This variable needs to be in byte so a division by 8 needs to happen.
This problem was introduced by 3cf94032b "nb/x4x/raminit: Rewrite SPD
decode and timing selection", but was probably not encountered because
such dimms are rather uncommon.
Change-Id: I2d57f5e584ac7fa1479791c239432005fe8c178d
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/22991
Reviewed-by: Felix Held <felix-coreboot@felixheld.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Displaying the whole receive enable procedure is very verbose should
only be done if CONFIG_DEBUG_RAM_SETUP is selected.
Change-Id: Ib568621e6d044624c1c0aeb6fb08945f561395c7
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/22912
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
This replaces magic values by macros and adds some comments to improve
readability.
Adds a convenient function to fetch the test address of a rank.
Also fixes the temporary memory map by changing a write to MCHBAR
0x100 to 0x110, since this is what vendor does. (No difference
observed thus far)
TESTED on DG43GT
Change-Id: I58923e4a8a756f4ae65f759e7d46e03fad39fab7
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/22328
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
This patch pushes these large default delay tables to a different file
to reduce cluttering up the actual raminit source. While doing so it
also uses more but smaller arrays and also adds the respective default
delays for DDR3 which are not yet used in this patch.
This patch add a function to set the read DQS delays instead of just
programming magic values. (This will prove useful for DQS read
training)
To prepare for adding trainings on the delay values it stores these
default delays in the sysinfo struct to program those. Later when
trainings are implemented those trained values will be used instead of
these safe default values, via using the cached sysinfo in 'mrc'
cache.
TESTED on DG43GT (still works fine)
Change-Id: I0e3676e06586ea84fc0729469946dbc9a8225934
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/22327
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-by: Felix Held <felix-coreboot@felixheld.de>
Stores information obtained from decoding dimms and receive enable
results for future use.
Depreciates using rtc nvram to store receive enable settings.
A notable change is that receive enable results are always reused, not
just on a resume from S3.
This requires cbmem to be initialized a bit earlier, right after the
raminit finished to be able to add the sysinfo struct to cbmem which
gets cached to the SPI flash in ramstage.
TESTED on Intel DG43GT with W25Q128.V. With 4 ddr2 dimms time in
raminit goes from 133,857ms (using i2c block read to fetch SPD) to
21,071ms for cached results.
Change-Id: I042dc5c52615d40781d9ef7ecd657ad0bf3ed08f
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/21677
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Felix Held <felix-coreboot@felixheld.de>
In order for ddr2.h and ddr3.h to be included in the same file it
cannot have conflicting definitions, therefore rename a few things and
move some things to a common header.
Change-Id: I6056148872076048e055f1d20a60ac31afd7cde6
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/23717
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Especially on ICH7 failing to do so results in i2c block read being
unusable. On ICH10 this problem doesn't manifest itself that much.
This moves disabling the watchdog reboot to the northbridge code like
i945 (even though it technically is southbridge stuff).
TESTED on Intel DG41WV: hacking on raminit is much nicer since no
need to do a hard power down for +4s are needed to clear the timeouts.
Change-Id: Icfd3789312704f61000a417f23a121d02d2e7fbe
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/22997
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-by: Felix Held <felix-coreboot@felixheld.de>
This is mostly written from scratch and uses common spd ddr2 decode
functions.
This improves the following:
* This fixes incorrect CAS/Freq detection on DDR2;
* Fixes tRFC computation; tRFC == 78 is a valid timing which is
excluded and 0 ends up being used; (TESTED)
* Timings selection does not use loops;
* Removes ddr3 spd decode and is re-added in follow-up patches using
common ddr3 spd functions;
* Raminit would bail out if a dimm was unsupported, now in some cases it
just marks the dimm slot as empty;
* It dramatically reduces stack usage since it does not allocate 4
times 256 bytes to store full SPDs, amongs other unused things that
were stored in sysinfo;
* Reports when no dimms are present;
* Uses i2c block read to read SPD which is about 5 times faster than
bytewise read, with a fallback to smbus mode in case of failure,
which does seem to happen when the system is forcefully powered
off.
Change-Id: I760eeaa3bd4f2bc25a517ddb1b9533c971454071
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/19143
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Martin Roth <martinroth@google.com>
This fixes some bitwise logic errors that caused the coarse offset not
to be programmed.
This fixes a regression introduced by 6d7a8c
"nb/intel/x4x/raminit: Rework receive enable calibration"
where the coarse offset doesn't get programmed anymore.
TESTED on Foxconn g41s-k on a DIMM where the final DQS receive enable
delays are close but above and below the edge of a coarse delay setting.
Change-Id: I41869815f782a2ea1178bdea006e3a7587441323
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/22816
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Port the ACPI opregion implementation that resides in
drivers/intel/gma to older platforms.
It allows to include a vbt.bin and allows GNU/Linux to load the
opregion as ASLS is being set.
Windows' Intel will likely ignore it as it relies on legacy VBIOS
to be loaded at 0xc0000.
