x86: apl: Add core init for the SoC

Set up MSRs required for Apollo Lake. This enables Linux to use the
timers correctly. Also write the fixed MSRs for this platform.

Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

7 files changed, 170 insertions, 21 deletions

diff --git a/arch/x86/cpu/apollolake/cpu.c b/arch/x86/cpu/apollolake/cpu.c
index a4c9c96..d37f91d 100644
--- a/arch/x86/cpu/apollolake/cpu.c
+++ b/arch/x86/cpu/apollolake/cpu.c
@@ -13,6 +13,9 @@
 #include <asm/cpu_x86.h>
 #include <asm/intel_acpi.h>
 #include <asm/msr.h>
+#include <asm/mtrr.h>
+#include <asm/arch/cpu.h>
+#include <asm/arch/iomap.h>
 #include <dm/acpi.h>
 
 #define CSTATE_RES(address_space, width, offset, address)		\
@@ -86,6 +89,86 @@ static int acpi_cpu_fill_ssdt(const struct udevice *dev, struct acpi_ctx *ctx)
 	return 0;
 }
 
+static void update_fixed_mtrrs(void)
+{
+	native_write_msr(MTRR_FIX_64K_00000_MSR,
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK),
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK));
+	native_write_msr(MTRR_FIX_16K_80000_MSR,
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK),
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK));
+	native_write_msr(MTRR_FIX_4K_E0000_MSR,
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK),
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK));
+	native_write_msr(MTRR_FIX_4K_E8000_MSR,
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK),
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK));
+	native_write_msr(MTRR_FIX_4K_F0000_MSR,
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK),
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK));
+	native_write_msr(MTRR_FIX_4K_F8000_MSR,
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK),
+			 MTRR_FIX_TYPE(MTRR_TYPE_WRBACK));
+}
+
+static void setup_core_msrs(void)
+{
+	wrmsrl(MSR_PMG_CST_CONFIG_CONTROL,
+	       PKG_C_STATE_LIMIT_C2_MASK | CORE_C_STATE_LIMIT_C10_MASK |
+	       IO_MWAIT_REDIRECT_MASK | CST_CFG_LOCK_MASK);
+	/* Power Management I/O base address for I/O trapping to C-states */
+	wrmsrl(MSR_PMG_IO_CAPTURE_ADR, ACPI_PMIO_CST_REG |
+	       (PMG_IO_BASE_CST_RNG_BLK_SIZE << 16));
+	/* Disable C1E */
+	msr_clrsetbits_64(MSR_POWER_CTL, 0x2, 0);
+	/* Disable support for MONITOR and MWAIT instructions */
+	msr_clrsetbits_64(MSR_IA32_MISC_ENABLE, MISC_ENABLE_MWAIT, 0);
+	/*
+	 * Enable and Lock the Advanced Encryption Standard (AES-NI)
+	 * feature register
+	 */
+	msr_clrsetbits_64(MSR_FEATURE_CONFIG, FEATURE_CONFIG_RESERVED_MASK,
+			  FEATURE_CONFIG_LOCK);
+
+	update_fixed_mtrrs();
+}
+
+static int soc_core_init(void)
+{
+	struct udevice *pmc;
+	int ret;
+
+	/* Clear out pending MCEs */
+	cpu_mca_configure();
+
+	/* Set core MSRs */
+	setup_core_msrs();
+	/*
+	 * Enable ACPI PM timer emulation, which also lets microcode know
+	 * location of ACPI_BASE_ADDRESS. This also enables other features
+	 * implemented in microcode.
+	 */
+	ret = uclass_first_device_err(UCLASS_ACPI_PMC, &pmc);
+	if (ret)
+		return log_msg_ret("PMC", ret);
+	enable_pm_timer_emulation(pmc);
+
+	return 0;
+}
+
+static int cpu_apl_probe(struct udevice *dev)
+{
+	if (gd->flags & GD_FLG_RELOC) {
+		int ret;
+
+		ret = soc_core_init();
+		if (ret)
+			return log_ret(ret);
+	}
+
+	return 0;
+}
+
 struct acpi_ops apl_cpu_acpi_ops = {
 	.fill_ssdt	= acpi_cpu_fill_ssdt,
 };
@@ -107,6 +190,7 @@ U_BOOT_DRIVER(intel_apl_cpu) = {
 	.id		= UCLASS_CPU,
 	.of_match	= cpu_x86_apl_ids,
 	.bind		= cpu_x86_bind,
+	.probe		= cpu_apl_probe,
 	.ops		= &cpu_x86_apl_ops,
 	ACPI_OPS_PTR(&apl_cpu_acpi_ops)
 	.flags		= DM_FLAG_PRE_RELOC,
diff --git a/arch/x86/cpu/apollolake/cpu_common.c b/arch/x86/cpu/apollolake/cpu_common.c
index ba6bda3..63f6999 100644
--- a/arch/x86/cpu/apollolake/cpu_common.c
+++ b/arch/x86/cpu/apollolake/cpu_common.c
@@ -4,8 +4,13 @@
  */
 
