Improve boot time performances by replacing the wbinvd instruction
with multiple clflush to ensure that the SIPI data is written back to
RAM.
According to some experimental measurements, the wbinvd execution
takes between 1.6 up and 6 milliseconds to complete. In the case of
the SIPI data, wbinvd unnecessarily flushes and invalidates the entire
cache. Indeed, the SIPI module is quite small (about 400 bytes) and
cflush'ing the associated cache lines is almost instantaneous,
typically less than 100 microseconds.
BUG=b/260455826
TEST=Successful boot on Skolas and Rex board
Change-Id: I0e00db8eaa6a3cb41bec3422572c8f2a9bec4057
Signed-off-by: Jeremy Compostella <jeremy.compostella@intel.com>
Suggested-by: Erin Park <erin.park@intel.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/75391
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Before the bootblock stage starts setting up the CAR mode, it sends
`POST_BOOTBLOCK_CAR` POST code. However, before the definition for
`POST_BOOTBLOCK_CAR` was introduced in the commit
0d34a50a36 , the value `0x20` was used.
At that point, `0x20` means "entry into CAR mode" and `0x21` means
"the cache memory region is cleared". Right now we are sending the
same POST code twice, which makes no sense.
So we can do the following (todo: drop me after we decided which one is
more appropriate):
1) Drop it (current patchset does exactly that)
2) Introduce POST code similar to POST_SOC_CLEARING_CAR and use it
before the cache memory region is cleared.
Change-Id: I5d9014c788abdf5a4338c9e199138d1e514450b3
Signed-off-by: Alexander Goncharov <chat@joursoir.net>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/73744
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Elyes Haouas <ehaouas@noos.fr>
Reviewed-by: Lean Sheng Tan <sheng.tan@9elements.com>
In ACPI 1.0 the processor objects were inside the \_PR scope, but since
ACPI 2.0 the \_SB scope can be used for that. Outside of coreboot some
firmwares still used the \_PR scope for a while for legacy ACPI 1.0 OS
compatibility, but apart from that the \_PR scope is deprecated.
coreboot already uses the \_SB scope for the processor devices
everywhere, so move the \_SB scope out of the ACPI_CPU_STRING to the
format string inside the 3 snprintf statements that use the
ACPI_CPU_STRING.
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Suggested-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Change-Id: I76f18594a3a623b437a163c270547d3e9618c31a
Reviewed-on: https://review.coreboot.org/c/coreboot/+/75167
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Tim Wawrzynczak <inforichland@gmail.com>
Reviewed-by: Matt DeVillier <matt.devillier@amd.corp-partner.google.com>
Reviewed-by: Fred Reitberger <reitbergerfred@gmail.com>
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Instead of having the maximum number of possible CPU objects defined in
the DSDT, dynamically generate the number of needed CPU devices in the
SSDT like it's done on all other x86 platforms in coreboot.
TEST=APU2 still boots and Linux doesn't show any ACPI errors with this
patch applied and it prints "ACPI: \_SB_.P000: Found 2 idle states".
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: Id6f057ad130a27b371722fa66ce0a982afc43c6c
Reviewed-on: https://review.coreboot.org/c/coreboot/+/73073
Reviewed-by: Martin Roth <martin.roth@amd.corp-partner.google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Michał Żygowski <michal.zygowski@3mdeb.com>
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Some of the chip.h files in the tree are missing the include guards.
This patch adds them in order to avoid potential redefinions of symbols
contained in these headers, when they are included multiple times in
static.c generated by sconfig.
Change-Id: I550a514e72a8dd4db602e7ceffccd81aa36446e3
Signed-off-by: Jan Samek <jan.samek@siemens.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/74749
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Felix Held <felix-coreboot@felixheld.de>
Use coreboot's native PCI access functions instead of using the
vendorcode's PCI access functions to set up the CPU IO routing in
function 1 of the HT PCI device. This file still has room for
improvement, but at least it's now using coreboot-native functionality.
Stoneyridge has a nicer implementation, but looking into possibly
unifying those is out of scope for this patch.
