diff options
author | Angel Pons <th3fanbus@gmail.com> | 2020-10-26 00:27:09 +0100 |
---|---|---|
committer | Michael Niewöhner <foss@mniewoehner.de> | 2020-11-13 13:17:49 +0000 |
commit | 3bd017356a7766c4884e55a28ca481c8a9110ceb (patch) | |
tree | 3d9ebcf9af8d5cbf5c3c63b25ed3bc2800baf573 /src/soc/intel/broadwell/cpu | |
parent | a0426267e330c60aa74528a06a0e130efe2ad32f (diff) |
soc/intel/broadwell: Relocate CPU files
Change-Id: Ib2ddce78db21db9c8deac632a77ecd71eb9887c2
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46794
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Diffstat (limited to 'src/soc/intel/broadwell/cpu')
-rw-r--r-- | src/soc/intel/broadwell/cpu/Makefile.inc | 29 | ||||
-rw-r--r-- | src/soc/intel/broadwell/cpu/bootblock.c | 65 | ||||
-rw-r--r-- | src/soc/intel/broadwell/cpu/cpu.c | 642 | ||||
-rw-r--r-- | src/soc/intel/broadwell/cpu/romstage.c | 32 | ||||
-rw-r--r-- | src/soc/intel/broadwell/cpu/smmrelocate.c | 264 | ||||
-rw-r--r-- | src/soc/intel/broadwell/cpu/tsc_freq.c | 14 |
6 files changed, 1046 insertions, 0 deletions
diff --git a/src/soc/intel/broadwell/cpu/Makefile.inc b/src/soc/intel/broadwell/cpu/Makefile.inc new file mode 100644 index 0000000000..e9f46376af --- /dev/null +++ b/src/soc/intel/broadwell/cpu/Makefile.inc @@ -0,0 +1,29 @@ +subdirs-y += ../../../../cpu/x86/lapic +subdirs-y += ../../../../cpu/x86/mtrr +subdirs-y += ../../../../cpu/x86/smm +subdirs-y += ../../../../cpu/x86/tsc +subdirs-y += ../../../../cpu/intel/microcode +subdirs-y += ../../../../cpu/intel/turbo +subdirs-y += ../../../../cpu/intel/common + +bootblock-y += bootblock.c +bootblock-y += ../../../../cpu/intel/car/bootblock.c +bootblock-y += ../../../../cpu/intel/car/non-evict/cache_as_ram.S +bootblock-y += ../../../../cpu/x86/early_reset.S + +romstage-y += romstage.c +romstage-y += ../../../../cpu/intel/car/romstage.c + +postcar-y += ../../../../cpu/intel/car/non-evict/exit_car.S + +ramstage-y += cpu.c +ramstage-y += smmrelocate.c + +bootblock-y += tsc_freq.c +ramstage-y += tsc_freq.c +romstage-y += tsc_freq.c +smm-y += tsc_freq.c +postcar-y += tsc_freq.c +verstage-y += tsc_freq.c + +cpu_microcode_bins += 3rdparty/blobs/soc/intel/broadwell/microcode.bin diff --git a/src/soc/intel/broadwell/cpu/bootblock.c b/src/soc/intel/broadwell/cpu/bootblock.c new file mode 100644 index 0000000000..d6883c6436 --- /dev/null +++ b/src/soc/intel/broadwell/cpu/bootblock.c @@ -0,0 +1,65 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ + +#include <stdint.h> +#include <arch/bootblock.h> +#include <arch/io.h> +#include <cpu/x86/msr.h> +#include <halt.h> +#include <soc/rcba.h> +#include <soc/msr.h> +#include <delay.h> + +static void set_flex_ratio_to_tdp_nominal(void) +{ + msr_t flex_ratio, msr; + u32 soft_reset; + u8 nominal_ratio; + + /* Check for Flex Ratio support */ + flex_ratio = rdmsr(MSR_FLEX_RATIO); + if (!(flex_ratio.lo & FLEX_RATIO_EN)) + return; + + /* Check for >0 configurable TDPs */ + msr = rdmsr(MSR_PLATFORM_INFO); + if (((msr.hi >> 1) & 3) == 0) + return; + + /* Use nominal TDP ratio for flex ratio */ + msr = rdmsr(MSR_CONFIG_TDP_NOMINAL); + nominal_ratio = msr.lo & 0xff; + + /* See if flex ratio is already set to nominal TDP ratio */ + if (((flex_ratio.lo >> 8) & 0xff) == nominal_ratio) + return; + + /* Set flex ratio to nominal TDP ratio */ + flex_ratio.lo &= ~0xff00; + flex_ratio.lo |= nominal_ratio << 8; + flex_ratio.lo |= FLEX_RATIO_LOCK; + wrmsr(MSR_FLEX_RATIO, flex_ratio); + + /* Set flex ratio in soft reset data register bits 11:6. + * RCBA region is enabled in southbridge bootblock */ + soft_reset = RCBA32(SOFT_RESET_DATA); + soft_reset &= ~(0x3f << 6); + soft_reset |= (nominal_ratio & 0x3f) << 6; + RCBA32(SOFT_RESET_DATA) = soft_reset; + + /* Set soft reset control to use register value */ + RCBA32_OR(SOFT_RESET_CTRL, 1); + + /* Delay before reset to avoid potential TPM lockout */ + mdelay(30); + + /* Issue warm reset, will be "CPU only" due to soft reset data */ + outb(0x0, 0xcf9); + outb(0x6, 