Tested successfully on DG43GT (x4x) with vbt.bin,
with X200 (gm45) with vendor option rom and
D945GCLF (i945) with fake vbt.
Change-Id: I1896411155592b343e48cbd116e2f70fb0dbfafa
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-on: https://review.coreboot.org/21766
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Moves receive enable calibration to a separate file to lighten
raminit.c a bit.
Receive enable calibration is quite similar to gm45 so it reuses some
of its function names.
The functional changes are:
* the minimum coarse is now reset for each channel;
* on the second fine search for DQS high, TAP overflow is handled by
increasing medium;
* start coarse at CAS + 1 instead of CAS - 1. Other Intel northbridges
do the same and the results are more in line with register dumps
from vendor bios.
These might improve stability.
TESTED on ga-g41m-es2l
Change-Id: I0c970455e609d3ce96a262cbf110336a2079da4d
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/18692
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
This allows the use of the native VGA init on boards featuring DVI-I
ports. Digital output is not supported.
Change-Id: I11a4dd68746e06c7e27ecf3e765bdd0d8cf40515
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/20890
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
This does the following:
* Clarify that settings are set to the same value for each rank;
* Allows to program coarse
* Fix some style issues like white spaces between arithmetic
operators.
Change-Id: I3a9e28cfec915a0bb15789c23bea259f621b5096
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/20136
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Martin Roth <martinroth@google.com>
Rename `FRAMEBUFFER_KEEP_VESA_MODE` to `LINEAR_FRAMEBUFFER` and put
it together with new `VGA_TEXT_FRAMEBUFFER` into a choice. There are
two versions of `LINEAR_FRAMEBUFFER` that differ only in the prompt
and help text (one for `HAVE_VBE_LINEAR_FRAMEBUFFER` and one for
`HAVE_LINEAR_FRAMEBUFFER`). Due to `kconfig_lint` we have to model
that with additional symbols.
Change-Id: I9144351491a14d9bb5e650c14933b646bc83fab0
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/19804
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Like HAVE_VGA_TEXT_FRAMEBUFFER, these are selected by graphics drivers
that support a linear framebuffer. Some related settings moved to the
drivers (i.e. for rockchip/rk3288 and nvidia/tegra124) since they are
hardcoded.
Change-Id: Iff6dac5a5f61af49456bc6312e7a376def02ab00
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/19800
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
* Rename it to HAVE_VGA_TEXT_FRAMEBUFFER.
* Let drivers select it if they are in charge.
* Don't select it on the mainboard level if a driver handles it.
Change-Id: I2d9d09be9aa6d019e77460e69a245ad2d8cda4ea
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: https://review.coreboot.org/19791
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
A left-over from 5e3cb72a71 (nb/x4x: Do not enable IGD when not
supported). Should fix coverity issue 1375009. Remove a redundant
line that uses the variable `gfxsize` out of its scope and move the
variable declaration. Make sure the variable is always initialized,
drop unneeded error-handling for `get_option()` and sanitize the
read value instead.
Change-Id: Iee2beda30d8c74df0f412622c3ff3357819e386b
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: https://review.coreboot.org/19680
Reviewed-by: Philippe Mathieu-Daudé <philippe.mathieu.daude@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Currently only one board uses this northbridge in coreboot but some
patches are pending to add more.
Change-Id: If035e442d1a23674667f46a07b44c4f2b81be48c
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/19650
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
According to "Intel ® 4 Series Chipset Family datasheet" in the
description about GGC and DEVEN, CAPID0 bit46 is said to reflect the
presence of an internal graphic device. This would allow the P43 and
P45 chipset variants to work.
Change-Id: Icdaa2862f82000de6d51278098365c63b7719f7f
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/18515
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
The NGI writes to legacy VGA registers which should not happen when
VGA cycles are assigned to a different device.
TESTED on ga-g41m-es2l
Change-Id: I0a03e35c0d7f2532edd6cc5e62d1cf07dab57f60
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/19607
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
This disables VGA cycles on IGD when an external VGA device is
found. This allows PCI or PCIe devices to be the 'main' VGA device if
found, while the IGD is still available.
TESTED on ga-g41m-es2l: SeaBIOS shows payload on external GPU while
linux (4.10) can use both as a framebuffer simultaneously without any
extra configuration.
Change-Id: I74890918feb0f1ff6b971c4aaa96f1f7b75266ac
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/18504
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Computes TSEG size dynamically.
Changes the size of legacy hole to match other Intel northbirdges.
Refactor this a little by needing one less variable.
Change-Id: I0e6898c06a2bc1016eeaa3f002ff6c39657018ae
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/18511
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
This code ought not to run if ME is disabled. It also prohibits
writing to some GMCH regs like GGC bit1.
Intel ® 4 Series Chipset Family datasheet refers to this as
"ME stolen Memory lock" without actually describing this
functionality.