 #include <common.h>
+#include <dm.h>
+#include <log.h>
 #include <asm/cpu_common.h>
 #include <asm/msr.h>
+#include <asm/arch/cpu.h>
+#include <asm/arch/iomap.h>
+#include <power/acpi_pmc.h>
 
 void cpu_flush_l1d_to_l2(void)
 {
@@ -15,3 +20,23 @@ void cpu_flush_l1d_to_l2(void)
 	msr.lo |= FLUSH_DL1_L2;
 	msr_write(MSR_POWER_MISC, msr);
 }
+
+void enable_pm_timer_emulation(const struct udevice *pmc)
+{
+	struct acpi_pmc_upriv *upriv = dev_get_uclass_priv(pmc);
+	msr_t msr;
+
+	/*
+	 * The derived frequency is calculated as follows:
+	 *    (CTC_FREQ * msr[63:32]) >> 32 = target frequency.
+	 *
+	 * Back-solve the multiplier so the 3.579545MHz ACPI timer frequency is
+	 * used.
+	 */
+	msr.hi = (3579545ULL << 32) / CTC_FREQ;
+
+	/* Set PM1 timer IO port and enable */
+	msr.lo = EMULATE_PM_TMR_EN | (upriv->acpi_base + R_ACPI_PM1_TMR);
+	debug("PM timer %x %x\n", msr.hi, msr.lo);
+	msr_write(MSR_EMULATE_PM_TIMER, msr);
+}
diff --git a/arch/x86/cpu/apollolake/cpu_spl.c b/arch/x86/cpu/apollolake/cpu_spl.c
index 9f32f2e..fafe4db 100644
--- a/arch/x86/cpu/apollolake/cpu_spl.c
+++ b/arch/x86/cpu/apollolake/cpu_spl.c
@@ -114,26 +114,6 @@ static int fast_spi_cache_bios_region(void)
 	return 0;
 }
 
-static void enable_pm_timer_emulation(struct udevice *pmc)
-{
-	struct acpi_pmc_upriv *upriv = dev_get_uclass_priv(pmc);
-	msr_t msr;
-
-	/*
-	 * The derived frequency is calculated as follows:
-	 *    (CTC_FREQ * msr[63:32]) >> 32 = target frequency.
-	 *
-	 * Back-solve the multiplier so the 3.579545MHz ACPI timer frequency is
-	 * used.
-	 */
-	msr.hi = (3579545ULL << 32) / CTC_FREQ;
-
-	/* Set PM1 timer IO port and enable */
-	msr.lo = EMULATE_PM_TMR_EN | (upriv->acpi_base + R_ACPI_PM1_TMR);
-	debug("PM timer %x %x\n", msr.hi, msr.lo);
-	msr_write(MSR_EMULATE_PM_TIMER, msr);
-}
-
 static void google_chromeec_ioport_range(uint *out_basep, uint *out_sizep)
 {
 	uint base;
diff --git a/arch/x86/cpu/intel_common/cpu.c b/arch/x86/cpu/intel_common/cpu.c
index 39aa0f6..a51bf86 100644
--- a/arch/x86/cpu/intel_common/cpu.c
+++ b/arch/x86/cpu/intel_common/cpu.c
@@ -306,3 +306,22 @@ int cpu_get_cores_per_package(void)
 