0xcf9); + halt(); +} + +void bootblock_early_cpu_init(void) +{ + /* Set flex ratio and reset if needed */ + set_flex_ratio_to_tdp_nominal(); +} diff --git a/src/soc/intel/broadwell/cpu/cpu.c b/src/soc/intel/broadwell/cpu/cpu.c new file mode 100644 index 0000000000..72efa3dc81 --- /dev/null +++ b/src/soc/intel/broadwell/cpu/cpu.c @@ -0,0 +1,642 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ + +#include <console/console.h> +#include <device/device.h> +#include <device/pci.h> +#include <arch/cpu.h> +#include <cpu/cpu.h> +#include <cpu/x86/mtrr.h> +#include <cpu/x86/msr.h> +#include <cpu/x86/lapic.h> +#include <cpu/x86/mp.h> +#include <cpu/intel/microcode.h> +#include <cpu/intel/smm_reloc.h> +#include <cpu/intel/speedstep.h> +#include <cpu/intel/turbo.h> +#include <cpu/x86/name.h> +#include <delay.h> +#include <soc/cpu.h> +#include <soc/msr.h> +#include <soc/pci_devs.h> +#include <soc/ramstage.h> +#include <soc/rcba.h> +#include <soc/systemagent.h> +#include <soc/intel/broadwell/chip.h> +#include <cpu/intel/common/common.h> + +/* Convert time in seconds to POWER_LIMIT_1_TIME MSR value */ +static const u8 power_limit_time_sec_to_msr[] = { + [0] = 0x00, + [1] = 0x0a, + [2] = 0x0b, + [3] = 0x4b, + [4] = 0x0c, + [5] = 0x2c, + [6] = 0x4c, + [7] = 0x6c, + [8] = 0x0d, + [10] = 0x2d, + [12] = 0x4d, + [14] = 0x6d, + [16] = 0x0e, + [20] = 0x2e, + [24] = 0x4e, + [28] = 0x6e, + [32] = 0x0f, + [40] = 0x2f, + [48] = 0x4f, + [56] = 0x6f, + [64] = 0x10, + [80] = 0x30, + [96] = 0x50, + [112] = 0x70, + [128] = 0x11, +}; + +/* Convert POWER_LIMIT_1_TIME MSR value to seconds */ +static const u8 power_limit_time_msr_to_sec[] = { + [0x00] = 0, + [0x0a] = 1, + [0x0b] = 2, + [0x4b] = 3, + [0x0c] = 4, + [0x2c] = 5, + [0x4c] = 6, + [0x6c] = 7, + [0x0d] = 8, + [0x2d] = 10, + [0x4d] = 12, + [0x6d] = 14, + [0x0e] = 16, + [0x2e] = 20, + [0x4e] = 24, + [0x6e] = 28, + [0x0f] = 32, + [0x2f] = 40, + [0x4f] = 48, + [0x6f] = 56, + [0x10] = 64, + [0x30] = 80, + [0x50] = 96, + [0x70] = 112, + [0x11] = 128, +}; + +/* The core 100MHz BCLK is disabled in deeper c-states. One needs to calibrate + * the 100MHz BCLK against the 24MHz BCLK to restore the clocks properly + * when a core is woken up. */ +static int pcode_ready(void) +{ + int wait_count; + const int delay_step = 10; + + wait_count = 0; + do { + if (!(MCHBAR32(BIOS_MAILBOX_INTERFACE) & MAILBOX_RUN_BUSY)) + return 0; + wait_count += delay_step; + udelay(delay_step); + } while (wait_count < 1000); + + return -1; +} + +static void calibrate_24mhz_bclk(void) +{ + int err_code; + + if (pcode_ready() < 0) { + printk(BIOS_ERR, "PCODE: mailbox timeout on wait ready.\n"); + return; + } + + /* A non-zero value initiates the PCODE calibration. */ + MCHBAR32(BIOS_MAILBOX_DATA) = ~0; + MCHBAR32(BIOS_MAILBOX_INTERFACE) = + MAILBOX_RUN_BUSY | MAILBOX_BIOS_CMD_FSM_MEASURE_INTVL; + + if (pcode_ready() < 0) { + printk(BIOS_ERR, "PCODE: mailbox timeout on completion.\n"); + return; + } + + err_code = MCHBAR32(BIOS_MAILBOX_INTERFACE) & 0xff; + + printk(BIOS_DEBUG, "PCODE: 24MHz BCLK calibration response: %d\n", + err_code); + + /* Read the calibrated value. */ + MCHBAR32(BIOS_MAILBOX_INTERFACE) = + MAILBOX_RUN_BUSY | MAILBOX_BIOS_CMD_READ_CALIBRATION; + + if (pcode_ready() < 0) { + printk(BIOS_ERR, "PCODE: mailbox timeout on read.\n"); + return; + } + + printk(BIOS_DEBUG, "PCODE: 24MHz BCLK calibration value: 0x%08x\n", + MCHBAR32(BIOS_MAILBOX_DATA)); +} + +static u32 pcode_mailbox_read(u32 command) +{ + if (pcode_ready() < 0) { + printk(BIOS_ERR, "PCODE: mailbox timeout on wait ready.\n"); + return 0; + } + + /* Send command and start transaction */ + MCHBAR32(BIOS_MAILBOX_INTERFACE) = command | MAILBOX_RUN_BUSY; + + if (pcode_ready() < 0) { + printk(BIOS_ERR, "PCODE: mailbox timeout on completion.\n"); + return 0; + } + + /* Read mailbox */ + return MCHBAR32(BIOS_MAILBOX_DATA); +} + +static int pcode_mailbox_write(u32 command, u32 data) +{ + if (pcode_ready() < 0) { + printk(BIOS_ERR, "PCODE: mailbox timeout on wait ready.