x86: fsp: Save stack address to CMOS for next S3 boot

At the end of pre-relocation phase, save the new stack address
to CMOS and use it as the stack on next S3 boot for fsp_init()
continuation function.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Tested-by: Stefan Roese <sr@denx.de>
diff --git a/arch/x86/cpu/cpu.c b/arch/x86/cpu/cpu.c
index f686ad4..f86eaa9 100644
--- a/arch/x86/cpu/cpu.c
+++ b/arch/x86/cpu/cpu.c
@@ -278,6 +278,14 @@
 	high_table_reserve();
 #endif
 
+#if defined(CONFIG_HAVE_ACPI_RESUME) && defined(CONFIG_HAVE_FSP)
+	/*
+	 * Save stack address to CMOS so that at next S3 boot,
+	 * we can use it as the stack address for fsp_contiue()
+	 */
+	fsp_save_s3_stack();
+#endif
+
 	return 0;
 }
 #endif
diff --git a/arch/x86/include/asm/cmos_layout.h b/arch/x86/include/asm/cmos_layout.h
new file mode 100644
index 0000000..0a0a51e
--- /dev/null
+++ b/arch/x86/include/asm/cmos_layout.h
@@ -0,0 +1,31 @@
+/*
+ * Copyright (C) 2017, Bin Meng <bmeng.cn@gmail.com>
+ *
+ * SPDX-License-Identifier:	GPL-2.0+
+ */
+
+#ifndef __CMOS_LAYOUT_H
+#define __CMOS_LAYOUT_H
+
+/*
+ * The RTC internal registers and RAM is organized as two banks of 128 bytes
+ * each, called the standard and extended banks. The first 14 bytes of the
+ * standard bank contain the RTC time and date information along with four
+ * registers, A - D, that are used for configuration of the RTC. The extended
+ * bank contains a full 128 bytes of battery backed SRAM.
+ *
+ * For simplicity in U-Boot we only support CMOS in the standard bank, and
+ * its base address starts from offset 0x10, which leaves us 112 bytes space.
+ */
+#define CMOS_BASE		0x10
+
+/*
+ * The file records all offsets off CMOS_BASE that is currently used by
+ * U-Boot for various reasons. It is put in such a unified place in order
+ * to be consistent across platforms.
+ */
+
+/* stack address for S3 boot in a FSP configuration, 4 bytes */
+#define CMOS_FSP_STACK_ADDR	CMOS_BASE
+
+#endif /* __CMOS_LAYOUT_H */
diff --git a/arch/x86/include/asm/u-boot-x86.h b/arch/x86/include/asm/u-boot-x86.h
index d2d6039..d55455f 100644
--- a/arch/x86/include/asm/u-boot-x86.h
+++ b/arch/x86/include/asm/u-boot-x86.h
@@ -54,6 +54,19 @@
 /* arch/x86/lib/... */
 int video_bios_init(void);
 
+/* arch/x86/lib/fsp/... */
+
+/**
+ * fsp_save_s3_stack() - save stack address to CMOS for next S3 boot
+ *
+ * At the end of pre-relocation phase, save the new stack address
+ * to CMOS and use it as the stack on next S3 boot for fsp_init()
+ * continuation function.
+ *
+ * @return:	0 if OK, -ve on error
+ */
+int fsp_save_s3_stack(void);
+
 void	board_init_f_r_trampoline(ulong) __attribute__ ((noreturn));
 void	board_init_f_r(void) __attribute__ ((noreturn));
 
diff --git a/arch/x86/lib/fsp/fsp_common.c b/arch/x86/lib/fsp/fsp_common.c
index f2d50ac..3397bb8 100644
--- a/arch/x86/lib/fsp/fsp_common.c
+++ b/arch/x86/lib/fsp/fsp_common.c
@@ -5,8 +5,12 @@
  */
 
 #include <common.h>
+#include <dm.h>
 #include <errno.h>
+#include <rtc.h>
 #include <asm/acpi_s3.h>
+#include <asm/cmos_layout.h>
+#include <asm/early_cmos.h>
 #include <asm/io.h>
 #include <asm/mrccache.h>
 #include <asm/post.h>
@@ -76,9 +80,36 @@
 	return cache->data;
 }
 
+#ifdef CONFIG_HAVE_ACPI_RESUME
+int fsp_save_s3_stack(void)
+{
+	struct udevice *dev;
+	int ret;
+
+	if (gd->arch.prev_sleep_state == ACPI_S3)
+		return 0;
+
+	ret = uclass_get_device(UCLASS_RTC, 0, &dev);
+	if (ret) {
+		debug("Cannot find RTC: err=%d\n", ret);
+		return -ENODEV;
+	}
+
+	/* Save the stack address to CMOS */
+	ret = rtc_write32(dev, CMOS_FSP_STACK_ADDR, gd->start_addr_sp);
+	if (ret) {
+		debug("Save stack address to CMOS: err=%d\n", ret);
+		return -EIO;
+	}
+
+	return 0;
+}
+#endif
+
 int arch_fsp_init(void)
 {
 	void *nvs;
+	int stack = CONFIG_FSP_TEMP_RAM_ADDR;
 	int boot_mode = BOOT_FULL_CONFIG;
 #ifdef CONFIG_HAVE_ACPI_RESUME
 	int prev_sleep_state = chipset_prev_sleep_state();
@@ -107,6 +138,11 @@
 				panic("Reboot System");
 			}
 
+			/*
+			 * DM is not avaiable yet at this point, hence call
+			 * CMOS access library which does not depend on DM.
+			 */
+			stack = cmos_read32(CMOS_FSP_STACK_ADDR);
 			boot_mode = BOOT_ON_S3_RESUME;
 		}
 #endif
@@ -115,7 +151,7 @@
 		 * Note the execution does not return to this function,
 		 * instead it jumps to fsp_continue().
 		 */
-		fsp_init(CONFIG_FSP_TEMP_RAM_ADDR, boot_mode, nvs);
+		fsp_init(stack, boot_mode, nvs);
 	} else {
 		/*
 		 * The second time we enter here, adjust the size of malloc()