It seems that we have some applications where we need to calculate a GCD
in 64 bits. Now, we could instantiate the algorithm multiple times for
different bit width combinations to be able to use the most efficient
one for each problem... but considering that the function usually only
gets called once per callsite per stage, and that software emulation of
64-bit division on 32-bit systems doesn't take *that* long either, we
would probably usually be paying more time loading the second instance
of the function than we save with faster divisions. So let's just make
things easy and always do it in 64-bit and then nobody has to spend time
thinking on which version to call.
Change-Id: I028361444c4048a0d76ba4f80c7334a9d9983c87
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/80319
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Yidi Lin <yidilin@google.com>
In case printk does not work the current exception handler will print a
simple "!" to notify the developer that coreboot is actually there but
something went wrong.
The "!" can be quite confusing when it actually happens that printk does
not work. Since "!" doesn't really say much (if you don't know the
exception arm64 code) the developer (like me) can easily assume that
something went wrong while configuring clocks or baud rate of UART,
since the output seemingly does not seem to make sense.
This adds a little bit more output to assure the developer that what was
printed was actually intended to be printed. Therefore it prints
"EXCEPT" which assures the developer that this was intended output.
It also adds a comment above so that developer can more easily grep
for this message.
It has intentionally not been written as:
```
const char *msg = "\r\n!EXCPT!";
while (*msg)
__uart_tx_byte(*msg++);
```
because in this case the compiler will generate code that will place
`msg` somewhere in bootblock and the code will try to access this using
a memory address. In rare cases (if you link bootblock at the wrong
address) this memory address can be wrong and coreboot will not print
the message. Using individual calls to `__uart_tx_byte` ensures that the
compiler will generate code which directly puts the character bytes into
the argument register without referencing a variable in bootblock.
Signed-off-by: Maximilian Brune <maximilian.brune@9elements.com>
Change-Id: I2f858730469fff3cae120fd7c32fec53b3d309ca
Reviewed-on: https://review.coreboot.org/c/coreboot/+/80184
Reviewed-by: Julius Werner <jwerner@chromium.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Back in the days of the APIC bus, the IOAPIC IDs mustn't overlap with
the LAPIC IDs (0 to CONFIG_MAX_CPUS - 1), but since the IOAPIC and LAPIC
nowadays talk to each other via the system bus, an IOAPIC ID of 0 is
valid. When set_ioapic_id gets called with an IOAPIC ID of 0, it skipped
writing the IOAPIC ID to the corresponding IOAPIC register, so the code
was relying of the register having the expected default value of the
IOAPIC IO 0 for things to work as expected. The case of the IOAPIC ID
being 0 is the most common case in coreboot, since that's what
register_new_ioapic_gsi0 will end up doing. Fix this issue by not making
the io_apic_write call conditional on ioapic_id being non-zero. The only
southbridge that doesn't call register_new_ioapic_gsi0, calls
set_ioapic_id with the IOAPIC ID 2 for which this won't cause any
changes in behavior.
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: Ic8538f82a6b10f16eeb228669db197dc8e326ffd
Reviewed-on: https://review.coreboot.org/c/coreboot/+/80330
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Matt DeVillier <matt.devillier@amd.corp-partner.google.com>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
This patch moves the IP checksum algorithm into commonlib to prepare for
it being shared with libpayload. The current implementation is ancient
and pretty hard to read (and does some unnecessary questionable things
like the type-punning stuff which leads to suboptimal code generation),
so this reimplements it from scratch (that also helps with the
licensing).
This algorithm is prepared to take in a pre-calculated "wide" checksum
in a machine-register-sized data type which is then narrowed down to 16
bits (see RFC 1071 for why that's valid). This isn't used yet (and the
code will get optimized out), but will be used later in this patch
series for architecture-specific optimization.