\n"); + return -1; + } + + MCHBAR32(BIOS_MAILBOX_DATA) = data; + + /* Send command and start transaction */ + MCHBAR32(BIOS_MAILBOX_INTERFACE) = command | MAILBOX_RUN_BUSY; + + if (pcode_ready() < 0) { + printk(BIOS_ERR, "PCODE: mailbox timeout on completion.\n"); + return -1; + } + + return 0; +} + +static void initialize_vr_config(void) +{ + config_t *conf = config_of_soc(); + msr_t msr; + + printk(BIOS_DEBUG, "Initializing VR config.\n"); + + /* Configure VR_CURRENT_CONFIG. */ + msr = rdmsr(MSR_VR_CURRENT_CONFIG); + /* Preserve bits 63 and 62. Bit 62 is PSI4 enable, but it is only valid + * on ULT systems. */ + msr.hi &= 0xc0000000; + msr.hi |= (0x01 << (52 - 32)); /* PSI3 threshold - 1A. */ + msr.hi |= (0x05 << (42 - 32)); /* PSI2 threshold - 5A. */ + msr.hi |= (0x14 << (32 - 32)); /* PSI1 threshold - 20A. */ + msr.hi |= (1 << (62 - 32)); /* Enable PSI4 */ + /* Leave the max instantaneous current limit (12:0) to default. */ + wrmsr(MSR_VR_CURRENT_CONFIG, msr); + + /* Configure VR_MISC_CONFIG MSR. */ + msr = rdmsr(MSR_VR_MISC_CONFIG); + /* Set the IOUT_SLOPE scalar applied to dIout in U10.1.9 format. */ + msr.hi &= ~(0x3ff << (40 - 32)); + msr.hi |= (0x200 << (40 - 32)); /* 1.0 */ + /* Set IOUT_OFFSET to 0. */ + msr.hi &= ~0xff; + /* Set entry ramp rate to slow. */ + msr.hi &= ~(1 << (51 - 32)); + /* Enable decay mode on C-state entry. */ + msr.hi |= (1 << (52 - 32)); + /* Set the slow ramp rate */ + msr.hi &= ~(0x3 << (53 - 32)); + /* Configure the C-state exit ramp rate. */ + if (conf->vr_slow_ramp_rate_enable) { + /* Configured slow ramp rate. */ + msr.hi |= ((conf->vr_slow_ramp_rate_set & 0x3) << (53 - 32)); + /* Set exit ramp rate to slow. */ + msr.hi &= ~(1 << (50 - 32)); + } else { + /* Fast ramp rate / 4. */ + msr.hi |= (0x01 << (53 - 32)); + /* Set exit ramp rate to fast. */ + msr.hi |= (1 << (50 - 32)); + } + /* Set MIN_VID (31:24) to allow CPU to have full control. */ + msr.lo &= ~0xff000000; + msr.lo |= (conf->vr_cpu_min_vid & 0xff) << 24; + wrmsr(MSR_VR_MISC_CONFIG, msr); + + /* Configure VR_MISC_CONFIG2 MSR. */ + msr = rdmsr(MSR_VR_MISC_CONFIG2); + msr.lo &= ~0xffff; + /* Allow CPU to control minimum voltage completely (15:8) and + * set the fast ramp voltage in 10mV steps. */ + if (cpu_family_model() == BROADWELL_FAMILY_ULT) + msr.lo |= 0x006a; /* 1.56V */ + else + msr.lo |= 0x006f; /* 1.60V */ + wrmsr(MSR_VR_MISC_CONFIG2, msr); + + /* Set C9/C10 VCC Min */ + pcode_mailbox_write(MAILBOX_BIOS_CMD_WRITE_C9C10_VOLTAGE, 0x1f1f); +} + +static void configure_pch_power_sharing(void) +{ + u32 pch_power, pch_power_ext, pmsync, pmsync2; + int i; + + /* Read PCH Power levels from PCODE */ + pch_power = pcode_mailbox_read(MAILBOX_BIOS_CMD_READ_PCH_POWER); + pch_power_ext = pcode_mailbox_read(MAILBOX_BIOS_CMD_READ_PCH_POWER_EXT); + + printk(BIOS_INFO, "PCH Power: PCODE Levels 0x%08x 0x%08x\n", + pch_power, pch_power_ext); + + pmsync = RCBA32(PMSYNC_CONFIG); + pmsync2 = RCBA32(PMSYNC_CONFIG2); + + /* Program PMSYNC_TPR_CONFIG PCH power limit values + * pmsync[0:4] = mailbox[0:5] + * pmsync[8:12] = mailbox[6:11] + * pmsync[16:20] = mailbox[12:17] + */ + for (i = 0; i < 3; i++) { + u32 level = pch_power & 0x3f; + pch_power >>= 6; + pmsync &= ~(0x1f << (i * 8)); + pmsync |= (level & 0x1f) << (i * 8); + } + RCBA32(PMSYNC_CONFIG) = pmsync; + + /* Program PMSYNC_TPR_CONFIG2 Extended PCH power limit values + * pmsync2[0:4] = mailbox[23:18] + * pmsync2[8:12] = mailbox_ext[6:11] + * pmsync2[16:20] = mailbox_ext[12:17] + * pmsync2[24:28] = mailbox_ext[18:22] + */ + pmsync2 &= ~0x1f; + pmsync2 |= pch_power & 0x1f; + + for (i = 1; i < 4; i++) { + u32 level = pch_power_ext & 0x3f; + pch_power_ext >>= 6; + pmsync2 &= ~(0x1f << (i * 8)); + pmsync2 |= (level & 0x1f) << (i * 8); + } + RCBA32(PMSYNC_CONFIG2) = pmsync2; +} + +int cpu_config_tdp_levels(void) +{ + msr_t platform_info; + + /* Bits 34:33 indicate how many levels supported */ + platform_info = rdmsr(MSR_PLATFORM_INFO); + return (platform_info.hi >> 1) & 3; +} + +/* + * Configure processor power limits if possible + * This must be done AFTER set of BIOS_RESET_CPL + */ +void set_power_limits(u8 power_limit_1_time) +{ + msr_t msr = rdmsr(MSR_PLATFORM_INFO); + msr_t limit; + unsigned int power_unit; + unsigned int tdp, min_power, max_power, max_time; + u8 power_limit_1_val; + + if (power_limit_1_time >= ARRAY_SIZE(power_limit_time_sec_to_msr)) + power_limit_1_time = ARRAY_SIZE(power_limit_time_sec_to_msr) - 1; + + if (!