TEST=None
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: Ieecc0e5f6576a838d79220b061de81e21b5d976c
Reviewed-on: https://review.coreboot.org/c/coreboot/+/74616
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Martin Roth <martin.roth@amd.corp-partner.google.com>
The BSP might have non-zero lapicid so set the topology accordingly,
without assuming it is 0. This fixes a cpu exception on at least Intel
Meteorlake. This was caused by FSP CPU PPI being giving incorrect
information about the BSP topology.
This problem was introduced by 8b8400a "drivers/fsp2_0/mp_service_ppi:
Use struct device to fill in buffer" which sets the PPI struct based on
struct device.
TESTED on google/rex
Change-Id: I3fae5efa86d8efc474c129b48bdfa1d1e2306acf
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/74374
Reviewed-by: Subrata Banik <subratabanik@google.com>
Reviewed-by: Kapil Porwal <kapilporwal@google.com>
Reviewed-by: Tarun Tuli <taruntuli@google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
On some systems the BSP cannot know how many CPUs are present in the
system. A typical use case is a multi socket system. Setting the enable
flag only on CPUs that actually exist makes it more flexible.
Change-Id: I6c8042b4d6127239175924f996f735bf9c83c6e8
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/68892
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
In the current design the relocatable parameters are used to know the
offset of the 32bit startpoint. This requires back and forward
interaction between the stub, the loader and the mp init code. This
makes the code hard to read.
This is static information known at buildtime, so a better way to deal
with this is to generate a header that contains this offset.
Change-Id: Ic01badd2af11a6e1dbc27c8e928916fedf104b5b
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/64625
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Maximilian Brune <maximilian.brune@9elements.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Use the 'verstage_x86' make target for the mode_switch.S compilation
unit instead of making adding it to the 'verstage' target depending on
VBOOT_STARTS_BEFORE_BOOTBLOCK not being selected. The only case where
VBOOT_STARTS_BEFORE_BOOTBLOCK is selected is the verstage on PSP case,
so I find using the 'verstage_x86' target here a bit easier to
understand.
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: Iab618d4b9e325b07a648b91fcdce99c63644fbfc
Reviewed-on: https://review.coreboot.org/c/coreboot/+/74196
Reviewed-by: Matt DeVillier <matt.devillier@amd.corp-partner.google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
When cbmem is initialized in romstage and postcar placed in the stage
cache + cbmem where it is run, the assumption is made that these are
all in UC memory such that calling INVD in postcar is OK.
For performance reasons (e.g. postcar decompression) it is desirable
to cache cbmem and the stage cache during romstage.
Another reason is that AGESA sets up MTRR during romstage to cache all
dram, which is currently worked around by using additional MTRR's to
make that UC.
TESTED on asus/p5ql-em, up/squared on both regular and S3 resume
bootpath. Sometimes there are minimal performance improvements
when cbmem is cached (few ms).
Change-Id: I7ff2a57aee620908b71829457ea0f5a0c410ec5b
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/37196
Reviewed-by: Lean Sheng Tan <sheng.tan@9elements.com>
Reviewed-by: Kapil Porwal <kapilporwal@google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
In certain cases data within protected memmory areas like SMRAM could
be leaked or modified if an attacker remaps PCI BARs to point within
that area. Add support to the existing SMM runtime to allow storing
PCI resources in SMRAM and then later retrieving them.
BRANCH=guybrush
BUG=b:186792595
TEST=builds
Signed-off-by: Robert Zieba <robertzieba@google.com>
Change-Id: I23fb1e935dd1b89f1cc5c834cc2025f0fe5fda37
Reviewed-on: https://review.coreboot.org/c/coreboot/+/67931
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Karthik Ramasubramanian <kramasub@google.com>
Add a Kconfig RUNTIME_CONFIGURABLE_SMM_LOGLEVEL that enables
mainboard to override mainboard_set_smm_log_level for SMM log level.
This can let SMM have different log level than other stages for
more flexibility.
Another reason is that getting certain data that requires searching
from flash VPD or CMOS is not very ideal to be done in SMM, so in this
change the value can be passed via the member variable in struct
smm_runtime and be referenced directly in SMM.
One example is that mainboard can get the desired SMM log level from
VPD/CMOS, and pass SMM console log level via the variable and in SMM
it can be referenced in get_console_loglevel() override function
directly.
Tested=On OCP Delta Lake, verified SMM log level can be overridden.
