From 751508ab019e0612f715ea481c550d73aa240112 Mon Sep 17 00:00:00 2001 From: Peter Stuge Date: Fri, 27 Jan 2012 22:17:09 +0100 Subject: northbridge/intel/i945: CHECK_SLFRCS_ON_RESUME Kconfig option MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Originally brought up by Sven Schnelle in March 2011 http://patchwork.coreboot.org/patch/2801/ http://www.coreboot.org/pipermail/coreboot/2011-March/064277.html On some mainboards it may be neccessary to reset early during resume from S3 if the SLFRCS register indicates that a memory channel is not guaranteed to be in self-refresh. On other mainboards, such as Lenovo X60 and T60, the check always creates false positives, effectively making it impossible to resume. The SLFRCS register is documented on page 197 of Mobile IntelĀ® 945 Express Chipset Family Datasheet Document Number: 309219-006 which is publically available, and the register indicates if a memory channel is guaranteed to be in self-refresh mode (if bit = 1), or that a memory channel *may or may not be* in self-refresh mode (if bit = 0). The register can thus only be used to positively learn that memory is in self-refresh. It is not known for sure that memory is *not* in self-refresh. The register is reset by the PWROK signal, which *should* go low during S3, and go high again when resuming, so it is unsurprising that SLFRCS has already been cleared when we read the register. Sven's measurements of the CKE signal on a ThinkPad shows that memory remains in self-refresh indefinitely, until coreboot re-initializes the memory controller, even when SLFRCS bits were = 0. Boards which require a warm reset when SLFRCS bits are cleared must now explicitly enable the check in the mainboard Kconfig file. This commit selects the new option in all existing i945 mainboards. A follow-up commit will remove the option for ThinkPads. Change-Id: I02320675efb8fde05c371ef243ba5093a4da6d11 Signed-off-by: Peter Stuge Reviewed-on: http://review.coreboot.org/590 Tested-by: build bot (Jenkins) Reviewed-by: Sven Schnelle --- src/northbridge/intel/i945/Kconfig | 9 +++++++++ src/northbridge/intel/i945/raminit.c | 4 ++-- 2 files changed, 11 insertions(+), 2 deletions(-) (limited to 'src/northbridge') diff --git a/src/northbridge/intel/i945/Kconfig b/src/northbridge/intel/i945/Kconfig index 9ba47daf2c..42cc7ce174 100644 --- a/src/northbridge/intel/i945/Kconfig +++ b/src/northbridge/intel/i945/Kconfig @@ -53,4 +53,13 @@ config MAXIMUM_SUPPORTED_FREQUENCY the board supports, despite what the chipset should be capable of. +config CHECK_SLFRCS_ON_RESUME + def_bool n + help + On some boards it may be neccessary to hard reset early + during resume from S3 if the SLFRCS register indicates that + a memory channel is not guaranteed to be in self-refresh. + On other boards the check always creates a false positive, + effectively making it impossible to resume. + endif diff --git a/src/northbridge/intel/i945/raminit.c b/src/northbridge/intel/i945/raminit.c index a7fbaa4439..d92c00672e 100644 --- a/src/northbridge/intel/i945/raminit.c +++ b/src/northbridge/intel/i945/raminit.c @@ -294,8 +294,8 @@ static void sdram_detect_errors(struct sys_info *sysinfo) reg8 |= (1<<7); pci_write_config8(PCI_DEV(0, 0x1f, 0), 0xa2, reg8); - /* clear self refresh if not wake-up from suspend */ - if (sysinfo->boot_path != 2) { + /* clear self refresh status if check is disabled or not a resume */ + if (!CONFIG_CHECK_SLFRCS_ON_RESUME || sysinfo->boot_path != 2) { MCHBAR8(0xf14) |= 3; } else { /* Validate self refresh config */ -- cgit v1.2.3