(msr.lo & PLATFORM_INFO_SET_TDP)) + return; + + /* Get units */ + msr = rdmsr(MSR_PKG_POWER_SKU_UNIT); + power_unit = 2 << ((msr.lo & 0xf) - 1); + + /* Get power defaults for this SKU */ + msr = rdmsr(MSR_PKG_POWER_SKU); + tdp = msr.lo & 0x7fff; + min_power = (msr.lo >> 16) & 0x7fff; + max_power = msr.hi & 0x7fff; + max_time = (msr.hi >> 16) & 0x7f; + + printk(BIOS_DEBUG, "CPU TDP: %u Watts\n", tdp / power_unit); + + if (power_limit_time_msr_to_sec[max_time] > power_limit_1_time) + power_limit_1_time = power_limit_time_msr_to_sec[max_time]; + + if (min_power > 0 && tdp < min_power) + tdp = min_power; + + if (max_power > 0 && tdp > max_power) + tdp = max_power; + + power_limit_1_val = power_limit_time_sec_to_msr[power_limit_1_time]; + + /* Set long term power limit to TDP */ + limit.lo = 0; + limit.lo |= tdp & PKG_POWER_LIMIT_MASK; + limit.lo |= PKG_POWER_LIMIT_EN; + limit.lo |= (power_limit_1_val & PKG_POWER_LIMIT_TIME_MASK) << + PKG_POWER_LIMIT_TIME_SHIFT; + + /* Set short term power limit to 1.25 * TDP */ + limit.hi = 0; + limit.hi |= ((tdp * 125) / 100) & PKG_POWER_LIMIT_MASK; + limit.hi |= PKG_POWER_LIMIT_EN; + /* Power limit 2 time is only programmable on server SKU */ + + wrmsr(MSR_PKG_POWER_LIMIT, limit); + + /* Set power limit values in MCHBAR as well */ + MCHBAR32(MCH_PKG_POWER_LIMIT_LO) = limit.lo; + MCHBAR32(MCH_PKG_POWER_LIMIT_HI) = limit.hi; + + /* Set DDR RAPL power limit by copying from MMIO to MSR */ + msr.lo = MCHBAR32(MCH_DDR_POWER_LIMIT_LO); + msr.hi = MCHBAR32(MCH_DDR_POWER_LIMIT_HI); + wrmsr(MSR_DDR_RAPL_LIMIT, msr); + + /* Use nominal TDP values for CPUs with configurable TDP */ + if (cpu_config_tdp_levels()) { + msr = rdmsr(MSR_CONFIG_TDP_NOMINAL); + limit.hi = 0; + limit.lo = msr.lo & 0xff; + wrmsr(MSR_TURBO_ACTIVATION_RATIO, limit); + } +} + +static void configure_c_states(void) +{ + msr_t msr; + + msr = rdmsr(MSR_PKG_CST_CONFIG_CONTROL); + msr.lo |= (1 << 31); // Timed MWAIT Enable + msr.lo |= (1 << 30); // Package c-state Undemotion Enable + msr.lo |= (1 << 29); // Package c-state Demotion Enable + msr.lo |= (1 << 28); // C1 Auto Undemotion Enable + msr.lo |= (1 << 27); // C3 Auto Undemotion Enable + msr.lo |= (1 << 26); // C1 Auto Demotion Enable + msr.lo |= (1 << 25); // C3 Auto Demotion Enable + msr.lo &= ~(1 << 10); // Disable IO MWAIT redirection + /* The deepest package c-state defaults to factory-configured value. */ + wrmsr(MSR_PKG_CST_CONFIG_CONTROL, msr); + + msr = rdmsr(MSR_MISC_PWR_MGMT); + msr.lo &= ~(1 << 0); // Enable P-state HW_ALL coordination + wrmsr(MSR_MISC_PWR_MGMT, msr); + + msr = rdmsr(MSR_POWER_CTL); + msr.lo |= (1 << 18); // Enable Energy Perf Bias MSR 0x1b0 + msr.lo |= (1 << 1); // C1E Enable + msr.lo |= (1 << 0); // Bi-directional PROCHOT# + wrmsr(MSR_POWER_CTL, msr); + + /* C-state Interrupt Response Latency Control 0 - package C3 latency */ + msr.hi = 0; + msr.lo = IRTL_VALID | IRTL_1024_NS | C_STATE_LATENCY_CONTROL_0_LIMIT; + wrmsr(MSR_C_STATE_LATENCY_CONTROL_0, msr); + + /* C-state Interrupt Response Latency Control 1 */ + msr.hi = 0; + msr.lo = IRTL_VALID | IRTL_1024_NS | C_STATE_LATENCY_CONTROL_1_LIMIT; + wrmsr(MSR_C_STATE_LATENCY_CONTROL_1, msr); + + /* C-state Interrupt Response Latency Control 2 - package C6/C7 short */ + msr.hi = 0; + msr.lo = IRTL_VALID | IRTL_1024_NS | C_STATE_LATENCY_CONTROL_2_LIMIT; + wrmsr(MSR_C_STATE_LATENCY_CONTROL_2, msr); + + /* C-state Interrupt Response Latency Control 3 - package C8 */ + msr.hi = 0; + msr.lo = IRTL_VALID | IRTL_1024_NS | + C_STATE_LATENCY_CONTROL_3_LIMIT; + wrmsr(MSR_C_STATE_LATENCY_CONTROL_3, msr); + + /* C-state