soc/intel/xeon_sp: Enable SMI handler

SMI handler was not installed for Xeon_sp platforms. This enables SMM
relocation and SMI handling.

TESTED:
- SMRR are correctly set
- The save state revision is correct (0x00030101)
- SMI's are properly generated and handled
- SMM MSR save state are not supported, so relocate SMM on all cores
in series
- Verified on OCP/Deltalake mainboard.

NOTE:
- Code for accessing a CPU save state is not working for SMMLOADERV2,
so some SMM features like GSMI, SMMSTORE, updating the ACPI GNVS
pointer are not supported.
- This hooks up to some soc/intel/common like TCO and ACPI GNVS. GNVS
is broken and needs to be fixed separately. It is unknown if TCO is
supported. This might require a cleanup in the future.

Change-Id: Iabee5c72f0245ab988d477ac8df3d8d655a2a506
Signed-off-by: Rocky Phagura <rphagura@fb.com>
Signed-off-by: Christian Walter <christian.walter@9elements.com>
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46231
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>

12 files changed, 205 insertions, 19 deletions

diff --git a/src/soc/intel/xeon_sp/Kconfig b/src/soc/intel/xeon_sp/Kconfig
index 2b1034b20605..2028a5e2199f 100644
--- a/src/soc/intel/xeon_sp/Kconfig
+++ b/src/soc/intel/xeon_sp/Kconfig
@@ -52,6 +52,7 @@ config	CPU_SPECIFIC_OPTIONS
 	select SOC_INTEL_COMMON_BLOCK_P2SB
 	select SOC_INTEL_COMMON_BLOCK_PMC
 	select SOC_INTEL_COMMON_BLOCK_PMC_DISCOVERABLE
+	select SOC_INTEL_COMMON_BLOCK_SMM
 	select SOC_INTEL_COMMON_BLOCK_TCO
 	select TSC_MONOTONIC_TIMER
 	select UDELAY_TSC
@@ -59,8 +60,11 @@ config	CPU_SPECIFIC_OPTIONS
 	select MICROCODE_BLOB_NOT_HOOKED_UP
 	select CPU_INTEL_FIRMWARE_INTERFACE_TABLE
 	select FSP_CAR
-	select NO_SMM
+	select CPU_INTEL_COMMON_SMM
 	select ACPI_INTEL_HARDWARE_SLEEP_VALUES
+	select SMM_TSEG
+	select HAVE_SMI_HANDLER
+	select X86_SMM_LOADER_VERSION2
 	select REG_SCRIPT
 
 config MAINBOARD_USES_FSP2_0
diff --git a/src/soc/intel/xeon_sp/Makefile.inc b/src/soc/intel/xeon_sp/Makefile.inc
index 2e50e64c25b2..cfda7d5d873f 100644
--- a/src/soc/intel/xeon_sp/Makefile.inc
+++ b/src/soc/intel/xeon_sp/Makefile.inc
@@ -12,6 +12,8 @@ ramstage-y += uncore.c reset.c util.c lpc.c spi.c gpio.c nb_acpi.c ramstage.c ch
 ramstage-y += memmap.c pch.c
 ramstage-$(CONFIG_SOC_INTEL_COMMON_BLOCK_PMC) += pmc.c
 ramstage-$(CONFIG_HAVE_ACPI_TABLES) += acpi.c
+ramstage-$(CONFIG_HAVE_SMI_HANDLER) += smmrelocate.c
+smm-$(CONFIG_HAVE_SMI_HANDLER) += smihandler.c pmc.c
 postcar-y += spi.c
 
 CPPFLAGS_common += -I$(src)/soc/intel/xeon_sp/include
diff --git a/src/soc/intel/xeon_sp/cpx/Kconfig b/src/soc/intel/xeon_sp/cpx/Kconfig
index 28e7b8338605..369d474552b5 100644
--- a/src/soc/intel/xeon_sp/cpx/Kconfig
+++ b/src/soc/intel/xeon_sp/cpx/Kconfig
@@ -78,6 +78,10 @@ config FSP_TEMP_RAM_SIZE
 	  documentation says this needs to be at least 128KiB, but practice
 	  show this needs to be 256KiB or more.
 
