EDAC, pnd2_edac: Add new EDAC driver for Intel SoC platforms

Initial target for this driver is the Intel Apollo Lake platform and
Denverton micro-server, they use the same internal memory controller IP
called Pondicherry2.

Memory controller registers are not in PCI config space like earlier
Intel memory controllers. For Apollo Lake platform they are accessed via
a "side-band" interface, for Denverton micro-server they are access via
PCI config space and memory map I/O. This driver is for Apollo Lake and
Denverton, but only the Denverton is fully enabled while we wait for the
sideband driver.

Apollo lake driver and initial cut at Denverton driver by Tony Luck.
Extensive cleanup, refactoring and basic verification by Qiuxu Zhuo.

Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Qiuxu Zhuo <qiuxu.zhuo@intel.com>
Cc: linux-edac <linux-edac@vger.kernel.org>
Link: http://lkml.kernel.org/r/20170308174539.14432-1-qiuxu.zhuo@intel.com
Signed-off-by: Borislav Petkov <bp@suse.de>

4 files changed, 1853 insertions, 0 deletions

diff --git a/drivers/edac/Kconfig b/drivers/edac/Kconfig
index 82d85cce81f8..be3eac6ad54d 100644
--- a/drivers/edac/Kconfig
+++ b/drivers/edac/Kconfig
@@ -259,6 +259,15 @@ config EDAC_SKX
 	  Support for error detection and correction the Intel
 	  Skylake server Integrated Memory Controllers.
 
+config EDAC_PND2
+	tristate "Intel Pondicherry2"
+	depends on EDAC_MM_EDAC && PCI && X86_64 && X86_MCE_INTEL
+	help
+	  Support for error detection and correction on the Intel
+	  Pondicherry2 Integrated Memory Controller. This SoC IP is
+	  first used on the Apollo Lake platform and Denverton
+	  micro-server but may appear on others in the future.
+
 config EDAC_MPC85XX
 	tristate "Freescale MPC83xx / MPC85xx"
 	depends on EDAC_MM_EDAC && FSL_SOC
diff --git a/drivers/edac/Makefile b/drivers/edac/Makefile
index 88e472e8b9a9..587107e90996 100644
--- a/drivers/edac/Makefile
+++ b/drivers/edac/Makefile
@@ -32,6 +32,7 @@ obj-$(CONFIG_EDAC_I7300)		+= i7300_edac.o
 obj-$(CONFIG_EDAC_I7CORE)		+= i7core_edac.o
 obj-$(CONFIG_EDAC_SBRIDGE)		+= sb_edac.o
 obj-$(CONFIG_EDAC_SKX)			+= skx_edac.o
+obj-$(CONFIG_EDAC_PND2)			+= pnd2_edac.o
 obj-$(CONFIG_EDAC_E7XXX)		+= e7xxx_edac.o
 obj-$(CONFIG_EDAC_E752X)		+= e752x_edac.o
 obj-$(CONFIG_EDAC_I82443BXGX)		+= i82443bxgx_edac.o
diff --git a/drivers/edac/pnd2_edac.c b/drivers/edac/pnd2_edac.c
new file mode 100644
index 000000000000..14d39f05226e
--- /dev/null
+++ b/drivers/edac/pnd2_edac.c
@@ -0,0 +1,1542 @@
+/*
+ * Driver for Pondicherry2 memory controller.
+ *
+ * Copyright (c) 2016, Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * [Derived from sb_edac.c]
+ *
+ * Translation of system physical addresses to DIMM addresses
+ * is a two stage process:
+ *
+ * First the Pondicherry 2 memory controller handles slice and channel interleaving
+ * in "sys2pmi()". This is (almost) completley common between platforms.
+ *
+ * Then a platform specific dunit (DIMM unit) completes the process to provide DIMM,
+ * rank, bank, row and column using the appropriate "dunit_ops" functions/parameters.
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/pci.h>
+#include <linux/pci_ids.h>
+#include <linux/slab.h>
+#include <linux/delay.h>
+#include <linux/edac.h>
+#include <linux/mmzone.h>
+#include <linux/smp.h>
+#include <linux/bitmap.h>
+#include <linux/math64.h>
+#include <linux/mod_devicetable.h>
+#include <asm/cpu_device_id.h>
+#include <asm/intel-family.h>
+#include <asm/processor.h>
+#include <asm/mce.h>
+
+#include "edac_mc.h"
+#include "edac_module.h"
+#include "pnd2_edac.h"
+
+#define APL_NUM_CHANNELS	4
+#define DNV_NUM_CHANNELS	2
+#define DNV_MAX_DIMMS		2 /* Max DIMMs per channel */
+
+enum type {
+	APL,
+	DNV, /* All requests go to PMI CH0 on each slice (CH1 disabled) */
+};
+
+struct dram_addr {
+	int chan;
+	int dimm;
+	int rank;
+	int bank;
+	int row;
+	int col;
+};
+
+struct pnd2_pvt {
+	int dimm_geom[APL_NUM_CHANNELS];
+	u64 tolm, tohm;
+};
+
+/*
+ * System address space is divided into multiple regions with
+ * different interleave rules in each. The as0/as1 regions
+ * have no interleaving at all. The as2 region is interleaved
+ * between two channels. The mot region is magic and may overlap
+ * other regions, with its interleave rules taking precedence.
+ * Addresses not in any of these regions are interleaved across
+ * all four channels.
