EDAC/igen6: Add Intel Tiger Lake SoC support

Tiger Lake SoC shares the same memory controller and In-Band ECC
(IBECC) IP with Elkhart Lake SoC. The main differences are that Tiger
Lake has two memory controllers each associated with one IBECC and
uses Machine Check for the memory error notification.

So add Tiger Lake compute die IDs, MCE decoding chain registration,
and memory slice decoding for Tiger Lake EDAC support.

Signed-off-by: Qiuxu Zhuo <qiuxu.zhuo@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Link: https://lore.kernel.org/r/20210611170123.1057025-6-tony.luck@intel.com

1 files changed, 253 insertions, 20 deletions

diff --git a/drivers/edac/igen6_edac.c b/drivers/edac/igen6_edac.c
index 0fedf2d206d3..9f653ace50e7 100644
--- a/drivers/edac/igen6_edac.c
+++ b/drivers/edac/igen6_edac.c
@@ -22,6 +22,7 @@
 #include <linux/io.h>
 #include <asm/mach_traps.h>
 #include <asm/nmi.h>
+#include <asm/mce.h>
 
 #include "edac_mc.h"
 #include "edac_module.h"
@@ -40,7 +41,7 @@
 
 #define GET_BITFIELD(v, lo, hi) (((v) & GENMASK_ULL(hi, lo)) >> (lo))
 
-#define NUM_IMC				1 /* Max memory controllers */
+#define NUM_IMC				2 /* Max memory controllers */
 #define NUM_CHANNELS			2 /* Max channels */
 #define NUM_DIMMS			2 /* Max DIMMs per channel */
 
@@ -54,6 +55,10 @@
 #define CAPID_C_OFFSET			0xec
 #define CAPID_C_IBECC			BIT(15)
 
+/* Capability register E */
+#define CAPID_E_OFFSET			0xf0
+#define CAPID_E_IBECC			BIT(12)
+
 /* Error Status */
 #define ERRSTS_OFFSET			0xc8
 #define ERRSTS_CE			BIT_ULL(6)
@@ -109,12 +114,20 @@
 #define CHANNEL_HASH_LSB_MASK_BIT(v)	GET_BITFIELD(v, 24, 26)
 #define CHANNEL_HASH_MODE(v)		GET_BITFIELD(v, 28, 28)
 
+/* Parameters for memory slice decode stage */
+#define MEM_SLICE_HASH_MASK(v)		(GET_BITFIELD(v, 6, 19) << 6)
+#define MEM_SLICE_HASH_LSB_MASK_BIT(v)	GET_BITFIELD(v, 24, 26)
+
 static struct res_config {
+	bool machine_check;
 	int num_imc;
+	u32 cmf_base;
+	u32 cmf_size;
+	u32 ms_hash_offset;
 	u32 ibecc_base;
 	bool (*ibecc_available)(struct pci_dev *pdev);
 	/* Convert error address logged in IBECC to system physical address */
-	u64 (*err_addr_to_sys_addr)(u64 eaddr);
+	u64 (*err_addr_to_sys_addr)(u64 eaddr, int mc);
 	/* Convert error address logged in IBECC to integrated memory controller address */
 	u64 (*err_addr_to_imc_addr)(u64 eaddr);
 } *res_cfg;
@@ -125,6 +138,7 @@ struct igen6_imc {
 	struct pci_dev *pdev;
 	struct device dev;
 	void __iomem *window;
+	u64 size;
 	u64 ch_s_size;
 	int ch_l_map;
 	u64 dimm_s_size[NUM_CHANNELS];
@@ -134,6 +148,9 @@ struct igen6_imc {
 
 static struct igen6_pvt {
 	struct igen6_imc imc[NUM_IMC];
+	u64 ms_hash;
+	u64 ms_s_size;
+	int ms_l_map;
 } *igen6_pvt;
 
 /* The top of low usable DRAM */
@@ -189,6 +206,9 @@ static struct work_struct ecclog_work;
 #define DID_ICL_SKU11	0x4589
 #define DID_ICL_SKU12	0x458d
 
+/* Compute die IDs for Tiger Lake with IBECC */
+#define DID_TGL_SKU	0x9a14
+
 static bool ehl_ibecc_available(struct pci_dev *pdev)
 {
 	u32 v;
@@ -199,7 +219,7 @@ static bool ehl_ibecc_available(struct pci_dev *pdev)
 	return !!(CAPID_C_IBECC & v);
 }
 
