Once we do CAR teardown between AmdInitResume() and
AmdS3LateRestore() we attempt to find our heap from the
temporary memory buffer instead of cache.
S3 resume is essentially broken anyways and this is not yet a
proper fix at all, but barely keeps system from halting on S3
resume.
Offset that seems arbitrary was taken from hudson/agesawrapper.c.
Change-Id: Idddf2ecde5a9d32d532071d6ba05032be730460c
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/19038
Tested-by: build bot (Jenkins)
Reviewed-by: Martin Roth <martinroth@google.com>
Producer and consumer of these buffers now appear in same file.
Also add test for uninitialized NonVolatileStorage in SPI.
Change-Id: Ibbf6581a0bf1d4bffda870fc055721627b538b92
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/19037
Tested-by: build bot (Jenkins)
Reviewed-by: Martin Roth <martinroth@google.com>
To gradually consolidate and improve AGESA board romstages,
fork the original CAR setup code as a separate file. It becomes
too messy with preprocessor to attempt make changes within the
same file, and at end of patchset original becomes obsolete.
Change-Id: I256b675b1ab9e13c2bcc956e0d67c6c03e91f2ed
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/18620
Tested-by: build bot (Jenkins)
Reviewed-by: Marc Jones <marc@marcjonesconsulting.com>
Reviewed-by: Martin Roth <martinroth@google.com>
We define AGESA_LEGACY as an implementation of mainboard
that has its romstage main completely under mainboard/
directory. We have learnt from other platforms this approach
has several downsides when it comes to making platform-wide
improvements.
We start by creating per-family romstage.c file, which
boards will gradually take into use by removing the
AGESA_LEGACY Kconfig option we here apply to all of them.
Change-Id: Id01931e185a023039a60af16a678de9966db8d65
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/18619
Tested-by: build bot (Jenkins)
Reviewed-by: Marc Jones <marc@marcjonesconsulting.com>
Reviewed-by: Martin Roth <martinroth@google.com>
When allowing use of SSE instructions, stack must be
aligned to 16 bytes. Adjust x86 entry to C accordingly,
by pushing values to maintain the alignment.
For some builds, new toolchain and GCC-6.3 could emit
SSE instruction 'andps (%esp),%xmm0' with incorrectly
aligned esp, raising exception and thus preventing boot.
Change-Id: I452d40eadac2b743d0d8431809c9a81bf28c330a
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/18691
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins)
Reviewed-by: Marshall Dawson <marshalldawson3rd@gmail.com>
The init files for the AMD families using the AGESA platform
initialization code are quite similar. So reduce the differences, by
using the same comments, variable names, console messages, and blank
lines.
Change-Id: Id4a3a5c3812a34627d726cdcbe8f4781a14be724
Signed-off-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-on: https://review.coreboot.org/18507
Tested-by: build bot (Jenkins)
Reviewed-by: Philippe Mathieu-Daudé <philippe.mathieu.daude@gmail.com>
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
This is useful for debugging S3 issues and in general
to understand AGESA memory allocator behaviour.
Change-Id: I422f2620ed0023f3920b8d2949ee1c33a6c227e0
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/18535
Reviewed-by: Nico Huber <nico.h@gmx.de>
Tested-by: build bot (Jenkins)
When allowing use of SSE instructions, stack must be
aligned to 16 bytes. Adjust x86 entry to C accordingly,
by pushing values to maintain the alignment.
Fixes regression with new toolchain using GCC-6.3 and
ec0a393 console: Enable printk for ENV_LIBAGESA
For some builds, the above-mentioned commit emitted
SSE instruction 'andps (%esp),%xmm0' with incorrectly
aligned esp, raising exception and thus preventing boot.
Change-Id: Ief57a2ea053c7497d50903838310b7f7800bff26
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/18622
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
No board with binaryPI currently supports HAVE_ACPI_RESUME. For
platforms with PSP the approach is also very different from what
we previously had here.
Furthermore, s3_resume.[ch] files under cpu/amd/pi do not
distinguish between NonVolatile and Volatile buffers of S3 storage.
This means the Volatile buffer that is maintained and available in
CBMEM is unnecessarily copied to SPI flash. This has been fixed on
open-source AGESA directory, so development of S3 suspend support
with binaryPI is better continued with that.
Unfortunately there are further complications and indications that
open-source AGESA may have always had a low-memory corruption
issue. This has to be investigated separately before restoring
or claiming S3 is supported on binaryPI.
Change-Id: I81585fff7aae7bcdd55e5e95bc373e0adef43ef0
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/18501
Tested-by: build bot (Jenkins)
Reviewed-by: Marc Jones <marc@marcjonesconsulting.com>
Relocate the enabling of the LAPIC out of the southbridge source and
surround it with a check for CONFIG_UDELAY_LAPIC (typical for AMD
systems). The LAPIC is now enabled for all cores; not only the BSP,
and not only when the UART is used.
