After aggressive read training, program nominal Vref for the current
channel, not only channel 0. This simple mistake can easily degrade
memory margins, especially when running at high speed (overclocking).
Tested on Asus P8H61-M PRO, still boots.
Change-Id: I12630fe33c5c786c8ec131c45c27180c3887d354
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47680
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-by: Felix Held <felix-coreboot@felixheld.de>
This function simply determines the best delay for the TX DQ PIs.
Change-Id: If44c4f661d8c81fe41532ce2bfe3718392b9fe94
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47625
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Write training needs to update mode register 1, but `write_mrreg` will
clobber the IOSAV sequence. Reference code uses one four-subsequence to
unset Qoff in MR1, run the test, and finally set Qoff again. This will
be implemented in future changes, and will use the newly-added helper.
Change-Id: I06a06a7bdd43dbde34af4ea2f90e00873eefe599
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47613
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
The intent here is to clear the register, so a simple write will work.
Change-Id: I547805059e911942ac2cac7bd2165af23d926a2b
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47608
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Also fuse two per-channel loops together.
Tested on Asus P8H61-M PRO, still boots.
Change-Id: Iacc66f4364290a66d60d483055abef6e98223d16
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47607
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
There's no need to use `struct timA_minmax`, since most cases only care
about the difference between logic delay deltas. The final step does use
the minimum logic delay across all lanes, but it's a special case.
Tested on Asus P8H61-M PRO, still boots.
Change-Id: I1da95520ac915ab003e1a839685cbf5f1970eb6a
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47604
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Constify variables, and also remove pointless and-masks on mr2reg.
Tested on Asus P8H61-M PRO, still boots.
Change-Id: I3829012ff7d41f4308ee84d6fbf3b1f2803431af
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47569
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Reference code never enables SRT for Sandy Bridge, and only enables it
for Ivy Bridge when the memory frequency is at most 1066 MHz.
Tested on Asus P8H61-M PRO, still boots.
Change-Id: I50527f311340584cf8290de2114ec2694cca3a83
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47568
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
This register must be programmed if Self-Refresh Temperature range is
enabled in MR2 (bit 7). Because the memory controller needs to reprogram
MR2 when entering Self-Refresh, it needs a copy of the MR2 settings. It
also needs to know about mirrored ranks to correctly issue MRS commands.
Tested on Asus P8H61-M PRO, still boots.
Change-Id: I2e459ac7907ead75826c7d2ded42328286eb9377
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47567
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
This function is only used in two places, so move its definition closer.
Change-Id: I21d3e04de45f58cef0603b6b75119cae4b1a7aae
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47517
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Frans Hendriks <fhendriks@eltan.com>
There's no need to use and-masks here.
Tested on Asus P8H61-M PRO, still boots.
Change-Id: If06352daf53ce278dfc64102e023e4f1ea78385c
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47516
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Use bitwise negations for AND-masks and shifts for bitfields.
Tested with BUILD_TIMELESS=1, Asus P8Z77-V LX2 remains identical.
Change-Id: Id265728c362a5035ac57f84766e883608f29c398
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47511
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-by: Frans Hendriks <fhendriks@eltan.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Create some functions to program commonly-used sequences.
Tested on Asus P8H61-M PRO, still boots.
Change-Id: I1b6474ab208fe5fc2bd7f1b68eff20541fdfce9b
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47503
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
This allows deduplicating them while preserving reproducibility.
Tested with BUILD_TIMELESS=1, Asus P8H61-M PRO remains identical.
Change-Id: Ic7d1a5732296bb678b9954f80508e9f7de7ff319
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47493
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Instead of programming subsequences one-by-one, we might as well take
the whole sequence as an array and program all subsequences in one go.
Since the number of subsequences is now known in advance, handling of
global state can be simplified, which allows reusing the last sequence.
Change-Id: Ica1b2b20e04ae368f10aa236ca24d12f69464430
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47492
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Put names and expand comments for some parts of the code.
