summaryrefslogtreecommitdiff
path: root/Documentation/AMD-S3.txt
diff options
context:
space:
mode:
authorStefan Reinauer <stefan.reinauer@coreboot.org>2015-05-05 13:10:04 -0700
committerStefan Reinauer <stefan.reinauer@coreboot.org>2015-05-06 19:09:47 +0200
commit4d8b843d3740cc90c98f181d49f0d4a67cb0a1b7 (patch)
tree19306425b170975b5317a63a84b0e6f4b94b5f40 /Documentation/AMD-S3.txt
parent29ed46caccd5cea8401c5d133895fa3a9d6f5030 (diff)
Rename documentation -> Documentation
In order to be closer to the Linux kernel source tree structure, rename documentation to Documentation. Change-Id: I8690f666638ef352d201bd3c3dc1923b0d24cb12 Signed-off-by: Stefan Reinauer <stefan.reinauer@coreboot.org> Reviewed-on: http://review.coreboot.org/10110 Tested-by: build bot (Jenkins) Reviewed-by: Patrick Georgi <pgeorgi@google.com>
Diffstat (limited to 'Documentation/AMD-S3.txt')
-rw-r--r--Documentation/AMD-S3.txt153
1 files changed, 153 insertions, 0 deletions
diff --git a/Documentation/AMD-S3.txt b/Documentation/AMD-S3.txt
new file mode 100644
index 0000000000..48d4c8f3e7
--- /dev/null
+++ b/Documentation/AMD-S3.txt
@@ -0,0 +1,153 @@
+ _____ ____ _____ ______ ____ ____ ____ _______
+ / ____/ __ \| __ \| ____| _ \ / __ \ / __ \__ __|
+ | | | | | | |__) | |__ | |_) | | | | | | | | |
+ | | | | | | _ /| __| | _ <| | | | | | | | |
+ | |___| |__| | | \ \| |____| |_) | |__| | |__| | | |
+ \_____\____/|_| \_\______|____/ \____/ \____/ |_|
+
+ __ __ _____ _____ ____
+ /\ | \/ | __ \ / ____| |___ \
+ / \ | \ / | | | | | (___ __) |
+ / /\ \ | |\/| | | | | \___ \ |__ <
+ / ____ \| | | | |__| | ____) | ___) |
+ /_/ \_\_| |_|_____/ |_____/ |____/
+
+
+ S3 in Coreboot (V 1.2)
+----------------------------------------
+ Zheng Bao
+ <zheng.bao@amd.com>
+ <fishbaozi@gmail.com>
+
+Introduction
+============
+This document is about how the feature S3 is implemented on coreboot,
+specifically on AMD platform. This topic deals with ACPI spec, hardware,
+BIOS, OS. We try to help coreboot users to realize their own S3.
+
+S3 in a nutshell
+================
+The S3 sleeping state is a low wake latency sleeping state where all
+system context is lost except system memory. [1]. S3 is a ACPI
+definition.
+To enter S3, write 3 in SLP_TYPx and set the SLP_EN bit (See ACPI
+registers). But if you do that, board can not resume at where it
+sleeps, because you don't save the context. More often than not, we
+make the board go into S3 by the tools which OSes provide. For
+windows, click Start->sleep. For linux, some distribution provide a
+tools called pm-suspend, which can make the system goto S3. If
+pm-suspend is not available, we can run "echo mem > /sys/power/state",
+but this way may not save all the needed context.
+In S3 state, the power is off. So when the power button is pressed,
+BIOS runs as it does in cold boot. If BIOS didn't detect whether
+board boots or resumes, it would go the same way as boot. It is not
+what we expect. BIOS detects the SLP_TYPx. If it is 3, it means BIOS
+are waking up.
+BIOS is responsible for restore the machine state as it is before
+sleep. It needs restore the memory controller, not overwriting memory
+which is not marked as reserved. For the peripheral which loses its
+registers, BIOS needs to write the original value.
+When everything is done, BIOS needs to find out the wakeup vector
+provided by OSes and jump there. OSes also have work to do. We can go
+to linux kernel or some other open source projects to find out how they
+handle S3 resume.
+
+ACPI registers
+==============
+ACPI specification defines a group of registers. OSes handle all these
+registers to read and write status to all the platform.
+On AMD platform, these registers are provided by southbridge. For
+example, Hudson uses PMIO 60:6F to define ACPI registers.
+OSes don't have any specific driver to know where these registers
+are. BIOS has the responsibility to allocated the IO resources and
+write all these address to FADT, a ACPI defined table.