+ */
+static struct region {
+	u64	base;
+	u64	limit;
+	u8	enabled;
+} mot, as0, as1, as2;
+
+static struct dunit_ops {
+	char *name;
+	enum type type;
+	int pmiaddr_shift;
+	int pmiidx_shift;
+	int channels;
+	int dimms_per_channel;
+	int (*rd_reg)(int port, int off, int op, void *data, size_t sz, char *name);
+	int (*get_registers)(void);
+	int (*check_ecc)(void);
+	void (*mk_region)(char *name, struct region *rp, void *asym);
+	void (*get_dimm_config)(struct mem_ctl_info *mci);
+	int (*pmi2mem)(struct mem_ctl_info *mci, u64 pmiaddr, u32 pmiidx,
+				   struct dram_addr *daddr, char *msg);
+} *ops;
+
+static struct mem_ctl_info *pnd2_mci;
+
+#define PND2_MSG_SIZE	256
+
+/* Debug macros */
+#define pnd2_printk(level, fmt, arg...)			\
+	edac_printk(level, "pnd2", fmt, ##arg)
+
+#define pnd2_mc_printk(mci, level, fmt, arg...)	\
+	edac_mc_chipset_printk(mci, level, "pnd2", fmt, ##arg)
+
+#define MOT_CHAN_INTLV_BIT_1SLC_2CH 12
+#define MOT_CHAN_INTLV_BIT_2SLC_2CH 13
+#define SELECTOR_DISABLED (-1)
+#define _4GB (1ul << 32)
+
+#define PMI_ADDRESS_WIDTH	31
+#define PND_MAX_PHYS_BIT	39
+
+#define APL_ASYMSHIFT		28
+#define DNV_ASYMSHIFT		31
+#define CH_HASH_MASK_LSB	6
+#define SLICE_HASH_MASK_LSB	6
+#define MOT_SLC_INTLV_BIT	12
+#define LOG2_PMI_ADDR_GRANULARITY	5
+#define MOT_SHIFT	24
+
+#define GET_BITFIELD(v, lo, hi)	(((v) & GENMASK_ULL(hi, lo)) >> (lo))
+#define U64_LSHIFT(val, s)	((u64)(val) << (s))
+
+#ifdef CONFIG_X86_INTEL_SBI_APL
+#include "linux/platform_data/sbi_apl.h"
+int sbi_send(int port, int off, int op, u32 *data)
+{
+	struct sbi_apl_message sbi_arg;
+	int ret, read = 0;
+
+	memset(&sbi_arg, 0, sizeof(sbi_arg));
+
+	if (op == 0 || op == 4 || op == 6)
+		read = 1;
+	else
+		sbi_arg.data = *data;
+
+	sbi_arg.opcode = op;
+	sbi_arg.port_address = port;
+	sbi_arg.register_offset = off;
+	ret = sbi_apl_commit(&sbi_arg);
+	if (ret || sbi_arg.status)
+		edac_dbg(2, "sbi_send status=%d ret=%d data=%x\n",
+				 sbi_arg.status, ret, sbi_arg.data);
+
+	if (ret == 0)
+		ret = sbi_arg.status;
+
+	if (ret == 0 && read)
+		*data = sbi_arg.data;
+
+	return ret;
+}
+#else
+int sbi_send(int port, int off, int op, u32 *data)
+{
+	return -EUNATCH;
+}
+#endif
+
+static int apl_rd_reg(int port, int off, int op, void *data, size_t sz, char *name)
+{
+	int	ret = 0;
+
+	edac_dbg(2, "Read %s port=%x off=%x op=%x\n", name, port, off, op);
+	switch (sz) {
+	case 8:
+		ret = sbi_send(port, off + 4, op, (u32 *)(data + 4));
+	case 4:
+		ret = sbi_send(port, off, op, (u32 *)data);
+		pnd2_printk(KERN_DEBUG, "%s=%x%08x ret=%d\n", name,
+					sz == 8 ? *((u32 *)(data + 4)) : 0, *((u32 *)data), ret);
+		break;
+	}
+
+	return ret;
+}
+
+static u64 get_mem_ctrl_hub_base_addr(void)
+{
+	struct b_cr_mchbar_lo_pci lo;
+	struct b_cr_mchbar_hi_pci hi;
+	struct pci_dev *pdev;
+
+	pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 0x1980, NULL);
+	if (pdev) {
+		pci_read_config_dword(pdev, 0x48, (u32 *)&lo);
+		pci_read_config_dword(pdev, 0x4c, (u32 *)&hi);
+		pci_dev_put(pdev);
+	} else {
+		return 0;
+	}
+
+	if (!lo.enable) {
+		edac_dbg(2, "MMIO via memory controller hub base address is disabled!\n");
+		return 0;
+	}
+
+	return U64_LSHIFT(hi.base, 32) | U64_LSHIFT(lo.base, 15);
+}
+
+static u64 get_sideband_reg_base_addr(void)
+{
+	struct pci_dev *pdev;
+	u32 hi, lo;
+
+	pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 0x19dd, NULL);
+	if (pdev) {
+		pci_read_config_dword(pdev, 0x10, &lo);
+		pci_read_config_dword(pdev, 0x14, &hi);
+		pci_dev_put(pdev);
+		return (U64_LSHIFT(hi, 32) | U64_LSHIFT(lo, 0));
+	} else {
+		return 0xfd000000;
+	}
+}
+
+static int dnv_rd_reg(int port, int off, int op, void *data, size_t sz, char *name)
+{
+	struct pci_dev *pdev;
+	char *base;
+	u64 addr;
+
+	if (op == 4) {
+		pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 0x1980, NULL);
+		if (!pdev)
+			return -ENODEV;
+
+		pci_read_config_dword(pdev, off, data);
+		pci_dev_put(pdev);