Change-Id: If1f83bf113ef08469768a9e4dd13819f76633f18
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47489
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
There's no need to use size_t to store a boolean.
Change-Id: I0069fa8d75583dc34b402004d753220943406a04
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47487
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
The only reason to write the MR values to the training result registers
is for EV (Electrical Validation) usage. The hardware doesn't need it.
Tested on Asus P8H61-M PRO, still boots.
Change-Id: I808174494729453f4ebcaa13258d735faae68d72
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47486
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
It is only used once, and can thus be moved to the same file.
Change-Id: I4ee0621449da7fa1970a475d5a2f6e66546357ea
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47485
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
It is usually written to right after programming a pattern, because its
lower byte contains the number of cachelines of the programmed pattern.
The other cases merely reset the WDB data write and compare pointers.
Change-Id: I97196d404bf70542db28499e0d2e24b7cdab07b6
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47484
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Currently the decision of whether or not to use mrc_cache in recovery
mode is made within the individual platforms' drivers (ie: fsp2.0,
fsp1.1, etc.). As this is not platform specific, but uses common
vboot infrastructure, the code can be unified and moved into
mrc_cache. The conditions are as follows:
1. If HAS_RECOVERY_MRC_CACHE, use mrc_cache data (unless retrain
switch is true)
2. If !HAS_RECOVERY_MRC_CACHE && VBOOT_STARTS_IN_BOOTBLOCK, this
means that memory training will occur after verified boot,
meaning that mrc_cache will be filled with data from executing
RW code. So in this case, we never want to use the training
data in the mrc_cache for recovery mode.
3. If !HAS_RECOVERY_MRC_CACHE && VBOOT_STARTS_IN_ROMSTAGE, this
means that memory training happens before verfied boot, meaning
that the mrc_cache data is generated by RO code, so it is safe
to use for a recovery boot.
4. Any platform that does not use vboot should be unaffected.
Additionally, we have removed the
MRC_CLEAR_NORMAL_CACHE_ON_RECOVERY_RETRAIN config because the
mrc_cache driver takes care of invalidating the mrc_cache data for
normal mode. If the platform:
1. !HAS_RECOVERY_MRC_CACHE, always invalidate mrc_cache data
2. HAS_RECOVERY_MRC_CACHE, only invalidate if retrain switch is set
BUG=b:150502246
BRANCH=None
TEST=1. run dut-control power_state:rec_force_mrc twice on lazor
ensure that memory retraining happens both times
run dut-control power_state:rec twice on lazor
ensure that memory retraining happens only first time
2. remove HAS_RECOVERY_MRC_CACHE from lazor Kconfig
boot twice to ensure caching of memory training occurred
on each boot.
Change-Id: I3875a7b4a4ba3c1aa8a3c1507b3993036a7155fc
Signed-off-by: Shelley Chen <shchen@google.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46855
Reviewed-by: Furquan Shaikh <furquan@google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Haswell Low Power variants do not have PEG at all.
Change-Id: Ia5577104b00bfc8713b54c3c43f8dcdd3bc367df
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46791
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
This change is just to align with Broadwell.
Change-Id: I25a481503f5df79502f5ae60c87e7dacb781adad
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46790
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
These are not used anywhere and are not present on Broadwell.
Change-Id: I2d1359286ac719cb5daefc955d5c6085e2949c1f
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46788
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
There's no need to perform manual shifting and masking when ACPI allows
one to painlessly describe bitfields of a register. The now-unused DVEN
definition will be dropped in a follow-up, alongside other definitions.
Change-Id: Iab6972c78c1114c8e3dfee28320ae233421ff154
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46787
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
There's no need to dynamically differentiate between traditional and Low
Power platforms at runtime, and doing so makes code reuse more complex.
Change-Id: Id40f2f5f41db00487af9115eabee8874c2399030
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46785
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
The regions TSEG, GSM, GMS should not be marked as cacheable
resources.