-static u64 ehl_err_addr_to_sys_addr(u64 eaddr)
+static u64 ehl_err_addr_to_sys_addr(u64 eaddr, int mc)
 {
 	return eaddr;
 }
@@ -229,20 +249,103 @@ static bool icl_ibecc_available(struct pci_dev *pdev)
 		(boot_cpu_data.x86_stepping >= 1);
 }
 
+static bool tgl_ibecc_available(struct pci_dev *pdev)
+{
+	u32 v;
+
+	if (pci_read_config_dword(pdev, CAPID_E_OFFSET, &v))
+		return false;
+
+	return !(CAPID_E_IBECC & v);
+}
+
+static u64 mem_addr_to_sys_addr(u64 maddr)
+{
+	if (maddr < igen6_tolud)
+		return maddr;
+
+	if (igen6_tom <= _4GB)
+		return maddr - igen6_tolud + _4GB;
+
+	if (maddr < _4GB)
+		return maddr - igen6_tolud + igen6_tom;
+
+	return maddr;
+}
+
+static u64 mem_slice_hash(u64 addr, u64 mask, u64 hash_init, int intlv_bit)
+{
+	u64 hash_addr = addr & mask, hash = hash_init;
+	u64 intlv = (addr >> intlv_bit) & 1;
+	int i;
+
+	for (i = 6; i < 20; i++)
+		hash ^= (hash_addr >> i) & 1;
+
+	return hash ^ intlv;
+}
+
+static u64 tgl_err_addr_to_mem_addr(u64 eaddr, int mc)
+{
+	u64 maddr, hash, mask, ms_s_size;
+	int intlv_bit;
+	u32 ms_hash;
+
+	ms_s_size = igen6_pvt->ms_s_size;
+	if (eaddr >= ms_s_size)
+		return eaddr + ms_s_size;
+
+	ms_hash = igen6_pvt->ms_hash;
+
+	mask = MEM_SLICE_HASH_MASK(ms_hash);
+	intlv_bit = MEM_SLICE_HASH_LSB_MASK_BIT(ms_hash) + 6;
+
+	maddr = GET_BITFIELD(eaddr, intlv_bit, 63) << (intlv_bit + 1) |
+		GET_BITFIELD(eaddr, 0, intlv_bit - 1);
+
+	hash = mem_slice_hash(maddr, mask, mc, intlv_bit);
+
+	return maddr | (hash << intlv_bit);
+}
+
+static u64 tgl_err_addr_to_sys_addr(u64 eaddr, int mc)
+{
+	u64 maddr = tgl_err_addr_to_mem_addr(eaddr, mc);
+
+	return mem_addr_to_sys_addr(maddr);
+}
+
+static u64 tgl_err_addr_to_imc_addr(u64 eaddr)
+{
+	return eaddr;
+}
+
 static struct res_config ehl_cfg = {
-	.num_imc	 = 1,
-	.ibecc_base	 = 0xdc00,
-	.ibecc_available = ehl_ibecc_available,
-	.err_addr_to_sys_addr  = ehl_err_addr_to_sys_addr,
-	.err_addr_to_imc_addr  = ehl_err_addr_to_imc_addr,
+	.num_imc		= 1,
+	.ibecc_base		= 0xdc00,
+	.ibecc_available	= ehl_ibecc_available,
+	.err_addr_to_sys_addr	= ehl_err_addr_to_sys_addr,
+	.err_addr_to_imc_addr	= ehl_err_addr_to_imc_addr,
 };
 
 static struct res_config icl_cfg = {
-	.num_imc	 = 1,
-	.ibecc_base	 = 0xd800,
-	.ibecc_available = icl_ibecc_available,
-	.err_addr_to_sys_addr  = ehl_err_addr_to_sys_addr,
-	.err_addr_to_imc_addr  = ehl_err_addr_to_imc_addr,
+	.num_imc		= 1,
+	.ibecc_base		= 0xd800,
+	.ibecc_available	= icl_ibecc_available,
+	.err_addr_to_sys_addr	= ehl_err_addr_to_sys_addr,
+	.err_addr_to_imc_addr	= ehl_err_addr_to_imc_addr,
+};
+
+static struct res_config tgl_cfg = {
+	.machine_check		= true,
+	.num_imc		= 2,
+	.cmf_base		= 0x11000,
+	.cmf_size		= 0x800,
+	.ms_hash_offset		= 0xac,
+	.ibecc_base		= 0xd400,
+	.ibecc_available	= tgl_ibecc_available,
+	.err_addr_to_sys_addr	= tgl_err_addr_to_sys_addr,
+	.err_addr_to_imc_addr	= tgl_err_addr_to_imc_addr,
 };
 