Change-Id: Ic04c714c00439a17fc04a8a6e730cc2aa19b8e68
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/80251
Reviewed-by: Yidi Lin <yidilin@google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jakub Czapiga <czapiga@google.com>
Multiple links are unused throughout the tree and make the code more
confusing as an iteration over all busses is needed to get downstream
devices. This also not done consistently e.g. the allocator does not
care about multiple links on busses. A better way of dealing multiple
links below a device is to feature dummy devices with each their
respective bus.
This drops the sconfig capability to declare the same device multiple
times which was previously used to declare multiple links.
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Change-Id: Iab6fe269faef46ae77ed1ea425440cf5c7dbd49b
Reviewed-on: https://review.coreboot.org/c/coreboot/+/78328
Reviewed-by: Felix Held <felix-coreboot@felixheld.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jincheng Li <jincheng.li@intel.com>
Reviewed-by: Lean Sheng Tan <sheng.tan@9elements.com>
The .inc suffix is confusing to various tools as it's not specific to
Makefiles. This means that editors don't recognize the files, and don't
open them with highlighting and any other specific editor functionality.
This issue is also seen in the release notes generation script where
Makefiles get renamed before running cloc.
Signed-off-by: Martin Roth <gaumless@gmail.com>
Change-Id: Ice5dadd3eaadfa9962225520a3a75b05b44518ca
Reviewed-on: https://review.coreboot.org/c/coreboot/+/80066
Reviewed-by: Maximilian Brune <maximilian.brune@9elements.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Michael Niewöhner <foss@mniewoehner.de>
Reviewed-by: Felix Singer <service+coreboot-gerrit@felixsinger.de>
Instead of checking if there is more than one PCI segment group and
erroring out in that case during the build, add this requirement as a
dependency to the GENERATE_MP_TABLE Kconfig option. The mpspec.c source
file only gets included in the build if GENERATE_MP_TABLE is selected.
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Suggested-by: Martin Roth <gaumless@gmail.com>
Change-Id: Ie532a401ad0161890d0fb4ca2889af022d5f6b47
Reviewed-on: https://review.coreboot.org/c/coreboot/+/79994
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Martin Roth <martin.roth@amd.corp-partner.google.com>
Add initial support for multiple PCI segment groups. Instead of
modifying secondary in the bus struct introduce a new segment_group
struct element and keep existing common code.
Since all platforms currently only use 1 segment this is not a
functional change. On platforms that support more than 1 segment the
segment has to be set when creating the PCI domain.
Signed-off-by: Patrick Rudolph <patrick.rudolph@9elements.com>
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: Ied3313c41896362dd989ee2ab1b1bcdced840aa8
Reviewed-on: https://review.coreboot.org/c/coreboot/+/79927
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Martin Roth <martin.roth@amd.corp-partner.google.com>
Use pm_acpi_smi_cmd_port() to get the APMC trigger IO port instead of
using the hard-coded APM_CNT define. This makes sure that the correct
APMC IO port will be used even when a system doesn't use the default
APM IO port.
TEST=SMMSTORE V2 still works with the EDK2 payload on Careena
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: Icb79c91cfcd75db760bd80cff7f3d0400d1f16cd
Reviewed-on: https://review.coreboot.org/c/coreboot/+/79568
Reviewed-by: Matt DeVillier <matt.devillier@amd.corp-partner.google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
The call_smm function is currently unused and the inline assembly code
for more or less the same functionality in drivers/smmstore/ramstage is
both a bit easier to understand since it uses the register names in the
'outb' instruction instead of positional arguments, and also tells the
compiler that this piece of code might change global memory. Having too
much in the clobber list might only have some performance impact, which
should however be negligible compared to the SMI handler being called,
while missing something in the clobber list might cause hard to debug
problems.
This is a preparation to make drivers/smmstore/ramstage use call_smm
instead of having its own inline assembly implementation for this.
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: I73837cab75429014897486b38a5c56f93a850f96
Reviewed-on: https://review.coreboot.org/c/coreboot/+/79827
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Matt DeVillier <matt.devillier@amd.corp-partner.google.com>
Introduce the HAVE_CONFIGURABLE_APMC_SMI_PORT Kconfig option that when
not selected will result in a default implementation of
pm_acpi_smi_cmd_port to be included in the build that returns APM_CNT.