Change-Id: I083b096cf3ed250bca722674abe9feffdb2436d1
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47174
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Provide necessary romstage hooks to allow unblocking the memory with
SCLEAN. Note that this is slow, and took four minutes with 4 GiB of RAM.
Tested on Asrock B85M Pro4 with tboot. When Linux has tboot support
compiled in, booting as well as S3 suspend and resume are functional.
However, SINIT will TXT reset when the iGPU is enabled, and using a dGPU
will result in DMAR-related problems as soon as the IOMMU is enabled.
However, SCLEAN seems to hang sometimes. This may be because the AP
initialization that reference code does before SCLEAN is missing, but
the ACM is still able to unblock the memory. Considering that SCLEAN is
critical to recover an otherwise-bricked platform but is hardly ever
necessary, prefer having a partially-working solution over none at all.
Change-Id: I60beb7d79a30f460bbd5d94e4cba0244318c124e
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46608
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
This is just to align the code with what Broadwell does.
Change-Id: I52fb1546d049ca9fa09d0c54304ca1d79f6c4c3e
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46756
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Align cosmetics and move CTDP-specific ASL into its own file.
Tested with BUILD_TIMELESS=1, Asrock B85M Pro4 does not change.
Change-Id: I476a4e01016caa3658177b0fa8916576f4a5e0e5
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46755
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Tested with BUILD_TIMELESS=1, Google Wolf does not change.
Change-Id: I029ab0dccbf7b61d641cccf79b491fabf97ab74a
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46720
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
The message was being printed too early, possibly because it was
relocated around alongside the rest of the code.
Change-Id: I4257f6f0baa1c398aa1df9bd3274458abfaf28a6
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46690
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
This is to reduce differences between Haswell and Broadwell.
Change-Id: I8d6a8ee02e24bee22f0a7b69098ea8430095ba90
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46689
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
MRC does not use the value of SSKPD, and will overwrite it with constant
values at the end of memory initialisation. Since coreboot does not rely
on this particular bit's value, it is safe to drop the writes to set it.
MCHBAR register 0x6120 is undocumented. It is nowhere to be found in any
documentation or code I have access to; not even for Sandy/Ivy Bridge,
the platform where this mysterious register write originally came from.
These workarounds were copied from Sandy Bridge, but do not apply to
Haswell. They were dropped on Broadwell, so drop them for Haswell too.
Tested on Asrock B85M Pro4, still boots.
Change-Id: I21d9656a7595d47ac8648c08d223b7cbafd213c3
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46683
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reorder register writes to match the locking order in Broadwell.
Tested on Asrock B85M Pro4, still boots and registers are still locked.
Change-Id: Ibe15c2598fabda752c9a54eba6362621e144ad77
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46682
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Broadwell uses a 32-bit or, so also use it on Haswell for consistency.
This has no effect because MRC already locks the memory controller down.
Tested on Asrock B85M Pro4, still boots and register is still locked.
Change-Id: Ida69cd9a95a658c24b4d2558dde88b94c167a3f9
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46681
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
The Haswell System Agent BIOS Spec revision 0.6.0 indicates this
register needs to be locked, and Broadwell already locks it.
Tested on Asrock B85M Pro4, still boots and register is locked.
Change-Id: Icdeb39e2fdde1403b6ab83faed214addca863f4b
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46680
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
This register has a lock bit. The Haswell System Agent BIOS Spec
revision 0.6.0 indicates it needs to be set, thus set it. Note that
Broadwell already locks this register.
Tested on Asrock B85M Pro4, still boots and register is locked.
Change-Id: Ie23b825e708edbfc04ec0d7783f868e8632eb608
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46679
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
This register had a lock bit on Sandy Bridge, but does not on Haswell.
Moreover, the bit remains cleared on Asrock B85M Pro4 with coreboot.
Therefore, remove the write to this bit, because it has no effect.
Tested on Asrock B85M Pro4, still boots.
Change-Id: I382a6d69233ced5af069767eb61b56741ed665be
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46678
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