 static const struct pci_device_id igen6_pci_tbl[] = {
@@ -261,6 +364,7 @@ static const struct pci_device_id igen6_pci_tbl[] = {
 	{ PCI_VDEVICE(INTEL, DID_ICL_SKU10), (kernel_ulong_t)&icl_cfg },
 	{ PCI_VDEVICE(INTEL, DID_ICL_SKU11), (kernel_ulong_t)&icl_cfg },
 	{ PCI_VDEVICE(INTEL, DID_ICL_SKU12), (kernel_ulong_t)&icl_cfg },
+	{ PCI_VDEVICE(INTEL, DID_TGL_SKU), (kernel_ulong_t)&tgl_cfg },
 	{ },
 };
 MODULE_DEVICE_TABLE(pci, igen6_pci_tbl);
@@ -519,7 +623,7 @@ static void ecclog_work_cb(struct work_struct *work)
 		eaddr = ECC_ERROR_LOG_ADDR(node->ecclog) <<
 			ECC_ERROR_LOG_ADDR_SHIFT;
 		res.mc	     = node->mc;
-		res.sys_addr = res_cfg->err_addr_to_sys_addr(eaddr);
+		res.sys_addr = res_cfg->err_addr_to_sys_addr(eaddr, res.mc);
 		res.imc_addr = res_cfg->err_addr_to_imc_addr(eaddr);
 
 		mci = igen6_pvt->imc[res.mc].mci;
@@ -569,6 +673,57 @@ static int ecclog_nmi_handler(unsigned int cmd, struct pt_regs *regs)
 	return NMI_HANDLED;
 }
 
+static int ecclog_mce_handler(struct notifier_block *nb, unsigned long val,
+			      void *data)
+{
+	struct mce *mce = (struct mce *)data;
+	char *type;
+
+	if (mce->kflags & MCE_HANDLED_CEC)
+		return NOTIFY_DONE;
+
+	/*
+	 * Ignore unless this is a memory related error.
+	 * We don't check the bit MCI_STATUS_ADDRV of MCi_STATUS here,
+	 * since this bit isn't set on some CPU (e.g., Tiger Lake UP3).
+	 */
+	if ((mce->status & 0xefff) >> 7 != 1)
+		return NOTIFY_DONE;
+
+	if (mce->mcgstatus & MCG_STATUS_MCIP)
+		type = "Exception";
+	else
+		type = "Event";
+
+	edac_dbg(0, "CPU %d: Machine Check %s: 0x%llx Bank %d: 0x%llx\n",
+		 mce->extcpu, type, mce->mcgstatus,
+		 mce->bank, mce->status);
+	edac_dbg(0, "TSC 0x%llx\n", mce->tsc);
+	edac_dbg(0, "ADDR 0x%llx\n", mce->addr);
+	edac_dbg(0, "MISC 0x%llx\n", mce->misc);
+	edac_dbg(0, "PROCESSOR %u:0x%x TIME %llu SOCKET %u APIC 0x%x\n",
+		 mce->cpuvendor, mce->cpuid, mce->time,
+		 mce->socketid, mce->apicid);
+	/*
+	 * We just use the Machine Check for the memory error notification.
+	 * Each memory controller is associated with an IBECC instance.
+	 * Directly read and clear the error information(error address and
+	 * error type) on all the IBECC instances so that we know on which
+	 * memory controller the memory error(s) occurred.
+	 */
+	if (!ecclog_handler())
+		return NOTIFY_DONE;
+
+	mce->kflags |= MCE_HANDLED_EDAC;
+
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block ecclog_mce_dec = {
+	.notifier_call	= ecclog_mce_handler,
+	.priority	= MCE_PRIO_EDAC,
+};
+
 static bool igen6_check_ecc(struct igen6_imc *imc)
 {
 	u32 activate = readl(imc->window + IBECC_ACTIVATE_OFFSET);
@@ -602,6 +757,8 @@ static int igen6_get_dimm_config(struct mem_ctl_info *mci)
 		imc->dimm_l_size[i] = MAD_DIMM_CH_DIMM_L_SIZE(mad_dimm);
 		imc->dimm_s_size[i] = MAD_DIMM_CH_DIMM_S_SIZE(mad_dimm);
 		imc->dimm_l_map[i]  = MAD_INTRA_CH_DIMM_L_MAP(mad_intra);
+		imc->size += imc->dimm_s_size[i];
+		imc->size += imc->dimm_l_size[i];
 		ndimms = 0;
 
 		for (j = 0; j < NUM_DIMMS; j++) {
@@ -637,6 +794,8 @@ static int igen6_get_dimm_config(struct mem_ctl_info *mci)
 		}
 	}
 
