> + */
> +int imr_add_range(phys_addr_t base, size_t size,
> +                 unsigned int rmask, unsigned int wmask, bool lock)
> +{
> +       phys_addr_t end;
> +       unsigned int i;
> +       struct imr_device *idev = &imr_dev;
> +       struct imr_regs imr;
> +       int reg;
> +       int ret;
> +
> +       if (idev->init == false)
> +               return -ENODEV;
> +
> +       ret = imr_check_range(base, size);
> +       if (ret)
> +               return ret;
> +
> +       if (size == 0)
> +               return -EINVAL;
> +
> +       /* Tweak the size value */
> +       size = imr_fixup_size(size);
> +       end = base + size;
> +
> +       /*
> +        * Check for reserved IMR value common to firmware, kernel and grub
> +        * indicating a disabled IMR
> +        */
> +       imr.addr_lo = phys_to_imr(base);
> +       imr.addr_hi = phys_to_imr(end);
> +       imr.rmask = rmask;
> +       imr.wmask = wmask;
> +       if (!imr_is_enabled(&imr))
> +               return -EINVAL;
> +
> +       mutex_lock(&idev->lock);
> +
> +       /*
> +        * Find a free IMR while checking for an existing overlapping range.
> +        * Note there's no restriction in silicon to prevent IMR overlaps.
> +        * For the sake of simplicity and ease in defining/debugging an IMR
> +        * memory map we exclude IMR overlaps.
> +        */
> +       reg = -1;
> +       for (i = 0; i < idev->max_imr; i++) {
> +               ret = imr_read(idev, i, &imr);
> +               if (ret)
> +                       goto done;
> +
> +               /* Find overlap @ base or end of requested range */
> +               if (imr_is_enabled(&imr)) {
> +                       if (imr_address_overlap(base, &imr)) {
> +                               ret = -EINVAL;
> +                               goto done;
> +                       }
> +                       if (imr_address_overlap(end, &imr)) {
> +                               ret = -EINVAL;
> +                               goto done;
> +                       }
> +               } else {
> +                       reg = i;
> +               }
> +       }
> +
> +       /* Error out if we have no free IMR entries */
> +       if (reg == -1) {
> +               ret = -ENODEV;
> +               goto done;
> +       }
> +
> +       pr_debug("IMR %d phys %pa-%pa rmask 0x%08x wmask 0x%08x\n",
> +               reg, &base, &end, rmask, wmask);
> +
> +       /* Allocate IMR */
> +       imr.addr_lo = phys_to_imr(base);
> +       imr.addr_hi = phys_to_imr(end);
> +       imr.rmask = rmask;
> +       imr.wmask = wmask;
> +
> +       ret = imr_write(idev, reg, &imr, lock);
> +
> +done:
> +       mutex_unlock(&idev->lock);
> +       return ret == 0 ? reg : ret;
> +}
> +EXPORT_SYMBOL(imr_add_range);
> +
> +/**
> + * imr_remove_range - delete an Isolated Memory Region
> + *
> + * This function allows you to delete an IMR by its index specified by reg or
> + * by address range specified by base and size respectively. If you specify an
> + * index on its own the base and size parameters are ignored.