+	} else {
+		/* MMIO via memory controller hub base address */
+		if (op == 0 && port == 0x4c) {
+			addr = get_mem_ctrl_hub_base_addr();
+			if (!addr)
+				return -ENODEV;
+		} else {
+			/* MMIO via sideband register base address */
+			addr = get_sideband_reg_base_addr();
+			if (!addr)
+				return -ENODEV;
+			addr += (port << 16);
+		}
+
+		base = ioremap((resource_size_t)addr, 0x10000);
+		if (!base)
+			return -ENODEV;
+
+		if (sz == 8)
+			*(u32 *)(data + 4) = *(u32 *)(base + off + 4);
+		*(u32 *)data = *(u32 *)(base + off);
+
+		iounmap(base);
+	}
+
+	edac_dbg(2, "Read %s=%.8x_%.8x\n", name,
+			(sz == 8) ? *(u32 *)(data + 4) : 0, *(u32 *)data);
+
+	return 0;
+}
+
+#define RD_REGP(regp, regname, port)	\
+	ops->rd_reg(port,					\
+		regname##_offset,				\
+		regname##_r_opcode,				\
+		regp, sizeof(struct regname),	\
+		#regname)
+
+#define RD_REG(regp, regname)			\
+	ops->rd_reg(regname ## _port,		\
+		regname##_offset,				\
+		regname##_r_opcode,				\
+		regp, sizeof(struct regname),	\
+		#regname)
+
+static u64 top_lm, top_hm;
+static bool two_slices;
+static bool two_channels; /* Both PMI channels in one slice enabled */
+
+static u8 sym_chan_mask;
+static u8 asym_chan_mask;
+static u8 chan_mask;
+
+static int slice_selector = -1;
+static int chan_selector = -1;
+static u64 slice_hash_mask;
+static u64 chan_hash_mask;
+
+static void mk_region(char *name, struct region *rp, u64 base, u64 limit)
+{
+	rp->enabled = 1;
+	rp->base = base;
+	rp->limit = limit;
+	edac_dbg(2, "Region:%s [%llx, %llx]\n", name, base, limit);
+}
+
+static void mk_region_mask(char *name, struct region *rp, u64 base, u64 mask)
+{
+	if (mask == 0) {
+		pr_info(FW_BUG "MOT mask cannot be zero\n");
+		return;
+	}
+	if (mask != GENMASK_ULL(PND_MAX_PHYS_BIT, __ffs(mask))) {
+		pr_info(FW_BUG "MOT mask not power of two\n");
+		return;
+	}
+	if (base & ~mask) {
+		pr_info(FW_BUG "MOT region base/mask alignment error\n");
+		return;
+	}
+	rp->base = base;
+	rp->limit = (base | ~mask) & GENMASK_ULL(PND_MAX_PHYS_BIT, 0);
+	rp->enabled = 1;
+	edac_dbg(2, "Region:%s [%llx, %llx]\n", name, base, rp->limit);
+}
+
+static bool in_region(struct region *rp, u64 addr)
+{
+	if (!rp->enabled)
+		return false;
+
+	return rp->base <= addr && addr <= rp->limit;
+}
+
+static int gen_sym_mask(struct b_cr_slice_channel_hash *p)
+{
+	int mask = 0;
+
+	if (!p->slice_0_mem_disabled)
+		mask |= p->sym_slice0_channel_enabled;
+
+	if (!p->slice_1_disabled)
+		mask |= p->sym_slice1_channel_enabled << 2;
+
+	if (p->ch_1_disabled || p->enable_pmi_dual_data_mode)
+		mask &= 0x5;
+
+	return mask;
+}
+
+static int gen_asym_mask(struct b_cr_slice_channel_hash *p,
+			 struct b_cr_asym_mem_region0_mchbar *as0,
+			 struct b_cr_asym_mem_region1_mchbar *as1,
+			 struct b_cr_asym_2way_mem_region_mchbar *as2way)
+{
+	const int intlv[] = { 0x5, 0xA, 0x3, 0xC };
+	int mask = 0;
+
+	if (as2way->asym_2way_interleave_enable)
+		mask = intlv[as2way->asym_2way_intlv_mode];
+	if (as0->slice0_asym_enable)
+		mask |= (1 << as0->slice0_asym_channel_select);
+	if (as1->slice1_asym_enable)
+		mask |= (4 << as1->slice1_asym_channel_select);
+	if (p->slice_0_mem_disabled)
+		mask &= 0xc;
+	if (p->slice_1_disabled)
+		mask &= 0x3;
+	if (p->ch_1_disabled || p->enable_pmi_dual_data_mode)
+		mask &= 0x5;
+
+	return mask;
+}
+
+static struct b_cr_tolud_pci tolud;
+static struct b_cr_touud_lo_pci touud_lo;
+static struct b_cr_touud_hi_pci touud_hi;
+static struct b_cr_asym_mem_region0_mchbar asym0;
+static struct b_cr_asym_mem_region1_mchbar asym1;
+static struct b_cr_asym_2way_mem_region_mchbar asym_2way;
+static struct b_cr_mot_out_base_mchbar mot_base;
+static struct b_cr_mot_out_mask_mchbar mot_mask;
+static struct b_cr_slice_channel_hash chash;
+
+/* Apollo Lake dunit */
+/*
+ * Validated on board with just two DIMMs in the [0] and [2] positions
+ * in this array. Other port number matches documentation, but caution
+ * advised.