 	return cores;
 }
+
+void cpu_mca_configure(void)
+{
+	msr_t msr;
+	int i;
+	int num_banks;
+
+	msr = msr_read(MSR_IA32_MCG_CAP);
+	num_banks = msr.lo & 0xff;
+	msr.lo = 0;
+	msr.hi = 0;
+	for (i = 0; i < num_banks; i++) {
+		/* Clear the machine check status */
+		msr_write(MSR_IA32_MC0_STATUS + (i * 4), msr);
+		/* Initialise machine checks */
+		msr_write(MSR_IA32_MC0_CTL + i * 4,
+			  (msr_t) {.lo = 0xffffffff, .hi = 0xffffffff});
+	}
+}
diff --git a/arch/x86/include/asm/arch-apollolake/cpu.h b/arch/x86/include/asm/arch-apollolake/cpu.h
index 5e906c5..67d48c6 100644
--- a/arch/x86/include/asm/arch-apollolake/cpu.h
+++ b/arch/x86/include/asm/arch-apollolake/cpu.h
@@ -15,6 +15,20 @@
 #ifndef __ASSEMBLY__
 /* Flush L1D to L2 */
 void cpu_flush_l1d_to_l2(void);
+
+/**
+ * Enable emulation of the PM timer
+ *
+ * Some legacy OSes cannot tolerate the ACPI timer stoping during idle states,
+ * and this results in higher power consumption. ACPI timer emulation allows
+ * disabling of the ACPI Timer (PM1_TMR) to have no impact on the system, with
+ * the exception that TMR_STS will not be set on an overflow condition. All
+ * aligned 32-bit reads from the ACPI Timer port are valid and will behave as if
+ * the ACPI timer remains enabled.
+ *
+ * @pmc: PMC device
+ */
+void enable_pm_timer_emulation(const struct udevice *pmc);
 #endif
 
 #endif /* _ASM_ARCH_CPU_H */
diff --git a/arch/x86/include/asm/cpu_common.h b/arch/x86/include/asm/cpu_common.h
index 48f56c2..2a5779a 100644
--- a/arch/x86/include/asm/cpu_common.h
+++ b/arch/x86/include/asm/cpu_common.h
@@ -184,4 +184,13 @@ int cpu_get_max_turbo_ratio(void);
  */
 int cpu_get_cores_per_package(void);
 
+/**
+ * cpu_mca_configure() - Set up machine-check exceptions ready for use
+ *
+ * These allow the SoC to report errors while running. See here for details:
+ *
+ * https://www.intel.com/content/dam/www/public/us/en/documents/white-papers/machine-check-exceptions-debug-paper.pdf
+ */
+void cpu_mca_configure(void);
+
 #endif
diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index 94e6b18..c49b422 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -68,7 +68,18 @@
 #define MSR_BSEL_CR_OVERCLOCK_CONTROL	0x000000cd
 #define MSR_PLATFORM_INFO		0x000000ce
 #define MSR_PMG_CST_CONFIG_CONTROL	0x000000e2
-#define SINGLE_PCTL			(1 << 11)
+/* Set MSR_PMG_CST_CONFIG_CONTROL[3:0] for Package C-State limit */
+#define   PKG_C_STATE_LIMIT_C2_MASK	BIT(1)
+/* Set MSR_PMG_CST_CONFIG_CONTROL[7:4] for Core C-State limit*/
+#define   CORE_C_STATE_LIMIT_C10_MASK	0x70
+/* Set MSR_PMG_CST_CONFIG_CONTROL[10] to IO redirect to MWAIT */
+#define   IO_MWAIT_REDIRECT_MASK	BIT(10)
+/* Set MSR_PMG_CST_CONFIG_CONTROL[15] to lock CST_CFG [0-15] bits */
+#define   CST_CFG_LOCK_MASK		BIT(15)
+#define   SINGLE_PCTL			BIT(11)
+
+/* ACPI PMIO Offset to C-state register */
+#define ACPI_PMIO_CST_REG	(ACPI_BASE_ADDRESS + 0x14)
 
 #define MSR_MTRRcap			0x000000fe
 #define MSR_IA32_BBL_CR_CTL		0x00000119
@@ -83,6 +94,10 @@
 #define  EMULATE_PM_TMR_EN		(1 << 16)
 #define  EMULATE_DELAY_VALUE		0x13
 
+#define MSR_FEATURE_CONFIG	0x13c
+#define   FEATURE_CONFIG_RESERVED_MASK	0x3ULL
+#define   FEATURE_CONFIG_LOCK	(1 << 0)
+
 #define MSR_IA32_SYSENTER_CS		0x00000174
 #define MSR_IA32_SYSENTER_ESP		0x00000175
 #define MSR_IA32_SYSENTER_EIP		0x00000176
@@ -453,6 +468,9 @@
 #define MSR_AMD_PERF_CTL		0xc0010062
 
 #define MSR_PMG_CST_CONFIG_CTL		0x000000e2
+/* CST Range (R/W) IO port block size */
+#define PMG_IO_BASE_CST_RNG_BLK_SIZE	0x5
+
 #define MSR_PMG_IO_CAPTURE_ADR		0x000000e4
 #define MSR_IA32_MPERF			0x000000e7
 #define MSR_IA32_APERF			0x000000e8