+config IED_REGION_SIZE
+	hex
+	default 0x400000
+
 config SOC_INTEL_COMMON_BLOCK_P2SB
 	def_bool y
 
diff --git a/src/soc/intel/xeon_sp/cpx/Makefile.inc b/src/soc/intel/xeon_sp/cpx/Makefile.inc
index b2e653e50e96..a5c684885229 100644
--- a/src/soc/intel/xeon_sp/cpx/Makefile.inc
+++ b/src/soc/intel/xeon_sp/cpx/Makefile.inc
@@ -5,6 +5,7 @@ ifeq ($(CONFIG_SOC_INTEL_COOPERLAKE_SP),y)
 subdirs-y += ../../../../cpu/intel/turbo
 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
 
diff --git a/src/soc/intel/xeon_sp/cpx/cpu.c b/src/soc/intel/xeon_sp/cpx/cpu.c
index d1bcbd79776c..2f4ab0663e0a 100644
--- a/src/soc/intel/xeon_sp/cpx/cpu.c
+++ b/src/soc/intel/xeon_sp/cpx/cpu.c
@@ -7,7 +7,9 @@
 #include <console/debug.h>
 #include <cpu/cpu.h>
 #include <cpu/intel/common/common.h>
+#include <cpu/intel/em64t101_save_state.h>
 #include <cpu/intel/microcode.h>
+#include <cpu/intel/smm_reloc.h>
 #include <cpu/intel/turbo.h>
 #include <cpu/x86/lapic.h>
 #include <cpu/x86/mp.h>
@@ -17,6 +19,7 @@
 #include <soc/cpu.h>
 #include <soc/msr.h>
 #include <soc/soc_util.h>
+#include <soc/smmrelocate.h>
 #include <soc/util.h>
 
 #include "chip.h"
@@ -172,16 +175,16 @@ static void post_mp_init(void)
 	/* Set Max Ratio */
 	set_max_turbo_freq();
 
-	/*
-	 * TODO: Now that all APs have been relocated as well as the BSP let SMIs
-	 * start flowing.
-	 */
-	if (0) global_smi_enable();
+	if (CONFIG(HAVE_SMI_HANDLER))
+		global_smi_enable();
 }
 
 static const struct mp_ops mp_ops = {
 	.pre_mp_init = pre_mp_init,
 	.get_cpu_count = get_thread_count,
+	.get_smm_info = get_smm_info,
+	.pre_mp_smm_init = smm_initialize,
+	.relocation_handler = smm_relocation_handler,
 	.get_microcode_info = get_microcode_info,
 	.post_mp_init = post_mp_init,
 };
diff --git a/src/soc/intel/xeon_sp/include/soc/nvs.h b/src/soc/intel/xeon_sp/include/soc/nvs.h
index becdd7657044..100a442178d3 100644
--- a/src/soc/intel/xeon_sp/include/soc/nvs.h
+++ b/src/soc/intel/xeon_sp/include/soc/nvs.h
@@ -8,10 +8,13 @@
 /* TODO - this requires xeon sp, server board support */
 /* NOTE: We do not use intelblocks/nvs.h since it includes
    mostly client specific attributes */
+
+/* TODO: This is not aligned with the ACPI asl code */
 struct __packed global_nvs {
 	uint8_t  pcnt;        /* 0x00 - Processor Count */
 	uint32_t cbmc;	      /* 0x01 - coreboot memconsole */
-	uint8_t rsvd3[251];
+	uint8_t uior;
+	uint8_t rsvd3[250];
 };
 
 #endif /* _SOC_NVS_H_ */
diff --git a/src/soc/intel/xeon_sp/include/soc/smmrelocate.h b/src/soc/intel/xeon_sp/include/soc/smmrelocate.h
new file mode 100644
index 000000000000..314ebd49dbf3
--- /dev/null
+++ b/src/soc/intel/xeon_sp/include/soc/smmrelocate.h
@@ -0,0 +1,9 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#ifndef _SOC_SMMRELOCATE_H_
+#define _SOC_SMMRELOCATE_H_
+
+void get_smm_info(uintptr_t *perm_smbase, size_t *perm_smsize,
+		  size_t *smm_save_state_size);
+
+#endif
diff --git a/src/soc/intel/xeon_sp/skx/Kconfig b/src/soc/intel/xeon_sp/skx/Kconfig
index 2e0778ee12af..08242b12e734 100644
--- a/src/soc/intel/xeon_sp/skx/Kconfig
+++ b/src/soc/intel/xeon_sp/skx/Kconfig
@@ -59,4 +59,8 @@ config HEAP_SIZE
 	hex
 	default 0x80000
 