+ */
+static const int apl_dports[APL_NUM_CHANNELS] = { 0x18, 0x10, 0x11, 0x19 };
+static struct d_cr_drp0 drp0[APL_NUM_CHANNELS];
+
+/* Denverton dunit */
+static const int dnv_dports[DNV_NUM_CHANNELS] = { 0x10, 0x12 };
+static struct d_cr_dsch dsch;
+static struct d_cr_ecc_ctrl ecc_ctrl[DNV_NUM_CHANNELS];
+static struct d_cr_drp drp[DNV_NUM_CHANNELS];
+static struct d_cr_dmap dmap[DNV_NUM_CHANNELS];
+static struct d_cr_dmap1 dmap1[DNV_NUM_CHANNELS];
+static struct d_cr_dmap2 dmap2[DNV_NUM_CHANNELS];
+static struct d_cr_dmap3 dmap3[DNV_NUM_CHANNELS];
+static struct d_cr_dmap4 dmap4[DNV_NUM_CHANNELS];
+static struct d_cr_dmap5 dmap5[DNV_NUM_CHANNELS];
+
+static void apl_mk_region(char *name, struct region *rp, void *asym)
+{
+	struct b_cr_asym_mem_region0_mchbar *a = asym;
+
+	mk_region(name, rp,
+			  U64_LSHIFT(a->slice0_asym_base, APL_ASYMSHIFT),
+			  U64_LSHIFT(a->slice0_asym_limit, APL_ASYMSHIFT) +
+			  GENMASK_ULL(APL_ASYMSHIFT - 1, 0));
+}
+
+static void dnv_mk_region(char *name, struct region *rp, void *asym)
+{
+	struct b_cr_asym_mem_region_denverton *a = asym;
+
+	mk_region(name, rp,
+			  U64_LSHIFT(a->slice_asym_base, DNV_ASYMSHIFT),
+			  U64_LSHIFT(a->slice_asym_limit, DNV_ASYMSHIFT) +
+			  GENMASK_ULL(DNV_ASYMSHIFT - 1, 0));
+}
+
+static int apl_get_registers(void)
+{
+	int i;
+
+	if (RD_REG(&asym_2way, b_cr_asym_2way_mem_region_mchbar))
+		return -ENODEV;
+
+	for (i = 0; i < APL_NUM_CHANNELS; i++)
+		if (RD_REGP(&drp0[i], d_cr_drp0, apl_dports[i]))
+			return -ENODEV;
+
+	return 0;
+}
+
+static int dnv_get_registers(void)
+{
+	int i;
+
+	if (RD_REG(&dsch, d_cr_dsch))
+		return -ENODEV;
+
+	for (i = 0; i < DNV_NUM_CHANNELS; i++)
+		if (RD_REGP(&ecc_ctrl[i], d_cr_ecc_ctrl, dnv_dports[i]) ||
+			RD_REGP(&drp[i], d_cr_drp, dnv_dports[i]) ||
+			RD_REGP(&dmap[i], d_cr_dmap, dnv_dports[i]) ||
+			RD_REGP(&dmap1[i], d_cr_dmap1, dnv_dports[i]) ||
+			RD_REGP(&dmap2[i], d_cr_dmap2, dnv_dports[i]) ||
+			RD_REGP(&dmap3[i], d_cr_dmap3, dnv_dports[i]) ||
+			RD_REGP(&dmap4[i], d_cr_dmap4, dnv_dports[i]) ||
+			RD_REGP(&dmap5[i], d_cr_dmap5, dnv_dports[i]))
+			return -ENODEV;
+
+	return 0;
+}
+
+/*
+ * Read all the h/w config registers once here (they don't
+ * change at run time. Figure out which address ranges have
+ * which interleave characteristics.
+ */
+static int get_registers(void)
+{
+	const int intlv[] = { 10, 11, 12, 12 };
+
+	if (RD_REG(&tolud, b_cr_tolud_pci) ||
+		RD_REG(&touud_lo, b_cr_touud_lo_pci) ||
+		RD_REG(&touud_hi, b_cr_touud_hi_pci) ||
+		RD_REG(&asym0, b_cr_asym_mem_region0_mchbar) ||
+		RD_REG(&asym1, b_cr_asym_mem_region1_mchbar) ||
+		RD_REG(&mot_base, b_cr_mot_out_base_mchbar) ||
+		RD_REG(&mot_mask, b_cr_mot_out_mask_mchbar) ||
+		RD_REG(&chash, b_cr_slice_channel_hash))
+		return -ENODEV;
+
+	if (ops->get_registers())
+		return -ENODEV;
+
+	if (ops->type == DNV) {
+		/* PMI channel idx (always 0) for asymmetric region */
+		asym0.slice0_asym_channel_select = 0;
+		asym1.slice1_asym_channel_select = 0;
+		/* PMI channel bitmap (always 1) for symmetric region */
+		chash.sym_slice0_channel_enabled = 0x1;
+		chash.sym_slice1_channel_enabled = 0x1;
+	}
+
+	if (asym0.slice0_asym_enable)
+		ops->mk_region("as0", &as0, &asym0);
+
+	if (asym1.slice1_asym_enable)
+		ops->mk_region("as1", &as1, &asym1);
+
+	if (asym_2way.asym_2way_interleave_enable) {
+		mk_region("as2way", &as2,
+				  U64_LSHIFT(asym_2way.asym_2way_base, APL_ASYMSHIFT),
+				  U64_LSHIFT(asym_2way.asym_2way_limit, APL_ASYMSHIFT) +
+				  GENMASK_ULL(APL_ASYMSHIFT - 1, 0));
+	}
+
+	if (mot_base.imr_en) {
+		mk_region_mask("mot", &mot,
+					   U64_LSHIFT(mot_base.mot_out_base, MOT_SHIFT),
+					   U64_LSHIFT(mot_mask.mot_out_mask, MOT_SHIFT));
+	}
+
+	top_lm = U64_LSHIFT(tolud.tolud, 20);
+	top_hm = U64_LSHIFT(touud_hi.touud, 32) | U64_LSHIFT(touud_lo.touud, 20);
+
+	two_slices = !chash.slice_1_disabled &&
+				 !chash.slice_0_mem_disabled &&
+				 (chash.sym_slice0_channel_enabled != 0) &&
+				 (chash.sym_slice1_channel_enabled != 0);
+	two_channels = !chash.ch_1_disabled &&
+				 !chash.enable_pmi_dual_data_mode &&
+				 ((chash.sym_slice0_channel_enabled == 3) ||
+				 (chash.sym_slice1_channel_enabled == 3));
+
+	sym_chan_mask = gen_sym_mask(&chash);
+	asym_chan_mask = gen_asym_mask(&chash, &asym0, &asym1, &asym_2way);
+	chan_mask = sym_chan_mask | asym_chan_mask;