SoCs that provide their own pm_acpi_smi_cmd_port implementation, need to
select this Kconfig option.
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: Iaceb61b0f2a630d7afe2e0780b6a2a9806ea62f9
Reviewed-on: https://review.coreboot.org/c/coreboot/+/79566
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Rudolph <patrick.rudolph@9elements.com>
Drop the first argument specifying the number of arguments pushed
to the stack. Instead always push the 3 arguments to stack and use
the first one as function pointer to call while in protected mode.
While on it add more comments and simplify register restore code.
Tested:
- On qemu can call x86_32 function and pass argument and return
value.
- Booted Lenovo X220 in x86_64 mode using x86_32 MRC.
Change-Id: I30809453a1800ba3c0df60acd7eca778841c520f
Signed-off-by: Patrick Rudolph <patrick.rudolph@9elements.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/79752
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-by: Jérémy Compostella <jeremy.compostella@intel.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
The physical address size of the System-on-Chip (SoC) can be different
from the CPU physical address size. These two different physical
address sizes should be used for settings of their respective field.
For instance, the physical address size related to the CPU should be
used for MTRR programming while the physical address size of the SoC
should be used for MMIO resource allocation.
Typically, on Meteor Lake, the CPUs physical address size is 46 if TME
is disabled and 42 if TME is enabled but Meteor Lake SoC physical
address size is always 42. As a result, MTRRs should reflect the TME
status while coreboot MMIO resource allocator should always use
42 bits.
This commit introduces `SOC_PHYSICAL_ADDRESS_WIDTH' Kconfig to set the
physical address size of the SoC for those SoCs.
BUG=b:314886709
TEST=MTRR are aligned between coreboot and FSP
Change-Id: Icb76242718581357e5c62c2465690cf489cb1375
Signed-off-by: Jeremy Compostella <jeremy.compostella@intel.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/79665
Reviewed-by: Paul Menzel <paulepanter@mailbox.org>
Reviewed-by: Subrata Banik <subratabanik@google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Having a separate romstage is only desirable:
- with advanced setups like vboot or normal/fallback
- boot medium is slow at startup (some ARM SOCs)
- bootblock is limited in size (Intel APL 32K)
When this is not the case there is no need for the extra complexity
that romstage brings. Including the romstage sources inside the
bootblock substantially reduces the total code footprint. Often the
resulting code is 10-20k smaller.
This is controlled via a Kconfig option.
TESTED: works on qemu x86, arm and aarch64 with and without VBOOT.
Change-Id: Id68390edc1ba228b121cca89b80c64a92553e284
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/55068
Reviewed-by: Julius Werner <jwerner@chromium.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Felix Held <felix-coreboot@felixheld.de>
The value from raw_read_cntfrq_el0() could be large enough to cause
overflow when multiplied by USECS_PER_SEC. To prevent this, both
USECS_PER_SEC and tfreq can be reduced by dividing them by their GCD.
BUG=b:307790895
TEST=emerge-geralt coreboot
TEST=boot to kernel and check the timestamps from `cbmem`
Change-Id: I366667de05392913150414f0fa9058725be71c52
Signed-off-by: Yidi Lin <yidilin@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/78800
Reviewed-by: Yu-Ping Wu <yupingso@google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Having a CBFS cache scratchpad offers a generic way to decompress CBFS
files through the cbfs_map() function without having to reserve a
per-file specific memory region.
This commit introduces the x86 `RAMSTAGE_CBFS_CACHE_SIZE' Kconfig to
set a ramstage CBFS cache size. A cache size of zero disables the
CBFS cache feature. The default size is 16 KB which seems a
reasonable minimal value large enough to satisfy basic needs such as
the decompression of a small configuration file. This setting can be
adjusted depending on the platform needs and capabilities.