Interrupt Response Latency Control 4 - package C9 */ + msr.hi = 0; + msr.lo = IRTL_VALID | IRTL_1024_NS | + C_STATE_LATENCY_CONTROL_4_LIMIT; + wrmsr(MSR_C_STATE_LATENCY_CONTROL_4, msr); + + /* C-state Interrupt Response Latency Control 5 - package C10 */ + msr.hi = 0; + msr.lo = IRTL_VALID | IRTL_1024_NS | + C_STATE_LATENCY_CONTROL_5_LIMIT; + wrmsr(MSR_C_STATE_LATENCY_CONTROL_5, msr); +} + +static void configure_thermal_target(void) +{ + config_t *conf; + struct device *lapic; + msr_t msr; + + /* Find pointer to CPU configuration */ + lapic = dev_find_lapic(SPEEDSTEP_APIC_MAGIC); + if (!lapic || !lapic->chip_info) + return; + conf = lapic->chip_info; + + /* Set TCC activation offset if supported */ + msr = rdmsr(MSR_PLATFORM_INFO); + if ((msr.lo & (1 << 30)) && conf->tcc_offset) { + msr = rdmsr(MSR_TEMPERATURE_TARGET); + msr.lo &= ~(0xf << 24); /* Bits 27:24 */ + msr.lo |= (conf->tcc_offset & 0xf) << 24; + wrmsr(MSR_TEMPERATURE_TARGET, msr); + } +} + +static void configure_misc(void) +{ + msr_t msr; + + msr = rdmsr(IA32_MISC_ENABLE); + msr.lo |= (1 << 0); /* Fast String enable */ + msr.lo |= (1 << 3); /* TM1/TM2/EMTTM enable */ + msr.lo |= (1 << 16); /* Enhanced SpeedStep Enable */ + wrmsr(IA32_MISC_ENABLE, msr); + + /* Disable Thermal interrupts */ + msr.lo = 0; + msr.hi = 0; + wrmsr(IA32_THERM_INTERRUPT, msr); + + /* Enable package critical interrupt only */ + msr.lo = 1 << 4; + msr.hi = 0; + wrmsr(IA32_PACKAGE_THERM_INTERRUPT, msr); +} + +static void set_max_ratio(void) +{ + msr_t msr, perf_ctl; + + perf_ctl.hi = 0; + + /* Check for configurable TDP option */ + if (get_turbo_state() == TURBO_ENABLED) { + msr = rdmsr(MSR_TURBO_RATIO_LIMIT); + perf_ctl.lo = (msr.lo & 0xff) << 8; + } else if (cpu_config_tdp_levels()) { + /* Set to nominal TDP ratio */ + msr = rdmsr(MSR_CONFIG_TDP_NOMINAL); + perf_ctl.lo = (msr.lo & 0xff) << 8; + } else { + /* Platform Info bits 15:8 give max ratio */ + msr = rdmsr(MSR_PLATFORM_INFO); + perf_ctl.lo = msr.lo & 0xff00; + } + wrmsr(IA32_PERF_CTL, perf_ctl); + + printk(BIOS_DEBUG, "CPU: frequency set to %d\n", + ((perf_ctl.lo >> 8) & 0xff) * CPU_BCLK); +} + +static void configure_mca(void) +{ + msr_t msr; + int i; + int num_banks; + + msr = rdmsr(IA32_MCG_CAP); + num_banks = msr.lo & 0xff; + msr.lo = msr.hi = 0; + /* TODO(adurbin): This should only be done on a cold boot. Also, some + * of these banks are core vs package scope. For now every CPU clears + * every bank. */ + for (i = 0; i < num_banks; i++) + wrmsr(IA32_MC0_STATUS + (i * 4), msr); +} + +/* All CPUs including BSP will run the following function. */ +static void cpu_core_init(struct device *cpu) +{ + /* Clear out pending MCEs */ + configure_mca(); + + /* Enable the local CPU APICs */ + enable_lapic_tpr(); + setup_lapic(); + + /* Set virtualization based on Kconfig option */ + set_vmx_and_lock(); + + /* Configure C States */ + configure_c_states(); + + /* Configure Enhanced SpeedStep and Thermal Sensors */ + configure_misc(); + + /* Thermal throttle activation offset */ + configure_thermal_target(); + + /* Enable Direct Cache Access */ + configure_dca_cap(); + + /* Set energy policy */ + set_energy_perf_bias(ENERGY_POLICY_NORMAL); + + /* Enable Turbo */ + enable_turbo(); +} + +/* MP initialization support. */ +static const void *microcode_patch; + +static void pre_mp_init(void) +{ + /* Setup MTRRs based on physical address size. */ + x86_setup_mtrrs_with_detect(); + x86_mtrr_check(); + + initialize_vr_config(); + calibrate_24mhz_bclk(); + configure_pch_power_sharing(); +} + +static int get_cpu_count(void) +{ + msr_t msr; + int num_threads; + int num_cores; + + msr = rdmsr(MSR_CORE_THREAD_COUNT); + num_threads = (msr.lo >> 0) & 0xffff; + num_cores = (msr.lo >> 16) & 0xffff; + printk(BIOS_DEBUG, "CPU has %u cores, %u threads enabled.