+
+	if (two_slices && !two_channels) {
+		if (chash.hvm_mode)
+			slice_selector = 29;
+		else
+			slice_selector = intlv[chash.interleave_mode];
+	} else if (!two_slices && two_channels) {
+		if (chash.hvm_mode)
+			chan_selector = 29;
+		else
+			chan_selector = intlv[chash.interleave_mode];
+	} else if (two_slices && two_channels) {
+		if (chash.hvm_mode) {
+			slice_selector = 29;
+			chan_selector = 30;
+		} else {
+			slice_selector = intlv[chash.interleave_mode];
+			chan_selector = intlv[chash.interleave_mode] + 1;
+		}
+	}
+
+	if (two_slices) {
+		if (!chash.hvm_mode)
+			slice_hash_mask = chash.slice_hash_mask << SLICE_HASH_MASK_LSB;
+		if (!two_channels)
+			slice_hash_mask |= BIT_ULL(slice_selector);
+	}
+
+	if (two_channels) {
+		if (!chash.hvm_mode)
+			chan_hash_mask = chash.ch_hash_mask << CH_HASH_MASK_LSB;
+		if (!two_slices)
+			chan_hash_mask |= BIT_ULL(chan_selector);
+	}
+
+	return 0;
+}
+
+/* Get a contiguous memory address (remove the MMIO gap) */
+static u64 remove_mmio_gap(u64 sys)
+{
+	return (sys < _4GB) ? sys : sys - (_4GB - top_lm);
+}
+
+/* Squeeze out one address bit, shift upper part down to fill gap */
+static void remove_addr_bit(u64 *addr, int bitidx)
+{
+	u64	mask;
+
+	if (bitidx == -1)
+		return;
+
+	mask = (1ull << bitidx) - 1;
+	*addr = ((*addr >> 1) & ~mask) | (*addr & mask);
+}
+
+/* XOR all the bits from addr specified in mask */
+static int hash_by_mask(u64 addr, u64 mask)
+{
+	u64 result = addr & mask;
+
+	result = (result >> 32) ^ result;
+	result = (result >> 16) ^ result;
+	result = (result >> 8) ^ result;
+	result = (result >> 4) ^ result;
+	result = (result >> 2) ^ result;
+	result = (result >> 1) ^ result;
+
+	return (int)result & 1;
+}
+
+/*
+ * First stage decode. Take the system address and figure out which
+ * second stage will deal with it based on interleave modes.
+ */
+static int sys2pmi(const u64 addr, u32 *pmiidx, u64 *pmiaddr, char *msg)
+{
+	u64 contig_addr, contig_base, contig_offset, contig_base_adj;
+	int mot_intlv_bit = two_slices ? MOT_CHAN_INTLV_BIT_2SLC_2CH :
+						MOT_CHAN_INTLV_BIT_1SLC_2CH;
+	int slice_intlv_bit_rm = SELECTOR_DISABLED;
+	int chan_intlv_bit_rm = SELECTOR_DISABLED;
+	/* Determine if address is in the MOT region. */
+	bool mot_hit = in_region(&mot, addr);
+	/* Calculate the number of symmetric regions enabled. */
+	int sym_channels = hweight8(sym_chan_mask);
+
+	/*
+	 * The amount we need to shift the asym base can be determined by the
+	 * number of enabled symmetric channels.
+	 * NOTE: This can only work because symmetric memory is not supposed
+	 * to do a 3-way interleave.
+	 */
+	int sym_chan_shift = sym_channels >> 1;
+
+	/* Give up if address is out of range, or in MMIO gap */
+	if (addr >= (1ul << PND_MAX_PHYS_BIT) ||
+	   (addr >= top_lm && addr < _4GB) || addr >= top_hm) {
+		snprintf(msg, PND2_MSG_SIZE, "Error address 0x%llx is not DRAM", addr);
+		return -EINVAL;
+	}
+
+	/* Get a contiguous memory address (remove the MMIO gap) */
+	contig_addr = remove_mmio_gap(addr);
+
+	if (in_region(&as0, addr)) {
+		*pmiidx = asym0.slice0_asym_channel_select;
+
+		contig_base = remove_mmio_gap(as0.base);
+		contig_offset = contig_addr - contig_base;
+		contig_base_adj = (contig_base >> sym_chan_shift) *
+						  ((chash.sym_slice0_channel_enabled >> (*pmiidx & 1)) & 1);
+		contig_addr = contig_offset + ((sym_channels > 0) ? contig_base_adj : 0ull);
+	} else if (in_region(&as1, addr)) {
+		*pmiidx = 2u + asym1.slice1_asym_channel_select;
+
+		contig_base = remove_mmio_gap(as1.base);
+		contig_offset = contig_addr - contig_base;
+		contig_base_adj = (contig_base >> sym_chan_shift) *
+						  ((chash.sym_slice1_channel_enabled >> (*pmiidx & 1)) & 1);
+		contig_addr = contig_offset + ((sym_channels > 0) ? contig_base_adj : 0ull);
+	} else if (in_region(&as2, addr) && (asym_2way.asym_2way_intlv_mode == 0x3ul)) {
+		bool channel1;
+
+		mot_intlv_bit = MOT_CHAN_INTLV_BIT_1SLC_2CH;
+		*pmiidx = (asym_2way.asym_2way_intlv_mode & 1) << 1;
+		channel1 = mot_hit ? ((bool)((addr >> mot_intlv_bit) & 1)) :
+			hash_by_mask(contig_addr, chan_hash_mask);
+		*pmiidx |= (u32)channel1;
+
+		contig_base = remove_mmio_gap(as2.base);
+		chan_intlv_bit_rm = mot_hit ? mot_intlv_bit : chan_selector;
+		contig_offset = contig_addr - contig_base;