+config IED_REGION_SIZE
+	hex
+	default 0x400000
+
 endif
diff --git a/src/soc/intel/xeon_sp/skx/Makefile.inc b/src/soc/intel/xeon_sp/skx/Makefile.inc
index f042d3689025..6ee610eded56 100644
--- a/src/soc/intel/xeon_sp/skx/Makefile.inc
+++ b/src/soc/intel/xeon_sp/skx/Makefile.inc
@@ -8,7 +8,7 @@ subdirs-y += ../../../../cpu/x86/lapic
 subdirs-y += ../../../../cpu/x86/mtrr
 subdirs-y += ../../../../cpu/x86/tsc
 subdirs-y += ../../../../cpu/x86/cache
-subdirs-$(CONFIG_HAVE_SMI_HANDLER) += ../../../cpu/x86/smm
+subdirs-$(CONFIG_HAVE_SMI_HANDLER) += ../../../../cpu/x86/smm
 
 postcar-y += soc_util.c
 
diff --git a/src/soc/intel/xeon_sp/skx/cpu.c b/src/soc/intel/xeon_sp/skx/cpu.c
index 874bcfd6d740..d36f11be669b 100644
--- a/src/soc/intel/xeon_sp/skx/cpu.c
+++ b/src/soc/intel/xeon_sp/skx/cpu.c
@@ -10,9 +10,13 @@
 #include <soc/msr.h>
 #include <soc/cpu.h>
 #include <soc/soc_util.h>
+#include <soc/smmrelocate.h>
 #include <soc/util.h>
 #include <assert.h>
 #include "chip.h"
+#include <cpu/intel/smm_reloc.h>
+#include <cpu/intel/em64t101_save_state.h>
+
 
 static const config_t *chip_config = NULL;
 
@@ -197,11 +201,8 @@ static void post_mp_init(void)
 	/* Set Max Ratio */
 	set_max_turbo_freq();
 
-	/*
-	 * TODO: Now that all APs have been relocated as well as the BSP let SMIs
-	 * start flowing.
-	 */
-	if (0) global_smi_enable();
+	if (CONFIG(HAVE_SMI_HANDLER))
+		global_smi_enable();
 }
 
 /*
@@ -214,12 +215,9 @@ static void post_mp_init(void)
 static const struct mp_ops mp_ops = {
 	.pre_mp_init = pre_mp_init,
 	.get_cpu_count = get_platform_thread_count,
-	//.get_smm_info = get_smm_info, /* TODO */
-	.get_smm_info = NULL,
-	//.pre_mp_smm_init = southcluster_smm_clear_state, /* TODO */
-	.pre_mp_smm_init = NULL,
-	//.relocation_handler = relocation_handler, /* TODO */
-	.relocation_handler = NULL,
+	.get_smm_info = get_smm_info,
+	.pre_mp_smm_init = smm_initialize,
+	.relocation_handler = smm_relocation_handler,
 	.post_mp_init = post_mp_init,
 };
 