\n", + num_cores, num_threads); + + return num_threads; +} + +static void get_microcode_info(const void **microcode, int *parallel) +{ + microcode_patch = intel_microcode_find(); + *microcode = microcode_patch; + *parallel = 1; +} + +static void per_cpu_smm_trigger(void) +{ + /* Relocate the SMM handler. */ + smm_relocate(); + + /* After SMM relocation a 2nd microcode load is required. */ + intel_microcode_load_unlocked(microcode_patch); +} + +static void post_mp_init(void) +{ + /* Set Max Ratio */ + set_max_ratio(); + + /* Now that all APs have been relocated as well as the BSP let SMIs + * start flowing. */ + global_smi_enable(); + + /* Lock down the SMRAM space. */ + smm_lock(); +} + +static const struct mp_ops mp_ops = { + .pre_mp_init = pre_mp_init, + .get_cpu_count = get_cpu_count, + .get_smm_info = smm_info, + .get_microcode_info = get_microcode_info, + .pre_mp_smm_init = smm_initialize, + .per_cpu_smm_trigger = per_cpu_smm_trigger, + .relocation_handler = smm_relocation_handler, + .post_mp_init = post_mp_init, +}; + +void broadwell_init_cpus(struct device *dev) +{ + struct bus *cpu_bus = dev->link_list; + + if (mp_init_with_smm(cpu_bus, &mp_ops)) + printk(BIOS_ERR, "MP initialization failure.\n"); +} + +static struct device_operations cpu_dev_ops = { + .init = cpu_core_init, +}; + +static const struct cpu_device_id cpu_table[] = { + { X86_VENDOR_INTEL, CPUID_HASWELL_ULT }, + { X86_VENDOR_INTEL, CPUID_BROADWELL_C0 }, + { X86_VENDOR_INTEL, CPUID_BROADWELL_D0 }, + { X86_VENDOR_INTEL, CPUID_BROADWELL_E0 }, + { 0, 0 }, +}; + +static const struct cpu_driver driver __cpu_driver = { + .ops = &cpu_dev_ops, + .id_table = cpu_table, +}; diff --git a/src/soc/intel/broadwell/cpu/romstage.c b/src/soc/intel/broadwell/cpu/romstage.c new file mode 100644 index 0000000000..c9f70a85d1 --- /dev/null +++ b/src/soc/intel/broadwell/cpu/romstage.c @@ -0,0 +1,32 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ + +#include <arch/cpu.h> +#include <console/console.h> +#include <cpu/x86/msr.h> +#include <soc/cpu.h> +#include <soc/msr.h> +#include <soc/romstage.h> + +void set_max_freq(void) +{ + msr_t msr, perf_ctl, platform_info; + + /* Check for configurable TDP option */ + platform_info = rdmsr(MSR_PLATFORM_INFO); + + if ((platform_info.hi >> 1) & 3) { + /* Set to nominal TDP ratio */ + msr = rdmsr(MSR_CONFIG_TDP_NOMINAL); + perf_ctl.lo = (msr.lo & 0xff) << 8; + } else { + /* Platform Info bits 15:8 give max ratio */ + msr = rdmsr(MSR_PLATFORM_INFO); + perf_ctl.lo = msr.lo & 0xff00; + } + + perf_ctl.hi = 0; + wrmsr(IA32_PERF_CTL, perf_ctl); + + printk(BIOS_DEBUG, "CPU: frequency set to %d MHz\n", + ((perf_ctl.lo >> 8) & 0xff) * CPU_BCLK); +} diff --git a/src/soc/intel/broadwell/cpu/smmrelocate.c b/src/soc/intel/broadwell/cpu/smmrelocate.c new file mode 100644 index 0000000000..3d4162abd0 --- /dev/null +++ b/src/soc/intel/broadwell/cpu/smmrelocate.c @@ -0,0 +1,264 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ + +#include <types.h> +#include <string.h> +#include <device/device.h> +#include <device/pci.h> +#include <device/pci_ops.h> +#include <cpu/x86/mp.h> +#include <cpu/x86/msr.h> +#include <cpu/x86/mtrr.h> +#include <cpu/x86/smm.h> +#include <cpu/intel/em64t101_save_state.h> +#include <cpu/intel/smm_reloc.h> +#include <console/console.h> +#include <smp/node.h> +#include <soc/cpu.h> +#include <soc/msr.h> +#include <soc/pci_devs.h> +#include <soc/systemagent.h> + +static void update_save_state(int cpu, uintptr_t curr_smbase, + uintptr_t staggered_smbase, + struct smm_relocation_params *relo_params) +{ + u32 smbase; + u32 iedbase; + + /* The relocated handler runs with all CPUs concurrently. Therefore + * stagger the entry points adjusting SMBASE downwards by save state + * size * CPU num. */ + smbase = staggered_smbase; + iedbase = relo_params->ied_base; + + printk(BIOS_DEBUG, "New SMBASE=0x%08x IEDBASE=0x%08x\n", + smbase, iedbase); + + /* All threads need to set IEDBASE and SMBASE to the relocated + * handler region. However, the save state location depends on the + * smm_save_state_in_msrs field in the relocation parameters. If + * smm_save_state_in_msrs is non-zero then the