+		remove_addr_bit(&contig_offset, chan_intlv_bit_rm);
+		contig_addr = (contig_base >> sym_chan_shift) + contig_offset;
+	} else {
+		/* Otherwise we're in normal, boring symmetric mode. */
+		*pmiidx = 0u;
+
+		if (two_slices) {
+			bool slice1;
+
+			if (mot_hit) {
+				slice_intlv_bit_rm = MOT_SLC_INTLV_BIT;
+				slice1 = (addr >> MOT_SLC_INTLV_BIT) & 1;
+			} else {
+				slice_intlv_bit_rm = slice_selector;
+				slice1 = hash_by_mask(addr, slice_hash_mask);
+			}
+
+			*pmiidx = (u32)slice1 << 1;
+		}
+
+		if (two_channels) {
+			bool channel1;
+
+			mot_intlv_bit = two_slices ? MOT_CHAN_INTLV_BIT_2SLC_2CH :
+							MOT_CHAN_INTLV_BIT_1SLC_2CH;
+
+			if (mot_hit) {
+				chan_intlv_bit_rm = mot_intlv_bit;
+				channel1 = (addr >> mot_intlv_bit) & 1;
+			} else {
+				chan_intlv_bit_rm = chan_selector;
+				channel1 = hash_by_mask(contig_addr, chan_hash_mask);
+			}
+
+			*pmiidx |= (u32)channel1;
+		}
+	}
+
+	/* Remove the chan_selector bit first */
+	remove_addr_bit(&contig_addr, chan_intlv_bit_rm);
+	/* Remove the slice bit (we remove it second because it must be lower */
+	remove_addr_bit(&contig_addr, slice_intlv_bit_rm);
+	*pmiaddr = contig_addr;
+
+	return 0;
+}
+
+/* Translate PMI address to memory (rank, row, bank, column) */
+#define C(n) (0x10 | (n))	/* column */
+#define B(n) (0x20 | (n))	/* bank */
+#define R(n) (0x40 | (n))	/* row */
+#define RS   (0x80)			/* rank */
+
+/* addrdec values */
+#define AMAP_1KB	0
+#define AMAP_2KB	1
+#define AMAP_4KB	2
+#define AMAP_RSVD	3
+
+/* dden values */
+#define DEN_4Gb		0
+#define DEN_8Gb		2
+
+/* dwid values */
+#define X8		0
+#define X16		1
+
+static struct dimm_geometry {
+	u8	addrdec;
+	u8	dden;
+	u8	dwid;
+	u8	rowbits, colbits;
+	u16	bits[PMI_ADDRESS_WIDTH];
+} dimms[] = {
+	{
+		.addrdec = AMAP_1KB, .dden = DEN_4Gb, .dwid = X16,
+		.rowbits = 15, .colbits = 10,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  B(0),  B(1),  B(2),  R(0),
+			R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),  R(8),  R(9),
+			R(10), C(7),  C(8),  C(9),  R(11), RS,    R(12), R(13), R(14),
+			0,     0,     0,     0
+		}
+	},
+	{
+		.addrdec = AMAP_1KB, .dden = DEN_4Gb, .dwid = X8,
+		.rowbits = 16, .colbits = 10,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  B(0),  B(1),  B(2),  R(0),
+			R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),  R(8),  R(9),
+			R(10), C(7),  C(8),  C(9),  R(11), RS,    R(12), R(13), R(14),
+			R(15), 0,     0,     0
+		}
+	},
+	{
+		.addrdec = AMAP_1KB, .dden = DEN_8Gb, .dwid = X16,
+		.rowbits = 16, .colbits = 10,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  B(0),  B(1),  B(2),  R(0),
+			R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),  R(8),  R(9),
+			R(10), C(7),  C(8),  C(9),  R(11), RS,    R(12), R(13), R(14),
+			R(15), 0,     0,     0
+		}
+	},
+	{
+		.addrdec = AMAP_1KB, .dden = DEN_8Gb, .dwid = X8,
+		.rowbits = 16, .colbits = 11,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  B(0),  B(1),  B(2),  R(0),
+			R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),  R(8),  R(9),
+			R(10), C(7),  C(8),  C(9),  R(11), RS,    C(11), R(12), R(13),
+			R(14), R(15), 0,     0
+		}
+	},
+	{
+		.addrdec = AMAP_2KB, .dden = DEN_4Gb, .dwid = X16,
+		.rowbits = 15, .colbits = 10,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  C(7),  B(0),  B(1),  B(2),
+			R(0),  R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),  R(8),
+			R(9),  R(10), C(8),  C(9),  R(11), RS,    R(12), R(13), R(14),
+			0,     0,     0,     0
+		}
+	},
+	{
+		.addrdec = AMAP_2KB, .dden = DEN_4Gb, .dwid = X8,
+		.rowbits = 16, .colbits = 10,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  C(7),  B(0),  B(1),  B(2),
+			R(0),  R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),  R(8),
+			R(9),  R(10), C(8),  C(9),  R(11), RS,    R(12), R(13), R(14),
+			R(15), 0,     0,     0
+		}
+	},
+	{
+		.addrdec = AMAP_2KB, .dden = DEN_8Gb, .dwid = X16,
+		.rowbits = 16, .colbits = 10,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  C(7),  B(0),  B(1),  B(2),
+			R(0),  R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),  R(8),
+			R(9),  R(10), C(8),  C(9),  R(11), RS,    R(12), R(13), R(14),
+			R(15), 0,     0,     0
+		}
+	},
+	{
+		.addrdec = AMAP_2KB, .dden = DEN_8Gb, .dwid = X8,
+		.rowbits = 16, .colbits = 11,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  C(7),  B(0),  B(1),  B(2),