diff --git a/src/soc/intel/xeon_sp/smihandler.c b/src/soc/intel/xeon_sp/smihandler.c
new file mode 100644
index 000000000000..db3b429e6107
--- /dev/null
+++ b/src/soc/intel/xeon_sp/smihandler.c
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+
+#include <intelblocks/smihandler.h>
+#include <soc/pm.h>
+#include <cpu/x86/smm.h>
+
+/* This is needed by common SMM code */
+const smi_handler_t southbridge_smi[SMI_STS_BITS] = {
+	[APM_STS_BIT] = smihandler_southbridge_apmc,
+	[PM1_STS_BIT] = smihandler_southbridge_pm1,
+#if CONFIG(SOC_INTEL_COMMON_BLOCK_SMM_TCO_ENABLE)
+	[TCO_STS_BIT] = smihandler_southbridge_tco,
+#endif
+};
diff --git a/src/soc/intel/xeon_sp/smmrelocate.c b/src/soc/intel/xeon_sp/smmrelocate.c
new file mode 100644
index 000000000000..a71a740955d4
--- /dev/null
+++ b/src/soc/intel/xeon_sp/smmrelocate.c
@@ -0,0 +1,144 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#include <assert.h>
+#include <string.h>
+#include <cpu/x86/mp.h>
+#include <cpu/intel/em64t101_save_state.h>
+#include <cpu/intel/smm_reloc.h>
+#include <console/console.h>
+#include <smp/node.h>
+#include <soc/msr.h>
+#include <soc/smmrelocate.h>
+
+static void fill_in_relocation_params(struct smm_relocation_params *params)
+{
+	uintptr_t tseg_base;
+	size_t tseg_size;
+
+	smm_region(&tseg_base, &tseg_size);
+
+	if (!IS_ALIGNED(tseg_base, tseg_size)) {
+		/*
+		 * Note SMRR2 is supported which might support base/size combinations.
+		 * For now it looks like FSP-M always uses aligned base/size, so let's
+		 * not care about that.
+		 */
+		printk(BIOS_WARNING,
+		       "TSEG base not aligned with TSEG SIZE! Not setting SMRR\n");
+		return;
+	}
+
+	/* SMRR has 32-bits of valid address aligned to 4KiB. */
+	if (!IS_ALIGNED(tseg_size, 4 * KiB)) {
+		printk(BIOS_WARNING,
+		       "TSEG size not aligned to the minimum 4KiB! Not setting SMRR\n");
+		return;
+	}
+
+	smm_subregion(SMM_SUBREGION_CHIPSET, &params->ied_base, &params->ied_size);
+
+	params->smrr_base.lo = tseg_base | MTRR_TYPE_WRBACK;
+	params->smrr_base.hi = 0;
+	params->smrr_mask.lo = ~(tseg_size - 1) | MTRR_PHYS_MASK_VALID;
+	params->smrr_mask.hi = 0;
+}
+
+static void setup_ied_area(struct smm_relocation_params *params)
+{
+	char *ied_base;
+
+	const struct ied_header ied = {
+		.signature = "INTEL RSVD",
+		.size = params->ied_size,
+		.reserved = {0},
+	};
+
+	ied_base = (void *)params->ied_base;
+
+	printk(BIOS_DEBUG, "IED base = 0x%08x\n", (u32)params->ied_base);
+	printk(BIOS_DEBUG, "IED size = 0x%08x\n", (u32)params->ied_size);
+
+	/* Place IED header at IEDBASE. */
+	memcpy(ied_base, &ied, sizeof(ied));
+
+	assert(params->ied_size > 1 * MiB + 32 * KiB);
+
+	/* Zero out 32KiB at IEDBASE + 1MiB */
+	memset(ied_base + 1 * MiB, 0, 32 * KiB);
+}
+
+void get_smm_info(uintptr_t *perm_smbase, size_t *perm_smsize,
+		  size_t *smm_save_state_size)
+{
+	fill_in_relocation_params(&smm_reloc_params);
+
+	smm_subregion(SMM_SUBREGION_HANDLER, perm_smbase, perm_smsize);
+
+	if (smm_reloc_params.ied_size)
+		setup_ied_area(&smm_reloc_params);
+
+	*smm_save_state_size = sizeof(em64t101_smm_state_save_area_t);
+}
+
+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;
+	int apic_id;
+	em64t101_smm_state_save_area_t *save_state;
+	/*
+	 * 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;
+
+	apic_id = cpuid_ebx(1) >> 24;
+	printk(BIOS_DEBUG, "New SMBASE=0x%08x IEDBASE=0x%08x\n apic_id=0x%x\n",
+		smbase, iedbase, apic_id);
+
+	save_state = (void *)(curr_smbase + SMM_DEFAULT_SIZE - sizeof(*save_state));
+
+	save_state->smbase = smbase;
+	save_state->iedbase = iedbase;
+}
+
+/*
+ * 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, "%s : CPU %d\n", __func__, cpu);
+
+	/* Make appropriate changes to the save state map. */
+	update_save_state(cpu, curr_smbase, staggered_smbase, relo_params);
+
+	/* Write SMRR MSRs based on indicated support. */
+	mtrr_cap = rdmsr(MTRR_CAP_MSR);
+	if (mtrr_cap.lo & SMRR_SUPPORTED)
+		write_smrr(relo_params);
+}
+
+void smm_initialize(void)
+{
+	/* Clear the SMM state in the southbridge. */
+	smm_southbridge_clear_state();
+	/* Run the relocation handler for on the BSP . */
+	smm_initiate_relocation();
+}
+
+void smm_relocate(void)
+{
+	/* Save states via MSR does not seem to be supported on CPX */
+	if (!boot_cpu())
+		smm_initiate_relocation();
+}