CPUs are relocating + * the SMM handler in parallel, and each CPUs save state area is + * located in their respective MSR space. If smm_save_state_in_msrs + * is zero then the SMM relocation is happening serially so the + * save state is at the same default location for all CPUs. */ + if (relo_params->smm_save_state_in_msrs) { + msr_t smbase_msr; + msr_t iedbase_msr; + + smbase_msr.lo = smbase; + smbase_msr.hi = 0; + + /* According the BWG the IEDBASE MSR is in bits 63:32. It's + * not clear why it differs from the SMBASE MSR. */ + iedbase_msr.lo = 0; + iedbase_msr.hi = iedbase; + + wrmsr(SMBASE_MSR, smbase_msr); + wrmsr(IEDBASE_MSR, iedbase_msr); + } else { + em64t101_smm_state_save_area_t *save_state; + + save_state = (void *)(curr_smbase + SMM_DEFAULT_SIZE - + sizeof(*save_state)); + + save_state->smbase = smbase; + save_state->iedbase = iedbase; + } +} + +/* Returns 1 if SMM MSR save state was set. */ +static int bsp_setup_msr_save_state(struct smm_relocation_params *relo_params) +{ + msr_t smm_mca_cap; + + smm_mca_cap = rdmsr(SMM_MCA_CAP_MSR); + if (smm_mca_cap.hi & SMM_CPU_SVRSTR_MASK) { + msr_t smm_feature_control; + + smm_feature_control = rdmsr(SMM_FEATURE_CONTROL_MSR); + smm_feature_control.hi = 0; + smm_feature_control.lo |= SMM_CPU_SAVE_EN; + wrmsr(SMM_FEATURE_CONTROL_MSR, smm_feature_control); + relo_params->smm_save_state_in_msrs = 1; + } + return relo_params->smm_save_state_in_msrs; +} + +/* The relocation work is actually performed in SMM context, but the code + * resides in the ramstage module. This occurs by trampolining from the default + * SMRAM entry point to here. */ +void smm_relocation_handler(int cpu, uintptr_t curr_smbase, + uintptr_t staggered_smbase) +{ + msr_t mtrr_cap; + struct smm_relocation_params *relo_params = &smm_reloc_params; + + printk(BIOS_DEBUG, "In relocation handler: CPU %d\n", cpu); + + /* Determine if the processor supports saving state in MSRs. If so, + * enable it before the non-BSPs run so that SMM relocation can occur + * in parallel in the non-BSP CPUs. */ + if (cpu == 0) { + /* If smm_save_state_in_msrs is 1 then that means this is the + * 2nd time through the relocation handler for the BSP. + * Parallel SMM handler relocation is taking place. However, + * it is desired to access other CPUs save state in the real + * SMM handler. Therefore, disable the SMM save state in MSRs + * feature. */ + if (relo_params->smm_save_state_in_msrs) { + msr_t smm_feature_control; + + smm_feature_control = rdmsr(SMM_FEATURE_CONTROL_MSR); + smm_feature_control.lo &= ~SMM_CPU_SAVE_EN; + wrmsr(SMM_FEATURE_CONTROL_MSR, smm_feature_control); + } else if (bsp_setup_msr_save_state(relo_params)) + /* Just return from relocation handler if MSR save + * state is enabled. In that case the BSP will come + * back into the relocation handler to setup the new + * SMBASE as well disabling SMM save state in MSRs. */ + return; + } + + /* Make appropriate changes to the save state map. */ + update_save_state(cpu, curr_smbase, staggered_smbase, relo_params); + + /* Write PRMRR and SMRR MSRs based on indicated support. */ + mtrr_cap = rdmsr(MTRR_CAP_MSR); + if (mtrr_cap.lo & SMRR_SUPPORTED) + write_smrr(relo_params); + + if (mtrr_cap.lo & PRMRR_SUPPORTED) { + write_prmrr(relo_params); + /* UNCORE_PRMRR msrs are package level. Therefore, only + * configure these MSRs on the BSP. */ + if (cpu == 0) + write_uncore_prmrr(relo_params); + } +} + +static void fill_in_relocation_params(struct smm_relocation_params *params) +{ + uintptr_t tseg_base; + size_t tseg_size; + u32 prmrr_base; + u32 prmrr_size; + int phys_bits; + /* All range registers are aligned to 4KiB */ + const u32 rmask = ~((1 << 12) - 1); + + /* Some of the range registers are dependent on the number of physical + * address bits supported. */ + phys_bits = cpuid_eax(0x80000008) & 0xff; + + /* The range bounded by the TSEGMB and BGSM registers encompasses the + * SMRAM range as well as the IED range. However, the SMRAM available + * to the handler is 