> + * imr_remove_range(0, size, base); delete IMR at index 0 base/size ignored
> + * imr_remove_range(-1, base, size); delete IMR from base to base+size
> + *
> + * @reg:       imr index to remove
> + * @base:      physical base address of region aligned to 1 KiB
> + * @size:      physical size of region in bytes aligned to 1 KiB
> + * @return:    -EINVAL on invalid range or out or range id
> + *             -ENODEV if reg is valid but no IMR exists or is locked
> + *             0 on success
> + */
> +int imr_remove_range(int reg, phys_addr_t base, size_t size)
> +{
> +       phys_addr_t end;
> +       bool found = false;
> +       unsigned int i;
> +       struct imr_device *idev = &imr_dev;
> +       struct imr_regs imr;
> +       int ret = 0;
> +
> +       if (idev->init == false)
> +               return -ENODEV;
> +
> +       if (imr_check_range(base, size))
> +               return -EINVAL;
> +
> +       if (reg >= idev->max_imr)
> +               return -EINVAL;
> +
> +       /* Tweak the size value */
> +       size = imr_fixup_size(size);
> +       end = base + size;
> +
> +       mutex_lock(&idev->lock);
> +
> +       if (reg >= 0) {
> +               /* If a specific IMR is given try to use it */
> +               ret = imr_read(idev, reg, &imr);
> +               if (ret)
> +                       goto done;
> +
> +               if (!imr_is_enabled(&imr) || imr.addr_lo & IMR_LOCK) {
> +                       ret = -ENODEV;
> +                       goto done;
> +               }
> +               found = true;
> +       } else {
> +               /* Search for match based on address range */
> +               for (i = 0; i < idev->max_imr; i++) {
> +                       ret = imr_read(idev, i, &imr);
> +                       if (ret)
> +                               goto done;
> +
> +                       if (!imr_is_enabled(&imr) || imr.addr_lo & IMR_LOCK)
> +                               continue;
> +
> +                       if ((imr_to_phys(imr.addr_lo) == base) &&
> +                           (imr_to_phys(imr.addr_hi) == end)) {
> +                               found = true;
> +                               reg = i;
> +                               break;
> +                       }
> +               }
> +       }
> +
> +       if (found == false) {
> +               ret = -ENODEV;
> +               goto done;
> +       }
> +
> +       /* Tear down the IMR */
> +       imr.addr_lo = 0;
> +       imr.addr_hi = 0;
> +       imr.rmask = IMR_READ_ACCESS_ALL;
> +       imr.wmask = IMR_WRITE_ACCESS_ALL;
> +
> +       ret = imr_write(idev, reg, &imr, false);
> +
> +done:
> +       mutex_unlock(&idev->lock);
> +       return ret;
> +}
> +EXPORT_SYMBOL(imr_remove_range);
> +
> +#ifdef CONFIG_DEBUG_IMR_SELFTEST
> +#define SELFTEST KBUILD_MODNAME ": self_test "
> +
> +/**
> + * imr_self_test_result - Print result string for self test
> + *
> + * @res:       result code - true if test passed false otherwise
> + * @fmt:       format string
> + * ...         variadic argument list
> + */
> +static void __init imr_self_test_result(int res, const char *fmt, ...)
> +{
> +       va_list vlist;
> +
> +       va_start(vlist, fmt);
> +       if (res)
> +               printk(KERN_INFO SELFTEST "pass ");
> +       else
> +               printk(KERN_ERR SELFTEST "fail ");
> +       vprintk(fmt, vlist);
> +       va_end(vlist);
> +       WARN(res == 0, "test failed");
> +}
> +#undef SELFTEST
> +
> +/**
> + * imr_self_test
> + *
> + * Verify IMR self_test with some simple tests to verify overlap,
> + * zero sized allocations and 1 KiB sized areas.