Change-Id: I81722a4f1bf75ec942cc06e403ad702dfe938e71
Signed-off-by: Johnny Lin <johnny_lin@wiwynn.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/49460
Reviewed-by: David Hendricks <david.hendricks@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jonathan Zhang <jonzhang@fb.com>
Instead of having a magic entry in the CPU device ID table list to tell
find_cpu_driver that it has reached the end of the list, introduce and
use CPU_TABLE_END. Since the vendor entry in the CPU device ID struct is
compared against X86_VENDOR_INVALID which is 0, use X86_VENDOR_INVALID
instead of the 0 in the CPU_TABLE_END definition.
TEST=Timeless build for Mandolin results in identical image.
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Suggested-by: Angel Pons <th3fanbus@gmail.com>
Change-Id: I0cae6d65b2265cf5ebf90fe1a9d885d0c489eb92
Reviewed-on: https://review.coreboot.org/c/coreboot/+/72888
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Fred Reitberger <reitbergerfred@gmail.com>
Instead of always doing exact matches between the CPUID read in
identify_cpu and the device entries of the CPU device ID table,
offer the possibility to use a bit mask in the CPUID matching. This
allows covering all steppings of a CPU family/model with one entry and
avoids that case of a missing new stepping causing the CPUs not being
properly initialized.
Some of the CPU device ID tables can now be deduplicated using the
CPUID_ALL_STEPPINGS_MASK define, but that's outside of the scope of this
patch.
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: I0540b514ca42591c0d3468307a82b5612585f614
Reviewed-on: https://review.coreboot.org/c/coreboot/+/72847
Reviewed-by: Fred Reitberger <reitbergerfred@gmail.com>
Reviewed-by: Martin Roth <martin.roth@amd.corp-partner.google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
To deprecate VBOOT_VBNV_CMOS [1], replace VBOOT_VBNV_CMOS with
VBOOT_VBNV_FLASH for samsung boards lumpy and stumpy. 0x8000 unused
flash space is allocated for RW_NVRAM.
Previously BOOT_DEVICE_SPI_FLASH_NO_EARLY_WRITES was selected for
CPU_INTEL_HASWELL, CPU_INTEL_MODEL_{2065X,206AX} and others (see [2]).
However, there seems to be no particular reason on those platforms.
We've dropped the config for haswell. Now drop it for
CPU_INTEL_MODEL_{2065X,206AX}, so that VBOOT_VBNV_FLASH can be enabled.
[1] https://web.archive.org/web/20230115020833/https://issuetracker.google.com/issues/235293589?pli=1
[2] commit 6c2568f4f5
("drivers/spi: Add BOOT_DEVICE_SPI_FLASH_NO_EARLY_WRITES config")
BUG=b:235293589
TEST=./util/abuild/abuild -a -t SAMSUNG_LUMPY -x
Change-Id: I833edd4f7a328b21e81c971ba8a9aec0aad7d3d3
Signed-off-by: Yu-Ping Wu <yupingso@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/70296
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Yidi Lin <yidilin@google.com>
This moves the definition for POST_BOOTBLOCK_CAR from the intel-specific
postcodes into the common postcode list, and uses it for the
cache-as-RAM init as needed.
Because POST_BOOTBLOCK_CAR was set to 0x20 in some spots and 0x21 in
most of the others, the values were consolidated into 0x21. This will
change the value on some platforms.
Any conflicts should get sorted out later in the conversion process.
Signed-off-by: Martin Roth <gaumless@gmail.com>
Change-Id: I8527334e679a23006b77a5645f919aea76dd4926
Reviewed-on: https://review.coreboot.org/c/coreboot/+/71596
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Elyes Haouas <ehaouas@noos.fr>
If selected, libgnat is linked into romstage. In addition, a call to
romstage_adainit() is added to support Ada program data
initialization.
BUG=b:252792591
BRANCH=firmware-brya-14505.B
TEST=Ada code compiles for romstage and loads successfully
Change-Id: I74f0460f6b14fde2b4bd6391e1782b2e5b217707
Signed-off-by: Jeremy Compostella <jeremy.compostella@intel.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/70274
Reviewed-by: Tarun Tuli <taruntuli@google.com>
Reviewed-by: Nick Vaccaro <nvaccaro@google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