4MiB since the IEDRAM lives TSEGMB + 4MiB. + */ + smm_region(&tseg_base, &tseg_size); + + /* SMRR has 32-bits of valid address aligned to 4KiB. */ + params->smrr_base.lo = (tseg_base & rmask) | MTRR_TYPE_WRBACK; + params->smrr_base.hi = 0; + params->smrr_mask.lo = (~(tseg_size - 1) & rmask) | MTRR_PHYS_MASK_VALID; + params->smrr_mask.hi = 0; + + smm_subregion(SMM_SUBREGION_CHIPSET, ¶ms->ied_base, ¶ms->ied_size); + + /* The PRMRR and UNCORE_PRMRR are at IEDBASE + 2MiB */ + prmrr_base = (params->ied_base + (2 << 20)) & rmask; + prmrr_size = params->ied_size - (2 << 20); + + /* PRMRR has 46 bits of valid address aligned to 4KiB. It's dependent + * on the number of physical address bits supported. */ + params->prmrr_base.lo = prmrr_base | MTRR_TYPE_WRBACK; + params->prmrr_base.hi = 0; + params->prmrr_mask.lo = (~(prmrr_size - 1) & rmask) + | MTRR_PHYS_MASK_VALID; + params->prmrr_mask.hi = (1 << (phys_bits - 32)) - 1; + + /* UNCORE_PRMRR has 39 bits of valid address aligned to 4KiB. */ + params->uncore_prmrr_base.lo = prmrr_base; + params->uncore_prmrr_base.hi = 0; + params->uncore_prmrr_mask.lo = (~(prmrr_size - 1) & rmask) | + MTRR_PHYS_MASK_VALID; + params->uncore_prmrr_mask.hi = (1 << (39 - 32)) - 1; +} + +static void setup_ied_area(struct smm_relocation_params *params) +{ + char *ied_base; + + struct ied_header ied = { + .signature = "INTEL RSVD", + .size = params->ied_size, + .reserved = {0}, + }; + + ied_base = (void *)params->ied_base; + + /* Place IED header at IEDBASE. */ + memcpy(ied_base, &ied, sizeof(ied)); + + /* Zero out 32KiB at IEDBASE + 1MiB */ + memset(ied_base + (1 << 20), 0, (32 << 10)); +} + +void smm_info(uintptr_t *perm_smbase, size_t *perm_smsize, + size_t *smm_save_state_size) +{ + printk(BIOS_DEBUG, "Setting up SMI for CPU\n"); + + fill_in_relocation_params(&smm_reloc_params); + + smm_subregion(SMM_SUBREGION_HANDLER, perm_smbase, perm_smsize); + + setup_ied_area(&smm_reloc_params); + + *smm_save_state_size = sizeof(em64t101_smm_state_save_area_t); +} + +void smm_initialize(void) +{ + /* Clear the SMM state in the southbridge. */ + smm_southbridge_clear_state(); + + /* + * Run the relocation handler for on the BSP to check and set up + * parallel SMM relocation. + */ + smm_initiate_relocation(); + + if (smm_reloc_params.smm_save_state_in_msrs) + printk(BIOS_DEBUG, "Doing parallel SMM relocation.\n"); +} + +/* The default SMM entry can happen in parallel or serially. If the + * default SMM entry is done in parallel the BSP has already setup + * the saving state to each CPU's MSRs. At least one save state size + * is required for the initial SMM entry for the BSP to determine if + * parallel SMM relocation is even feasible. */ +void smm_relocate(void) +{ + /* + * If smm_save_state_in_msrs is non-zero then parallel SMM relocation + * shall take place. Run the relocation handler a second time on the + * BSP to do * the final move. For APs, a relocation handler always + * needs to be run. + */ + if (smm_reloc_params.smm_save_state_in_msrs) + smm_initiate_relocation_parallel(); + else if (!boot_cpu()) + smm_initiate_relocation(); +} + +void smm_lock(void) +{ + struct device *sa_dev = pcidev_path_on_root(SA_DEVFN_ROOT); + /* LOCK the SMM memory window and enable normal SMM. + * After running this function, only a full reset can + * make the SMM registers writable again. + */ + printk(BIOS_DEBUG, "Locking SMM.\n"); + pci_write_config8(sa_dev, SMRAM, D_LCK | G_SMRAME | C_BASE_SEG); +} diff --git a/src/soc/intel/broadwell/cpu/tsc_freq.c b/src/soc/intel/broadwell/cpu/tsc_freq.c new file mode 100644 index 0000000000..4a8a34349f --- /dev/null +++ b/src/soc/intel/broadwell/cpu/tsc_freq.c @@ -0,0 +1,14 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ + +#include <cpu/x86/msr.h> +#include <cpu/x86/tsc.h> +#include <soc/cpu.h> +#include <soc/msr.h> + +unsigned long tsc_freq_mhz(void) +{ + msr_t platform_info; + + platform_info = rdmsr(MSR_PLATFORM_INFO); + return CPU_BCLK * ((platform_info.lo >> 8) & 0xff); +} |