> + *
> + */
> +static void __init imr_self_test(void)
> +{
> +       phys_addr_t base  = virt_to_phys(&_text);
> +       size_t size = virt_to_phys(&__end_rodata) - base;
> +       const char *fmt_over = "overlapped IMR @ (0x%08lx - 0x%08lx)\n";
> +       int idx;
> +
> +       /* Test zero zero */
> +       idx = imr_add_range(0, 0, 0, 0, false);
> +       imr_self_test_result(idx < 0, "zero sized IMR\n");
> +
> +       /* Test exact overlap */
> +       idx = imr_add_range(base, size, IMR_CPU, IMR_CPU, false);
> +       imr_self_test_result(idx < 0, fmt_over, __va(base), __va(base + size));
> +
> +       /* Test overlap with base inside of existing */
> +       base += size - IMR_ALIGN;
> +       idx = imr_add_range(base, size, IMR_CPU, IMR_CPU, false);
> +       imr_self_test_result(idx < 0, fmt_over, __va(base), __va(base + size));
> +
> +       /* Test overlap with end inside of existing */
> +       base -= size + IMR_ALIGN * 2;
> +       idx = imr_add_range(base, size, IMR_CPU, IMR_CPU, false);
> +       imr_self_test_result(idx < 0, fmt_over, __va(base), __va(base + size));
> +
> +       /* Test that a 1 KiB IMR @ zero with read/write all will bomb out */
> +       idx = imr_add_range(0, IMR_ALIGN, IMR_READ_ACCESS_ALL,
> +                           IMR_WRITE_ACCESS_ALL, false);
> +       imr_self_test_result(idx < 0, "1KiB IMR @ 0x00000000 - access-all\n");
> +
> +       /* Test that a 1 KiB IMR @ zero with CPU only will work */
> +       idx = imr_add_range(0, IMR_ALIGN, IMR_CPU, IMR_CPU, false);
> +       imr_self_test_result(idx >= 0, "1KiB IMR @ 0x00000000 - cpu access\n");
> +       if (idx >= 0) {
> +               /* For an index removal address doesn't matter */
> +               idx = imr_remove_range(idx, 0, 0);
> +               imr_self_test_result(idx == 0, "index-teardown - cpu access\n");
> +       }
> +
> +       /* Test 2 KiB works - remove based on index */
> +       size = IMR_ALIGN * 2;
> +       idx = imr_add_range(0, size, IMR_READ_ACCESS_ALL,
> +                           IMR_WRITE_ACCESS_ALL, false);
> +       imr_self_test_result(idx >= 0, "2KiB IMR @ 0x00000000\n");
> +       if (idx >= 0) {
> +               /* For an index removal address doesn't matter */
> +               idx = imr_remove_range(idx, 0, 0);
> +               imr_self_test_result(idx == 0, "index-teardown 2KiB\n");
> +       }
> +
> +       /* Test 1 KiB - remove based on address range */
> +       idx = imr_add_range(0, size, IMR_READ_ACCESS_ALL,
> +                           IMR_WRITE_ACCESS_ALL, false);
> +       imr_self_test_result(idx >= 0, "2KiB IMR @ 0x00000000\n");
> +       if (idx >= 0) {
> +               idx = imr_remove_range(-1, 0, size);
> +               imr_self_test_result(idx == 0, "addr-teardown 2KiB\n");
> +       }
> +}
> +#endif /* CONFIG_DEBUG_IMR_SELFTEST */
> +
> +/**
> + * imr_fixup_memmap - Tear down IMRs used during bootup.
> + *
> + * BIOS and Grub both setup IMRs around compressed kernel, initrd memory
> + * that need to be removed before the kernel hands out one of the IMR
> + * encased addresses to a downstream DMA agent such as the SD or Ethernet.
> + * IMRs on Galileo are setup to immediately reset the system on violation.
> + * As a result if you're running a root filesystem from SD - you'll need
> + * the boot-time IMRs torn down or you'll find seemingly random resets when
> + * using your filesystem.
> + *
> + * @idev:      pointer to imr_device structure
> + * @return:
> + */
> +static void __init imr_fixup_memmap(struct imr_device *idev)
> +{
> +       phys_addr_t base = virt_to_phys(&_text);
> +       size_t size = virt_to_phys(&__end_rodata) - base;
> +       int i;
> +       int ret;
> +
> +       /* Tear down all existing unlocked IMRs */
> +       for (i = 0; i < idev->max_imr; i++)
> +               imr_remove_range(i, 0, 0);
> +
> +       /*
> +        * Setup a locked IMR around the physical extent of the kernel
> +        * from the beginning of the .text secton to the end of the
> +        * .rodata section as one physically contiguous block.