+			R(0),  R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),  R(8),
+			R(9),  R(10), C(8),  C(9),  R(11), RS,    C(11), R(12), R(13),
+			R(14), R(15), 0,     0
+		}
+	},
+	{
+		.addrdec = AMAP_4KB, .dden = DEN_4Gb, .dwid = X16,
+		.rowbits = 15, .colbits = 10,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  C(7),  C(8),  B(0),  B(1),
+			B(2),  R(0),  R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),
+			R(8),  R(9),  R(10), C(9),  R(11), RS,    R(12), R(13), R(14),
+			0,     0,     0,     0
+		}
+	},
+	{
+		.addrdec = AMAP_4KB, .dden = DEN_4Gb, .dwid = X8,
+		.rowbits = 16, .colbits = 10,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  C(7),  C(8),  B(0),  B(1),
+			B(2),  R(0),  R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),
+			R(8),  R(9),  R(10), C(9),  R(11), RS,    R(12), R(13), R(14),
+			R(15), 0,     0,     0
+		}
+	},
+	{
+		.addrdec = AMAP_4KB, .dden = DEN_8Gb, .dwid = X16,
+		.rowbits = 16, .colbits = 10,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  C(7),  C(8),  B(0),  B(1),
+			B(2),  R(0),  R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),
+			R(8),  R(9),  R(10), C(9),  R(11), RS,    R(12), R(13), R(14),
+			R(15), 0,     0,     0
+		}
+	},
+	{
+		.addrdec = AMAP_4KB, .dden = DEN_8Gb, .dwid = X8,
+		.rowbits = 16, .colbits = 11,
+		.bits = {
+			C(2),  C(3),  C(4),  C(5),  C(6),  C(7),  C(8),  B(0),  B(1),
+			B(2),  R(0),  R(1),  R(2),  R(3),  R(4),  R(5),  R(6),  R(7),
+			R(8),  R(9),  R(10), C(9),  R(11), RS,    C(11), R(12), R(13),
+			R(14), R(15), 0,     0
+		}
+	}
+};
+
+static int bank_hash(u64 pmiaddr, int idx, int shft)
+{
+	int bhash = 0;
+
+	switch (idx) {
+	case 0:
+		bhash ^= ((pmiaddr >> (12 + shft)) ^ (pmiaddr >> (9 + shft))) & 1;
+		break;
+	case 1:
+		bhash ^= (((pmiaddr >> (10 + shft)) ^ (pmiaddr >> (8 + shft))) & 1) << 1;
+		bhash ^= ((pmiaddr >> 22) & 1) << 1;
+		break;
+	case 2:
+		bhash ^= (((pmiaddr >> (13 + shft)) ^ (pmiaddr >> (11 + shft))) & 1) << 2;
+		break;
+	}
+
+	return bhash;
+}
+
+static int rank_hash(u64 pmiaddr)
+{
+	return ((pmiaddr >> 16) ^ (pmiaddr >> 10)) & 1;
+}
+
+/* Second stage decode. Compute rank, bank, row & column. */
+static int apl_pmi2mem(struct mem_ctl_info *mci, u64 pmiaddr, u32 pmiidx,
+		       struct dram_addr *daddr, char *msg)
+{
+	struct d_cr_drp0 *cr_drp0 = &drp0[pmiidx];
+	struct pnd2_pvt *pvt = mci->pvt_info;
+	int g = pvt->dimm_geom[pmiidx];
+	struct dimm_geometry *d = &dimms[g];
+	int column = 0, bank = 0, row = 0, rank = 0;
+	int i, idx, type, skiprs = 0;
+
+	for (i = 0; i < PMI_ADDRESS_WIDTH; i++) {
+		int	bit = (pmiaddr >> i) & 1;
+
+		if (i + skiprs >= PMI_ADDRESS_WIDTH) {
+			snprintf(msg, PND2_MSG_SIZE, "Bad dimm_geometry[] table\n");
+			return -EINVAL;
+		}
+
+		type = d->bits[i + skiprs] & ~0xf;
+		idx = d->bits[i + skiprs] & 0xf;
+
+		/*
+		 * On single rank DIMMs ignore the rank select bit
+		 * and shift remainder of "bits[]" down one place.
+		 */
+		if (type == RS && (cr_drp0->rken0 + cr_drp0->rken1) == 1) {
+			skiprs = 1;
+			type = d->bits[i + skiprs] & ~0xf;
+			idx = d->bits[i + skiprs] & 0xf;
+		}
+
+		switch (type) {
+		case C(0):
+			column |= (bit << idx);
+			break;
+		case B(0):
+			bank |= (bit << idx);
+			if (cr_drp0->bahen)
+				bank ^= bank_hash(pmiaddr, idx, d->addrdec);
+			break;
+		case R(0):
+			row |= (bit << idx);
+			break;
+		case RS:
+			rank = bit;
+			if (cr_drp0->rsien)
+				rank ^= rank_hash(pmiaddr);
+			break;
+		default:
+			if (bit) {
+				snprintf(msg, PND2_MSG_SIZE, "Bad translation\n");
+				return -EINVAL;
+			}
+			goto done;
+		}
+	}
+
+done:
+	daddr->col = column;
+	daddr->bank = bank;
+	daddr->row = row;
+	daddr->rank = rank;
+	daddr->dimm = 0;
+
+	return 0;
+}
+
+/* Pluck bit "in" from pmiaddr and return value shifted to bit "out" */
+#define dnv_get_bit(pmi, in, out) ((int)(((pmi) >> (in)) & 1u) << (out))
+
+static int dnv_pmi2mem(struct mem_ctl_info *mci, u64 pmiaddr, u32 pmiidx,
+					   struct dram_addr *daddr, char *msg)
+{
+	/* Rank 0 or 1 */
+	daddr->rank = dnv_get_bit(pmiaddr, dmap[pmiidx].rs0 + 13, 0);
+	/* Rank 2 or 3 */
+	daddr->rank |= dnv_get_bit(pmiaddr, dmap[pmiidx].rs1 + 13, 1);
+
+	/*
+	 * Normally ranks 0,1 are DIMM0, and 2,3 are DIMM1, but we
+	 * flip them if DIMM1 is larger than DIMM0.
