This CL has changes that allow us to enable a configurable
ramstage, and one change that allows us to minimize PCI
scanning. Minimal scanning is a frequently requested feature.
To enable it, we add two new variables to src/Kconfig
CONFIGURABLE_RAMSTAGE
is the overall variable controlling other options for minimizing the
ramstage.
MINIMAL_PCI_SCANNING is how we indicate we wish to enable minimal
PCI scanning.
Some devices must be scanned in all cases, such as 0:0.0.
To indicate which devices we must scan, we add a new mandatory
keyword to sconfig
It is used in place of on, off, or hidden, and indicates
a device is enabled and mandatory. Mandatory
devices are always scanned. When MINIMAL_PCI_SCANNING is enabled,
ONLY mandatory devices are scanned.
We further add support in src/device/pci_device.c to manage
both MINIMAL_PCI_SCANNING and mandatory devices.
Finally, to show how this works in practice, we add mandatory
keywords to 3 devices on the qemu-q35.
TEST=
1. This is tested and working on the qemu-q35 target.
2. On CML-Hatch
Before CL:
Total Boot time: ~685ms
After CL:
Total Boot time: ~615ms
Change-Id: I2073d9f8e9297c2b02530821ebb634ea2a5c758e
Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/36221
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jeremy Soller <jeremy@system76.com>
This change adds support for allocating resources for PCI express hotplug
bridges when PCIEXP_HOTPLUG is selected. By default, this will add 32 PCI
subordinate numbers (buses), 256 MiB of prefetchable memory, 8 MiB of
non-prefetchable memory, and 8 KiB of I/O space to any device with the
PCI_EXP_SLTCAP_HPC bit set in the PCI_EXP_SLTCAP register, which
indicates hot-plugging capability. The resource allocation is configurable,
please see the PCIEXP_HOTPLUG_* variables in src/device/Kconfig.
In order to support the allocation of hotplugged PCI buses, a new field
is added to struct device called hotplug_buses. This is defaulted to
zero, but when set, it adds the hotplug_buses value to the subordinate
value of the PCI bridge. This allows devices to be plugged in and
unplugged after boot.
This code was tested on the System76 Darter Pro (darp6). Before this
change, there are not enough resources allocated to the Thunderbolt
PCI bridge to allow plugging in new devices after boot. This can be
worked around in the Linux kernel by passing a boot param such as:
pci=assign-busses,hpbussize=32,realloc
This change makes it possible to use Thunderbolt hotplugging without
kernel parameters, and attempts to match closely what our motherboard
manufacturer's firmware does by default.
Signed-off-by: Jeremy Soller <jeremy@system76.com>
Change-Id: I500191626584b83e6a8ae38417fd324b5e803afc
Reviewed-on: https://review.coreboot.org/c/coreboot/+/35946
Reviewed-by: Philipp Deppenwiese <zaolin.daisuki@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
To print times with 1 us resolution just adds unnecessary noise
when comparing logs across different boots. Furthermore, just
the printk itself is 1 ms if some slow console is enabled.
Change-Id: Ibea43124a1937f404a6e71fd9431086b2b72290a
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/37425
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
-1 shouldn't be assigned to an unsigned variable, so use an otherwise
unused constant here. Since 7 is the highest virtual LDN number, using
0xffff as PNP_SKIP_FUNCTION marker has no unwanted side effects.
Change-Id: I5e31e7ef9dad5fedfd5552963c298336c533a5e9
Signed-off-by: Felix Held <felix-coreboot@felixheld.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/37741
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
According to the POSIX standard, %p is supposed to print a pointer "as
if by %#x", meaning the "0x" prefix should automatically be prepended.
All other implementations out there (glibc, Linux, even libpayload) do
this, so we should make coreboot match. This patch changes vtxprintf()
accordingly and removes any explicit instances of "0x%p" from existing
format strings.
How to handle zero padding is less clear: the official POSIX definition
above technically says there should be no automatic zero padding, but in
practice most other implementations seem to do it and I assume most
programmers would prefer it. The way chosen here is to always zero-pad
to 32 bits, even on a 64-bit system. The rationale for this is that even
on 64-bit systems, coreboot always avoids using any memory above 4GB for
itself, so in practice all pointers should fit in that range and padding
everything to 64 bits would just hurt readability. Padding it this way
also helps pointers that do exceed 4GB (e.g. prints from MMU config on
some arm64 systems) stand out better from the others.
Change-Id: I0171b52f7288abb40e3fc3c8b874aee14b9bdcd6
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/37626
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Patrick Georgi <pgeorgi@google.com>
Reviewed-by: David Guckian
As preparation for x86_64 clean the assembly code and introduce
arch/ram_segs.h similar to existing arch/rom_segs.h.
Replace open coded segment values with the defines from the new
header.
Change-Id: Ib006cd4df59951335506b8153e9347450ec3403e
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/36321
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
There are mainboards that do not have any graphics ports connected to
the SoC. It would be senseless to initialize the iGD, thus add a new
mainboard Kconfig to hide the GOP option.
Change-Id: Ica3b3a7a0c8120c95412369a24d8d669fb59fded
Signed-off-by: Michael Niewöhner <foss@mniewoehner.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/36348
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
The value stored to 'spd_bytes_total' is never read. Now it is fixed.