+	edac_dbg(0, "MC %d, total size %llu MiB\n", mc, imc->size >> 20);
+
 	return 0;
 }
 
@@ -886,6 +1045,77 @@ static void igen6_unregister_mcis(void)
 	}
 }
 
+static int igen6_mem_slice_setup(u64 mchbar)
+{
+	struct igen6_imc *imc = &igen6_pvt->imc[0];
+	u64 base = mchbar + res_cfg->cmf_base;
+	u32 offset = res_cfg->ms_hash_offset;
+	u32 size = res_cfg->cmf_size;
+	u64 ms_s_size, ms_hash;
+	void __iomem *cmf;
+	int ms_l_map;
+
+	edac_dbg(2, "\n");
+
+	if (imc[0].size < imc[1].size) {
+		ms_s_size = imc[0].size;
+		ms_l_map  = 1;
+	} else {
+		ms_s_size = imc[1].size;
+		ms_l_map  = 0;
+	}
+
+	igen6_pvt->ms_s_size = ms_s_size;
+	igen6_pvt->ms_l_map  = ms_l_map;
+
+	edac_dbg(0, "ms_s_size: %llu MiB, ms_l_map %d\n",
+		 ms_s_size >> 20, ms_l_map);
+
+	cmf = ioremap(base, size);
+	if (!cmf) {
+		igen6_printk(KERN_ERR, "Failed to ioremap cmf 0x%llx\n", base);
+		return -ENODEV;
+	}
+
+	ms_hash = readq(cmf + offset);
+	igen6_pvt->ms_hash = ms_hash;
+
+	edac_dbg(0, "MEM_SLICE_HASH: 0x%llx\n", ms_hash);
+
+	iounmap(cmf);
+
+	return 0;
+}
+
+static int register_err_handler(void)
+{
+	int rc;
+
+	if (res_cfg->machine_check) {
+		mce_register_decode_chain(&ecclog_mce_dec);
+		return 0;
+	}
+
+	rc = register_nmi_handler(NMI_SERR, ecclog_nmi_handler,
+				  0, IGEN6_NMI_NAME);
+	if (rc) {
+		igen6_printk(KERN_ERR, "Failed to register NMI handler\n");
+		return rc;
+	}
+
+	return 0;
+}
+
+static void unregister_err_handler(void)
+{
+	if (res_cfg->machine_check) {
+		mce_unregister_decode_chain(&ecclog_mce_dec);
+		return;
+	}
+
+	unregister_nmi_handler(NMI_SERR, IGEN6_NMI_NAME);
+}
+
 static int igen6_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 {
 	u64 mchbar;
@@ -909,6 +1139,12 @@ static int igen6_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 			goto fail2;
 	}
 
+	if (res_cfg->num_imc > 1) {
+		rc = igen6_mem_slice_setup(mchbar);
+		if (rc)
+			goto fail2;
+	}
+
 	ecclog_pool = ecclog_gen_pool_create();
 	if (!ecclog_pool) {
 		rc = -ENOMEM;
@@ -921,12 +1157,9 @@ static int igen6_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	/* Check if any pending errors before registering the NMI handler */
 	ecclog_handler();
 
-	rc = register_nmi_handler(NMI_SERR, ecclog_nmi_handler,
-				  0, IGEN6_NMI_NAME);
-	if (rc) {
-		igen6_printk(KERN_ERR, "Failed to register NMI handler\n");
+	rc = register_err_handler();
+	if (rc)
 		goto fail3;
-	}
 
 	/* Enable error reporting */
 	rc = errcmd_enable_error_reporting(true);
@@ -954,7 +1187,7 @@ static void igen6_remove(struct pci_dev *pdev)
 
 	igen6_debug_teardown();
 	errcmd_enable_error_reporting(false);
-	unregister_nmi_handler(NMI_SERR, IGEN6_NMI_NAME);
+	unregister_err_handler();
 	irq_work_sync(&ecclog_irq_work);
 	flush_work(&ecclog_work);
 	gen_pool_destroy(ecclog_pool);