Instead of counting consecutive matches (in `j`), check for a second
match directly in the control flow. Also, add some dedicated variables:
* `tap`: Keeps track of the tap value that resulted in a match and
is eventually programmed into the hardware.
* `tap2`: Is just temporarily used to search for another edge.
Keeping `tap` sync'ed with the hardware has the benefit that we don't
need to read the programmed value back for later fixups.
Change-Id: I3ae541c39efdc695f5ca74bc757b2f009239ec93
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51903
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Move the last block of the sync DLL programming up. It's independent
of the switch/case statement that it's moved around.
Change-Id: I71bc1ca1c629e4f2f4a13474c7e2c22d1a3b65d9
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51904
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
There is no nb/amd/pi northbridge left in coreboot that could be paired
with the Bolton FCH, since the remaining nb/amd/pi northbridges all use
an integrated FCH (Avalon on Mullins and Kern on Carrizo) while Bolton
is a discrete FCH. I ran into this when verifying if the common soc/amd
GPIO functionality that gets added by selecting
SOC_AMD_COMMON_BLOCK_BANKED_GPIOS is valid for all chips selecting it
and that code isn't valid for Bolton that uses the old GPIO 100
interface.
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: Iffe876bee96e42645e1be10730b78959b1c06d59
Reviewed-on: https://review.coreboot.org/c/coreboot/+/52222
Reviewed-by: Raul Rangel <rrangel@chromium.org>
Reviewed-by: Marshall Dawson <marshalldawson3rd@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Wrap `r` in parentheses to avoid unexpected behavior with compound
expressions. Fortunately, all uses of this macro do not cause issues.
Tested with BUILD_TIMELESS=1, Roda RK9 remains identical.
Change-Id: Id0f05a507c5e7e8c50e9765261d86bae73c7b5a6
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51879
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Some cases break reproducibility if refactored, and are left as-is.
Tested with BUILD_TIMELESS=1, Asus P5QL PRO remains identical.
Change-Id: I163995c0b107860449c2f36ad63e4e4ca52decb2
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51878
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
The `CLKCFG_UPDATE` macro is copied from gm45 and unused. Correct it and
use the CLKCFG macros instead of magic values.
Tested with BUILD_TIMELESS=1, Asus P5QL PRO remains identical.
Change-Id: I17e972eba21282ac84c7afe10b7149cd1131fd07
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51877
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Breaking strings across multiple lines hurts greppability. Refactor the
code a bit to drop one indentation level, and then reflow the strings.
Change-Id: I0accdfd0d2c5f58e4da493ba0d4b5c6a067d92c3
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51876
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Bit 4 needs to be set then polled for after changing sync DLL taps.
Change-Id: I61b73998dec84710eec0d2561a6f4d88068e3373
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51872
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
These changes are not reproducible for some reason.
Change-Id: If1fcd0285c3a14686f7deb70d83a4c63d57d62fe
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51871
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
These changes are not reproducible for some reason.
Change-Id: I43b445b8af8871db87fb86747db8a35cec75716a
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51867
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
There are some cases in `northbridge_topology_init` where condensing the
operation using one macro changes the binary, and have been left as-is.
Tested with BUILD_TIMELESS=1, Asrock B85M Pro4 remains identical.
Change-Id: I59c7d1f8d816b95e86d39dcbf7bc7ce8c34f0770
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51865
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
The {MCH,DMI,EP}BAR macros can be used for both reading and writing.
While this can sometimes be useful, compile-time overflow checking is
limited. Moreover, and-masks need to be bit-wise negated, which is easy
to forget and may result in spurious overflow warnings, and silencing
them with a cast also suppresses true integer overflow issues.
To address these limitations and for consistency with the existing MMIO
API (arch/mmio.h and device/mmio.h), these macros will be replaced with
prefixed wrappers around MMIO API functions. However, existing platform
code needs to be refactored, and the risk of introducing regressions is
substantial. To minimize the risk of breakage, the bulk of the platform
code changes will be verified using reproducible builds.
This patch introduces the new accessors, to be put to use in follow-ups.
These accessors are implemented as macros so that subsequent commits can
be verified using reproducible builds. They will be replaced with actual
functions after refactoring all platforms.
Change-Id: I85376a9e2f6cd042b41036f90de7f9edc7ad4508
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51864
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
These p-suffixed helpers allow dropping pointer casts in call-sites,
which is particularly useful when accessing registers at an offset from
a base address. Move existing helpers in chipset code to arch/mmio.h and
create the rest accordingly.