+	 */
+	daddr->dimm = (daddr->rank >= 2) ^ drp[pmiidx].dimmflip;
+
+	daddr->bank = dnv_get_bit(pmiaddr, dmap[pmiidx].ba0 + 6, 0);
+	daddr->bank |= dnv_get_bit(pmiaddr, dmap[pmiidx].ba1 + 6, 1);
+	daddr->bank |= dnv_get_bit(pmiaddr, dmap[pmiidx].bg0 + 6, 2);
+	if (dsch.ddr4en)
+		daddr->bank |= dnv_get_bit(pmiaddr, dmap[pmiidx].bg1 + 6, 3);
+	if (dmap1[pmiidx].bxor) {
+		if (dsch.ddr4en) {
+			daddr->bank ^= dnv_get_bit(pmiaddr, dmap3[pmiidx].row6 + 6, 0);
+			daddr->bank ^= dnv_get_bit(pmiaddr, dmap3[pmiidx].row7 + 6, 1);
+			if (dsch.chan_width == 0)
+				/* 64/72 bit dram channel width */
+				daddr->bank ^= dnv_get_bit(pmiaddr, dmap5[pmiidx].ca3 + 6, 2);
+			else
+				/* 32/40 bit dram channel width */
+				daddr->bank ^= dnv_get_bit(pmiaddr, dmap5[pmiidx].ca4 + 6, 2);
+			daddr->bank ^= dnv_get_bit(pmiaddr, dmap2[pmiidx].row2 + 6, 3);
+		} else {
+			daddr->bank ^= dnv_get_bit(pmiaddr, dmap2[pmiidx].row2 + 6, 0);
+			daddr->bank ^= dnv_get_bit(pmiaddr, dmap3[pmiidx].row6 + 6, 1);
+			if (dsch.chan_width == 0)
+				daddr->bank ^= dnv_get_bit(pmiaddr, dmap5[pmiidx].ca3 + 6, 2);
+			else
+				daddr->bank ^= dnv_get_bit(pmiaddr, dmap5[pmiidx].ca4 + 6, 2);
+		}
+	}
+
+	daddr->row = dnv_get_bit(pmiaddr, dmap2[pmiidx].row0 + 6, 0);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap2[pmiidx].row1 + 6, 1);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap2[pmiidx].row2 + 6, 2);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap2[pmiidx].row3 + 6, 3);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap2[pmiidx].row4 + 6, 4);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap2[pmiidx].row5 + 6, 5);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap3[pmiidx].row6 + 6, 6);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap3[pmiidx].row7 + 6, 7);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap3[pmiidx].row8 + 6, 8);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap3[pmiidx].row9 + 6, 9);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap3[pmiidx].row10 + 6, 10);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap3[pmiidx].row11 + 6, 11);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap4[pmiidx].row12 + 6, 12);
+	daddr->row |= dnv_get_bit(pmiaddr, dmap4[pmiidx].row13 + 6, 13);
+	if (dmap4[pmiidx].row14 != 31)
+		daddr->row |= dnv_get_bit(pmiaddr, dmap4[pmiidx].row14 + 6, 14);
+	if (dmap4[pmiidx].row15 != 31)
+		daddr->row |= dnv_get_bit(pmiaddr, dmap4[pmiidx].row15 + 6, 15);
+	if (dmap4[pmiidx].row16 != 31)
+		daddr->row |= dnv_get_bit(pmiaddr, dmap4[pmiidx].row16 + 6, 16);
+	if (dmap4[pmiidx].row17 != 31)
+		daddr->row |= dnv_get_bit(pmiaddr, dmap4[pmiidx].row17 + 6, 17);
+
+	daddr->col = dnv_get_bit(pmiaddr, dmap5[pmiidx].ca3 + 6, 3);
+	daddr->col |= dnv_get_bit(pmiaddr, dmap5[pmiidx].ca4 + 6, 4);
+	daddr->col |= dnv_get_bit(pmiaddr, dmap5[pmiidx].ca5 + 6, 5);
+	daddr->col |= dnv_get_bit(pmiaddr, dmap5[pmiidx].ca6 + 6, 6);
+	daddr->col |= dnv_get_bit(pmiaddr, dmap5[pmiidx].ca7 + 6, 7);
+	daddr->col |= dnv_get_bit(pmiaddr, dmap5[pmiidx].ca8 + 6, 8);
+	daddr->col |= dnv_get_bit(pmiaddr, dmap5[pmiidx].ca9 + 6, 9);
+	if (!dsch.ddr4en && dmap1[pmiidx].ca11 != 0x3f)
+		daddr->col |= dnv_get_bit(pmiaddr, dmap1[pmiidx].ca11 + 13, 11);
+
+	return 0;
+}
+
+static int check_channel(int ch)
+{
+	if (drp0[ch].dramtype != 0) {
+		pnd2_printk(KERN_INFO, "Unsupported DIMM in channel %d\n", ch);
+		return 1;
+	} else if (drp0[ch].eccen == 0) {
+		pnd2_printk(KER