> +        */
> +       ret = imr_add_range(base, size, IMR_CPU, IMR_CPU, true);
> +       if (ret < 0)
> +               pr_err("unable to setup IMR for kernel: (%p - %p)\n",
> +                       &_text, &__end_rodata);
> +       else
> +               pr_info("protecting kernel .text - .rodata: %d KiB (%p - %p)\n",
> +                       size / 1024, &_text, &__end_rodata);
> +
> +}
> +
> +static const struct x86_cpu_id imr_ids[] __initconst = {
> +       { X86_VENDOR_INTEL, 5, 9 },     /* Intel Quark SoC X1000 */
> +       {}
> +};
> +MODULE_DEVICE_TABLE(x86cpu, imr_ids);
> +
> +/**
> + * imr_probe - entry point for IMR driver
> + *
> + * return: -ENODEV for no IMR support 0 if good to go
> + */
> +static int __init imr_init(void)
> +{
> +       struct imr_device *idev = &imr_dev;
> +       int ret;
> +
> +       if (!x86_match_cpu(imr_ids) || !iosf_mbi_available())
> +               return -ENODEV;
> +
> +       idev->max_imr = QUARK_X1000_IMR_MAX;
> +       idev->reg_base = QUARK_X1000_IMR_REGBASE;
> +       idev->init = true;
> +
> +       mutex_init(&idev->lock);
> +       ret = imr_debugfs_register(idev);
> +       if (ret != 0)
> +               pr_warn("debugfs register failed!\n");
> +       imr_fixup_memmap(idev);
> +
> +#ifdef CONFIG_DEBUG_IMR_SELFTEST
> +       /* Run optional self test */
> +       imr_self_test();
> +#endif
> +       return 0;
> +}
> +
> +/**
> + * imr_exit - exit point for IMR code
> + *
> + * Deregisters debugfs, leave IMR state as-is.
> + *
> + * return:
> + */
> +static void __exit imr_exit(void)
> +{
> +       imr_debugfs_unregister(&imr_dev);
> +}
> +
> +module_init(imr_init);
> +module_exit(imr_exit);
> +
> +MODULE_AUTHOR("Bryan O'Donoghue <pure.logic@nexus-software.ie>");
> +MODULE_DESCRIPTION("Intel Isolated Memory Region driver");
> +MODULE_LICENSE("Dual BSD/GPL");
> diff --git a/drivers/platform/x86/Kconfig b/drivers/platform/x86/Kconfig
> index 638e7970..9752761 100644
> --- a/drivers/platform/x86/Kconfig
> +++ b/drivers/platform/x86/Kconfig
> @@ -735,6 +735,31 @@ config INTEL_IPS
>           functionality.  If in doubt, say Y here; it will only load on
>           supported platforms.
>
> +config INTEL_IMR
> +       bool "Intel Isolated Memory Region support"
> +       default n
> +       depends on X86_INTEL_QUARK && IOSF_MBI
> +       ---help---
> +         This option provides a means to manipulate Isolated Memory Regions.
> +         IMRs are a set of registers that define read and write access masks
> +         to prohibit certain system agents from accessing memory with 1 KiB
> +         granularity.
> +
> +         IMRs make it possible to control read/write access to an address
> +         by hardware agents inside the SoC. Read and write masks can be
> +         defined for:
> +               - eSRAM flush
> +               - Dirty CPU snoop (write only)
> +               - RMU access
> +               - PCI Virtual Channel 0/Virtual Channel 1
> +               - SMM mode
> +               - Non SMM mode
> +
> +         Quark contains a set of eight IMR registers and makes use of those
> +         registers during its bootup process.
> +
> +         If you are running on a Galileo/Quark say Y here.
> +
>  config IBM_RTL
>         tristate "Device driver to enable PRTL support"
>         depends on X86 && PCI
> --
> 1.9.1
>



-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH v5 1/2] x86: Add Isolated Memory Regions for Quark X1000

[Bryan O'Donoghue]

> One optional (though I suppose not bad) idea is to split test to
> separate module which might go as a separate patch in the series.
>
> After addressing those comments:
> Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>

No issue with that.

Will generate a V6 based on above feedback