This is spotted using clang-tool v9.
Also add a check if spd_bytes_used and/or spd_bytes_total are reserved
and make sure that spd_bytes_used is not greater than spd_bytes_total.
Change-Id: I426a7e64cc4c0bcced91d03387e02c8d965a21dc
Signed-off-by: Elyes HAOUAS <ehaouas@noos.fr>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/35558
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
So, the PCI to PCI bridge specification had a pitfall for us:
Originally, when decoding i/o ports for legacy VGA cycles, bridges
should only consider the 10 least significant bits of the port address.
This means all VGA registers were aliased every 1024 ports!
e.g. 0x3b0 was also decoded as 0x7b0, 0xbb0 etc.
However, it seems, we never reserved the aliased ports, resulting in
silent conflicts we preallocated resources. We neither use much
external VGA nor many i/o ports these days, so nobody noticed.
To avoid this mess, a bridge control bit (VGA16) was introduced in
2003 to enable decoding of 16-bit port addresses. As older systems
seem rather safe and well tested, and newer systems should support
this bit, we'll use it if possible and only warn if not.
With old (AGP era) hardware one will likely encounter a warning like
this:
found VGA at PCI: 06:00.0
A bridge on the path doesn't support 16-bit VGA decoding!
This is not generally fatal, but makes unnoticed resource conflicts
more likely.
Change-Id: Id7a07f069dd54331df79f605c6bcda37882a602d
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/35516
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-by: Michael Niewöhner
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Before oprom is executed, no check is performed if rom passes verification.
Add call to verified_boot_should_run_oprom() to verify the oprom.
verified_boot_should_run_oprom() expects and rom address as input pointer.
*rom is added as input parameter to should_run_oprom() which must be parsed
to verified_boot_should_run_oprom()..
BUG=N/A
TEST=Created verified binary and verify logging on Facebook FBG1701
Change-Id: Iec5092e85d34940ea3a3bb1192ea49f3bc3e5b27
Signed-off-by: Frans Hendriks <fhendriks@eltan.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/30810
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Philipp Deppenwiese <zaolin.daisuki@gmail.com>
The simple PCI config accessors are always available
under names pci_s_[read|write]_configX.
We have some use for PCI bridge configurations and
resets in romstages, so expose them.
Change-Id: Ia97a4e1f1b4c80b3dae800d80615bdc118414ed3
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/35672
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
dev_find_slot() can sometimes fail to return the desired device object
prior to full PCI enumeration. Comment the declaration and
implementation accordingly to help the user understand the problem and
avoid its usage.
Change-Id: I3fe1f24ff015d3e4f272323947f057e4c910186c
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/35632
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Read-modify-write needs to access the same register. Numerically
both used defines are 0x3e, while register implementations are
not identical but only similar.
Change-Id: I9348b855320f86868e2d3ef76d3b8d7a4ab7fae0
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/35518
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Michael Niewöhner
As discussed on the mailing list and voted upon, the coreboot project
is going to move the majority of copyrights out of the headers and into
an AUTHORS file. This will happen a bit at a time, as we'll be unifying
license headers at the same time.
Additional cleanup - Unify "Inc" to "Inc." and "LLC." to "LLC"
Signed-off-by: Martin Roth <martin@coreboot.org>
Change-Id: Ie03a3ce1f6085494bd5f38da76e2467970cf301a
Reviewed-on: https://review.coreboot.org/c/coreboot/+/35430
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Georgi <pgeorgi@google.com>
Every vga init implementation needs to cache the framebuffer state
to be able to fill the lb_framebuffer struct later on in the
fill_lb_framebuffer call. Showing the bootsplash afterwards
guarantees to have the same interface into all the vga drivers.
This is by far from ideal, as it only allows for a single driver at
compile-time and should be adapted in the future.
It was tested on the wip razer blade stealth using vgabios @ 1280x1024
and also in Qemu @ 1280x1024.
By default the qemu framebuffer will be initialized in 800x600@32.
This can be overwriten by configuration by setting
CONFIG_DRIVERS_EMULATION_QEMU_BOCHS_{X,Y}RES .
Change-Id: I4bec06d22423627e8f429c4b47e0dc9920f1464e
Signed-off-by: Johanna Schander <coreboot@mimoja.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/34599
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Many peripheral drivers across different SoCs regularly face the same
task of piping a transfer buffer into (or reading it out of) a 32-bit
FIFO register. Sometimes it's just one register, sometimes a whole array
of registers. Sometimes you actually transfer 4 bytes per register
read/write, sometimes only 2 (or even 1). Sometimes writes need to be
prefixed with one or two command bytes which makes the actual payload
buffer "misaligned" in relation to the FIFO and requires a bunch of
tricky bit packing logic to get right. Most of the times transfer
lengths are not guaranteed to be divisible by 4, which also requires a
bunch of logic to treat the potential unaligned end of the transfer
correctly.
We have a dozen different implementations of this same pattern across
coreboot. This patch introduces a new family of helper functions that
aims to solve all these use cases once and for all (*fingers crossed*).
Change-Id: Ia71f66c1cee530afa4c77c46a838b4de646ffcfb
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/34850
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Georgi <pgeorgi@google.com>