Change-Id: Iaa8646e535e13c44c010ccd434a5af954cf7dfbc
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/18513
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
The checkreset() function checks if raminit previously
succeeded (pmcon2 bit7 == 0). If this is not the case it will issue a
hot reset (writing 0x6 to 0xcf9). On the next attempt to boot the
system BOOT_PATH_RESET path will be taken. This boot path can only
successfully initialize memory if the system was reset from a state
where raminit succeeded, which is not the case here.
This can be fixed by issuing a cold reset instead of a hot reset.
Change-Id: Idbcf034c3777a64cc3fb92dc603d10470a6c8cb6
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/19506
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
The x4x northbridge can be paired with either an ICH7 (in the case of
g41) or an ICH10 (all other cases: g45, q45, p45, ...). Only ICH10
sometimes occurs with a descriptor, gbe and an ME region.
ICH7 is always descriptorless so it makes no sense to fix CBFS to
accommodate for those other objects.
Change-Id: I4a01dfdbce1807e44932a3ac812110382332abd8
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/19181
Tested-by: build bot (Jenkins)
Reviewed-by: Nico Huber <nico.h@gmx.de>
It makes no sense to read SPDs if the system will reset anyway.
Change-Id: Id2ad9b04860b3e4939a149eef6b619a496179ff8
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/17661
Tested-by: build bot (Jenkins)
Reviewed-by: Nico Huber <nico.h@gmx.de>
It rewrites the results of receive enable stored in the upper nvram
region, to avoid running receive enable again.
Some debug info is also printed about the self-refresh registers.
(Not enforcing a reset here, since 0 does not necessarily mean it's
not in self-refresh).
Change-Id: Ib54bc5c7b0fed6d975ffc31f037b5179d9e5600b
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/17998
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Previously the raminit failed on hot reset and to work around this
issue it unconditionally did a cold reset.
This has the following issues:
* it's slow;
* when the OS issues a hot reset some disk drives expect their 5V
power supply to remain on, which gets cut off by a cold reset,
causing data corruption.
To fix this some steps in raminit must be ommited on the reset path.
This includes receive enable calibration.
To achieve this it stores receive enable results in RTC nvram for them
to be rewritten on the resume path.
Note: The same thing needs to be done on the S3 resume path.
Calling a hot reset after raminit "outb(0x6, 0cf9)" works.
Change-Id: I6601dd90aebd071a0de7cec070487b0f9845bc30
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/18009
Tested-by: build bot (Jenkins)
Reviewed-by: Nico Huber <nico.h@gmx.de>
The results were obtained by comparing the MCHBAR registers of vendor bios
with coreboot at the same dram timings.
This fixes 2 issues:
* 1333MHz fsb CPUs were limited to 667MHz ddr2 speeds, because with
800MHz raminit failed;
* 1067MHz fsb CPUs did not boot when second dimm slot was populated.
TESTED on ga-g41m-es2l on 800, 1067 and 1333MHz CPUs with
DDR2 667 and 800MHz dimms.
Change-Id: I70f554f97b44947c2c78713b4d73a47c06d7ba60
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/18022
Tested-by: build bot (Jenkins)
Reviewed-by: Nico Huber <nico.h@gmx.de>
Nothing from that header is used or even declared since
CONFIG_HYPERTRANSPORT_PLUGIN_SUPPORT is not selected on Intel
hardware.
Change-Id: I9101eb6ffa6664a2ab45bc0b247279c916266537
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/18044
Tested-by: build bot (Jenkins)
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-by: Idwer Vollering <vidwer@gmail.com>
Currently only there is only one eaglelake board in coreboot
(ga-g41m-es2l) featuring a G41 variant northbridge.
Adding boards with a different variant (Q43, Q45, G43, G45, B43) will
require this change for graphic initialisation.
Change-Id: Ida32c563a99576b66685dfdadf9a534fd6e197dc
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/17900
Tested-by: build bot (Jenkins)
Reviewed-by: Martin Roth <martinroth@google.com>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Adapt implementation from skylake to prepare for removal of
HIGH_MEMORY_SAVE and moving on to RELOCATABLE_RAMSTAGE.
With this change, CBMEM region is set early-on as WRBACK
with MTRRs and romstage ram stack is moved to CBMEM.
Change-Id: Idee5072fd499aa3815b0d78f54308c273e756fd1
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/15791
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Fix-up for 696abfc
nb/intel/x4x: Fix and deflate `dimm_config` in raminit
It didn't fix the channel-number shifting issue as intended.
The channel index is either 0 or 1. DIMMs are counted from 0
to 3 where 0..1 covers channel 0, and 2..3 covers channel 1.
Since we have two DIMMs per channel, we have to multiply the
channel index by 2 (or shift it left by 1) to get the index
of the first DIMM in the channel. Finally, to get the offset
of a DIMM in the channel we take its index modulo 2 (again,
the number of DIMMs per channel).
Change-Id: I2784b0cb655bfe823bf5fa48b722623dfca1ddc3
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/17612
Tested-by: build bot (Jenkins)
Reviewed-by: Damien Zammit <damien@zamaudio.com>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