To support S3 suspend/resume use-case, the CBFS cache memory cannot be
released to the operating system. There are two options to meet this
requirement:
1. Define a static CBFS cache buffer (located in the .bss section)
2. Create a new CBMEM entry
Option #2 seems more powerful but considering that:
1. The CBFS cache is actually not a cache but just a scratch pad
designed to be isolated between stages
2. postcar is a very short stage not really needing CBFS cache
3. The static initialization of the `cbfs_cache' global
variable (cf. src/lib/cbfs.c) offers a simple and robust design
=> It is simpler to use a static buffer and limit the support to
ramstage.
Since some AMD SoCs (cf. `SOC_AMD_COMMON_BLOCK_NONCAR' Kconfig) define
a `_cbfs_cache' region, an extra `POSTRAM_CBFS_CACHE_IN_BSS' Kconfig
must be set to enable the use of a static buffer as the CBFS cache
scratchpad.
TEST=Decompression of vbt.bin in ramstage on rex using cbfs_map()
Change-Id: I7fbb1b51cda9f84842992e365b16c5ced1010b89
Signed-off-by: Jeremy Compostella <jeremy.compostella@intel.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/77885
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Having a CBFS cache scratchpad offers a generic way to decompress CBFS
files through the cbfs_map() function without having to reserve a
per-file specific memory region.
This commit introduces the x86 `PRERAM_CBFS_CACHE_SIZE' Kconfig to set
the pre-memory stages CBFS cache size. A cache size of zero disables
the CBFS cache feature. The default value is 16 KB which seems a
reasonable minimal value enough to satisfy basic needs such as the
decompression of a small configuration file. This setting can be
adjusted depending on the platform needs and capabilities.
We have set this size to zero for all the platforms without enough
space in Cache-As-RAM to accommodate the default size.
TEST=Decompression of vbt.bin in romstage on rex using cbfs_map()
Change-Id: Iee493f9947fddcc57576f04c3d6a2d58c7368e09
Signed-off-by: Jeremy Compostella <jeremy.compostella@intel.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/77290
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
In the cpuid helper functions eax is always written to
by the cpuid instruction, so add it to the output clobbered list.
This prevents GCC from generating code with undefined behaviour
when the function is inlined.
Test: Verified that the generated assembly is sane and runtime
tests showed no "strange" behaviour when calling cpuid
functions.
Change-Id: I5dc0bb620184a355716b9c8d4206d55554b41ab9
Signed-off-by: Patrick Rudolph <patrick.rudolph@9elements.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/78192
Reviewed-by: Felix Held <felix-coreboot@felixheld.de>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Krystian Hebel <krystian.hebel@3mdeb.com>
Reviewed-by: Paul Menzel <paulepanter@mailbox.org>
Since also some AMD CPUs have reserved physical address bits that can't
be used as normal address bits, introduce the
RESERVED_PHYSICAL_ADDRESS_BITS_SUPPORT Kconfig option which gets
selected by CPU_INTEL_COMMON, and use the new common option to configure
if the specific SoC/CPU code implements get_reserved_phys_addr_bits or
if the default of this returning 0 is used instead.
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: I0059e63a160e60ddee280635bba72d363deca7f7
Reviewed-on: https://review.coreboot.org/c/coreboot/+/78073
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Marshall Dawson <marshalldawson3rd@gmail.com>
Reviewed-by: Jérémy Compostella <jeremy.compostella@intel.com>
Reviewed-by: Matt DeVillier <matt.devillier@amd.corp-partner.google.com>
Before the cpuid(0x80000001) read in smbios_write_type4, it was
previously checked in a slightly convoluted way if the result from
cpu_cpuid_extended_level was larger than 0x80000001, but the check
should be if it is larger or equal to 0x80000001.
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Change-Id: Iabcfdb2b8b90d80baf8f4c4d2fd79f1f44866dc7
Reviewed-on: https://review.coreboot.org/c/coreboot/+/78107
Reviewed-by: Matt DeVillier <matt.devillier@amd.corp-partner.google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