+
+Memory Layout
+=============
+Restoring memory is the most important job done by BIOS. When the
+power is off, the memory is maintained by standby power. BIOS need to
+make sure that when flow goes to OS, everything in memory should be
+the same as it was.
+
+The chip vendor will provide a way, or code, to wake up the memory
+from sleeping. In AGESA 2008 arch, it is called AmdInitResume.
+
+The BIOS itself needs some memory to run. Either, BIOS marks the erea
+as reserved in e820, or BIOS saves the content into reserved space.
+
+Here is the address Map for S3 Resume. Assumingly the total memory is 1GB.
+00000000 --- 00100000 BIOS Reserved area.
+00100000 --- 00200000 Free
+00200000 --- 01000000 Coreboot ramstage area.
+01000000 --- 2e160000 Free
+2e160000 --- 2e170000 ACPI table
+2e170000 --- 2ef70000 OSRAM
+2ef70000 --- 2efe0000 Stack in highmem
+2efe0000 --- 2f000000 heap in highmem
+2f000000 TOM
+
+AMD requirements in S3
+======================
+Chip vendor like AMD will provide bunch of routines to restore the
+board.[2]
+ * AmdS3Save: It is called in cold boot, save required register into
+ non-volatile storage. Currently, we use SPI flash to store the data.
+ * AmdInitResume: Restore the memory controller.
+ * AmdS3LateRestore: Called after AmdInitResume, restore other
+ register that memory.
+ * (SouthBridge)InitS3EarlyRestore, (SouthBridge)InitS3LateRestore:
+ Provided by Southbridge vendor code. Early is called before PCI
+ enumeration, and Late is called after that.
+
+Lifecycle of booting, sleeping and waking Coreboot and Ubuntu
+=============================================================
+1. Cold boot.
+For a system with S3 feature, the BIOS needs to save some data to
+non-volatile storage at cold boot stage. What data need to be save are
+provided by AmdS3Save. After the wrapper calls the AmdS3Save, it gets
+the VolatileStorage and NvStorage, which are where the data are
+located. It is the wrappers's responsibility to save the data.[3][4]
+Currently, the wrappers allocate a CBFS modules in BIOS image. To do
+that, the wrapper needs to have the ability to write flash chips. It
+is not as comprehensive as flashrom. But for the SST chip on Parmer,
+MX chip on Thather, coreboot works well.[5]
+
+2. OS goes in S3.
+For Linux, besides the kernel needs to do some saving, most distributions
+run some scripts. For Ubuntu, scripts are located at /usr/lib/pm-utils/sleep.d.
+ # ls /usr/lib/pm-utils/sleep.d
+ 000kernel-change 49bluetooth 90clock 95led
+ 00logging 55NetworkManager 94cpufreq 98video-quirk-db-handler
+ 00powersave 60_wpa_supplicant 95anacron 99video
+ 01PulseAudio 75modules 95hdparm-apm
+The script with lower prefix runs before the one with higher prefix.
+99video is the last one.
+Those scripts have hooks called hibernate, suspend, thaw, resume. For
+each script, suspend is called when system sleeps and wakeup is called
+when system wakeups.
+
+3. Firmware detects S3 wakeup
+As we mentioned, Firmware detects the SLP_TYPx to find out if the board
+wakes up. In romstage.c, AmdInitReset and AmdInitEarly are called
+as they are during cold boot. AmdInitResume and AmdS3LateRestore are
+called only during resume. For whole ramstage, Coreboot goes through
+almost the same way as cold boot, other than not calling the AmdInitMid,
+AmdInitLate and AmdS3Save, and restoring all the MTRRs.
+At last step of coreboot stage, coreboot finds out the wakeup vector in FADT,
+written by OS, and jump.
+
+4. OS resumes.
+When Linux resumes, all the sleeping scripts call their resume
+hooks. If we are more lucky, all the scripts can go through. More
+chances that the 99video hangs or fails to get the display
+back. Sometimes it can fixed if CONFIG_S3_VGA_ROM_RUN is unset in
+Coreboot/Kconfig. That needs more troubleshooting.
+
+
+Reference
+=========
+[1] ACPI40a, http://www.acpi.info/spec40a.htm
+[2] Coreboot Vendorcode, {top}/src/vendorcode/amd/agesa/{family}/Proc/Common/
+[3] Coreboot AGESA wrapper, {top}/src/mainboard/amd/parmer/agesawrapper.c
+[4] Coreboot AGESA wrapper, {top}/src/cpu/amd/agesa/s3_resume.c
+[5] Coreboot Southbridge, {top}/src/southbridge/amd/agesa/hudson/spi.c