Change-Id: I36a015456f7b0af1f1bf2fdff9e1ccd1e3b11747
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51862
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-by: Furquan Shaikh <furquan@google.com>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Haswell MRC.bin can return zero even when raminit did not complete
successfully. When this happens, the memory controller will not have
acknowledged raminit: the mc_init_done_ack bit in the MC_INIT_STATE_G
register will be zero, and memory accesses will lock up the system.
To handle this situation more gracefully, check the mc_init_done_ack bit
after running MRC. If the bit is not set, log a fatal error and halt.
Tested on Asrock B85M Pro4:
- With badly-seated DIMMs, MRC raminit fails and coreboot dies.
- After reseating the DIMMs, the board still boots successfully.
Change-Id: I144bf827f65cd0be319c44bf3d407ddc116b129d
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51940
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
There's no good reason to use values smaller than 2 GiB here. Well, it
increases available DRAM in 32-bit space. However, as this is a 64-bit
platform, it's highly unlikely that 32-bit limitations would cause any
issues anymore. It's more likely to have the allocator give up because
memory-mapped resources in 32-bit space don't fit within the specified
MMIO size, which can easily occur when using a discrete graphics card.
Change-Id: If585b6044f58b1e5397457f3bfa906aafc7f9297
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/52072
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
There's no good reason to use values smaller than 2 GiB here. Well, it
increases available DRAM in 32-bit space. However, as this is a 64-bit
platform, it's highly unlikely that 32-bit limitations would cause any
issues anymore. It's more likely to have the allocator give up because
memory-mapped resources in 32-bit space don't fit within the specified
MMIO size, which can easily occur when using a discrete graphics card.
Change-Id: I6cdce5f56bc94cca7065ee3e38af60d1de66e45c
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/52070
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
The `pdwm` part was supposed to be an abbreviation of `power down`, but
it is neither self-explanatory nor properly-spelled. Rename the enum.
Change-Id: I7b83c71d4534b62e18ced04eebe6a65089e1d874
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51901
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Use the actual value as it is more informative.
Change-Id: Id3bd8ccdf79d1e3fdf97cda049f81271bb017ef7
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/52073
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
These typedefs are not necessary. Remove them, and rename some elements
to avoid any confusion with other DRAM generations, such as DDR4.
Change-Id: Ibe40f33372358262c540e371f7866b06a4ac842a
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51895
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reference code does an and-or operation with zero as or-value, reading
and writing to the same address. The accessed register is 32-bit, and
reference code programs bits 22, 21, 20, 16 to zero. However, coreboot
code reads the value from bits 7..0 instead. Correct this.
Change-Id: I33bf268449c2f799321be81a02bbccff855ee1fe
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51861
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reference code does a 32-bit write, and the values don't fit in 16 bits.
Change-Id: I1195c0637b5c215a45328ebae312cf620cd4c950
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51860
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reference code uses the `0x06` as an or-mask, which makes more sense.
Change-Id: I04e5262d9ab36ae866fccd90255e4a0f85328e85
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51859
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
While the macro value is the same, the DMIBAR register is not HTBONUS1.
Tested with BUILD_TIMELESS=1, Foxconn D41S remains identical.
Change-Id: I5025f115f5a55dc782092989f3d158802d1d9353
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51858
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Commit 56823f53dc (nb/intel/ironlake:
Rewrite early QPI init) rewrote this part, but the or-value is missing
one zero. Correct this magic value to align with MRC binaries.
Change-Id: Id7a6766b3f0fe415dea70cbc54afc30f808c8b16
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51857
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
This was copied from Sandy Bridge and does not apply to Ironlake. These
offsets go past the MCHBAR window (MCHBAR size is 16 KiB on Ironlake).
Some of these writes would have collided with `DEFAULT_HECIBAR` if the
PCI resource had been reported as fixed. Remove the copy-pasted code.
Change-Id: I7688921ad7517cbd68a0c48262b29ecf7b4c396c
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51856
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
While 64-bit writes seem to work properly, there could be unknown
side-effects in some cases, e.g. when running in long mode. Since
reference code uses two 32-bit writes, follow suit.
Change-Id: I48ed3d94c7865b3a3cce52108e99cf1656b57fc2
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51855
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Both EHCI and xHCI USB controllers are inside the PCH (southbridge).
Now that mainboard USB configuration no longer depends on pei_data.h
definitions, the API declarations can be placed in southbridge code.
Tested with BUILD_TIMELESS=1, Asrock B85M Pro4 remains identical.
Change-Id: Ia21991b225482b33c5bc0dc52884674d301b28ba
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51569
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
With this change, only raminit.c uses pei_data.h definitions. With MRC
cornered, making it optional is just a matter of writing a replacement.
USB config definitions will be moved to Lynx Point code in a follow-up.
Tested on Asrock B85M Pro4, still boots and still resumes from S3.
Change-Id: I4bc405213e9b0828d9ced18677335533c7dd381d
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51440
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
There are at most 14 USB2 ports and 6 USB3 ports on LynxPoint-H, and
there are at most 10 USB2 ports and 4 USB3 ports on LynxPoint-LP. Limit
the array lengths accordingly to cause build errors on invalid configs.
Change-Id: Ieda7a1320d78dbbcb651f1715a87cd1d202a79f2
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51451
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
It's common to use the raw, unshifted I2C address in coreboot. Adapt
mainboards accordingly and perform the shift in MRC glue code.
Tested on Asrock B85M Pro4, still boots and still resumes from S3.
Change-Id: I4e4978772744ea27f4c5a88def60a8ded66520e1
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51458
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reorganize romstage.c to resemble sandybridge, and move everything that
needs `pei_data` into raminit.c function `perform_raminit`. Barring USB
settings, coreboot code no longer depends on pei_data.h definitions. It
still depends on MRC, though. For now.
Tested on Asrock B85M Pro4, still boots and still resumes from S3.
Change-Id: I433f88db5fe7a7533ab6837015647ec31fb45e88
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51449
Reviewed-by: Patrick Georgi <pgeorgi@google.com>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Mainboards do not need to know about `pei_data` to tell northbridge code
where to find the SPD data. Adjust `mb_get_spd_map` to take a pointer to
a struct instead of an array, and update all the mainboards accordingly.
Currently, the only board with memory-down in the tree is google/slippy.
Mainboard code now obtains the SPD index in `mb_get_spd_map` and adjusts
the channel population accordingly. Then, northbridge code reads the SPD
file and uses the index that was read in `mb_get_spd_map`, and copies it
to channel 0 slot 0 unconditionally. MRC only uses the first position of
the `spd_data` array, and ignores the other positions. In coreboot code,
`setup_sdram_meminfo` uses the data of each SPD index, so `copy_spd` has
to account for this.
Tested on Asrock B85M Pro4, still boots and still resumes from S3.
Change-Id: Ibaed5c6de9853db6abd08f53bbfda8800d207c3e
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51448
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Matt DeVillier <matt.devillier@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
MRC only uses the SPD data for the first index, and ignores the rest.
Moreover, index 1 corresponds to the second DIMM on the first channel,
which does not exist on ULT (only one DIMM per channel is supported).
Copy the SPD to the first DIMM on channel 1 instead. Adjust northbridge
code to retrieve the serial number from the correct SPD data block.
Tested on Google Wolf, both channels are still correctly detected.
Change-Id: Ic60ff75043e6b96a59baa9e5ebffb712a100a934
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51443
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Done for consistency with other platforms. This also drops redundant S3
resume logging, as `southbridge_detect_s3_resume` already prints it.
Tested on Asrock B85M Pro4, still boots and still resumes from S3.
Change-Id: Id96c5aedad80702ebf343dd0a351fbd4e7b1c6c1
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51438
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
The `HPET_ADDRESS` Kconfig option has the same value. Use it instead.
Change-Id: I268e949d4396aa20e38f719b36cc4e6226efe082
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/49743
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
There's no need to finalize the northbridge in SMM. This also makes
unification with Broadwell easier.
Tested on Asrock B85M Pro4, still boots and registers get locked.
Change-Id: I8b2c0d14a79e4fcd2e8985ce58542791cef9b1fe
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51157
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Add devicetree configuration parameters for mainboard-specific settings,
and provide reasonable defaults, which should usually be good enough.
This is based on Haswell SA Reference Code version 1.9.0 (Nov 2014).
Tested on Asrock B85M Pro4, registers now have the expected values.
Change-Id: I0dcdd4ca431c2ae1e62f2719c376d8bdef3054bd
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47223
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Nico Huber