This solves the problem of APs not having their APICs enabled when
the timer is expected to be functional, e.g. verstage often uses
do_printk_va_list() instead of do_printk() which exits early for
APs when CONFIG_SQUELCH_EARLY_SMP=y.
The changes were tested with two Gardenia builds, one using verstage
and another with CONFIG_SQUELCH_EARLY_SMP=n.
Original-Signed-off-by: Marshall Dawson <marshalldawson3rd@gmail.com>
Original-Reviewed-by: Marc Jones <marcj303@gmail.com>
(cherry picked from commit 93ffc311165f19d4192a5489051fa4264cd8e0ad)
Change-Id: Ieaecc0bf921ee0d2691a8082f2431ea4d0c33749
Signed-off-by: Marc Jones <marcj303@gmail.com>
Signed-off-by: Marshall Dawson <marshalldawson3rd@gmail.com>
Reviewed-on: https://review.coreboot.org/18436
Tested-by: build bot (Jenkins)
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Previously, all romstages for this northbridge family
would compile via 1 single C file with everything
included into the romstage.c file (!)
This patch separates the build into separate .o modules
and links them accordingly.
Currently compiles and links all fam10 roms without
breaking other roms.
Both DDR2 and DDR3 have been completed
TESTED on REACTS: passes all boot tests for 2 boards
ASUS KGPE-D16
ASUS KFSN4-DRE
Some extra changes were required to make it compile
otherwise there were unused functions in included "c" files.
This is because I needed to exchange CIMX
for the native southbridge routines. See in particular:
advansus/a785e-i
asus/m5a88-v
avalue/eax-785e
A followup patch may be required to fix the above boards.
See FIXME, XXX tags
Change-Id: Id0f9849578fd0f8b1eab83aed910902c27354426
Signed-off-by: Damien Zammit <damien@zamaudio.com>
Reviewed-on: https://review.coreboot.org/17625
Tested-by: build bot (Jenkins)
Tested-by: Raptor Engineering Automated Test Stand <noreply@raptorengineeringinc.com>
Reviewed-by: Timothy Pearson <tpearson@raptorengineering.com>
The value for _size was not evaluated correctly if ramstage
is relocated, make the calculation runtime.
While touching it, move symbol declarations to header file.
Change-Id: I4402315945771acf1c86a81cac6d43f1fe99a2a2
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/17784
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
The same pattern was being used throughout the code base
for initializing the romstage handoff structure. Provide
a helper function to initialize the structure with the S3
resume state then utilize it at all the existing call sites.
Change-Id: I1e9d588ab6b9ace67757387dbb5963ae31ceb252
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: https://review.coreboot.org/17646
Tested-by: build bot (Jenkins)
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-by: Furquan Shaikh <furquan@google.com>
We don't need to do explicit pci_io_read/write operations,
as we can use MMCONF everywhere. AGESA code still enables
extended cf8/cfc should it be needed by payload or OS.
Change-Id: Ib08028bda1b5226bb3b6b67e91f514480a9fc5ee
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/17536
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Vendorcode always does PCI MMCONF access once it is
enabled via MSR.
In coreboot proper, we don't give opportunity to make
pci_read/write calls before PCI MMCONF is enabled via MSR.
This happens early in romstage amd_initmmio() for all cores.
Change-Id: Id6ec25706b52441259e7dc1582f9a4ce8b154083
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/17534
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
We don't need to do explicit pci_io_read/write operations,
as we can use MMCONF everywhere. AGESA code still enables
extended cf8/cfc should it be required by payload or OS.
Change-Id: I278e5e26eb9a247f67927cbc67e04f081ca50f7b
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/17535
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Vendorcode always does PCI MMCONF access once it is
enabled via MSR.
In coreboot proper, we don't give opportunity to make
pci_read/write calls before PCI MMCONF is enabled via MSR.
This happens early in romstage amd_initmmio() for all cores.
Change-Id: If31bc0a67b480bcc1d955632f413f5cdeec51a54
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/17533
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
RW flag was added to spi_slave structure to get around a requirement on
some AMD flash controllers that need to group together all spi volatile
operations (write/erase). This rw flag is not a property or attribute of
the SPI slave or controller. Thus, instead of saving it in spi_slave
structure, clean up the SPI flash driver interface. This allows
chipsets/mainboards (that require volatile operations to be grouped) to
indicate beginning and end of such grouped operations.
New user APIs are added to allow users to perform probe, read, write,
erase, volatile group begin and end operations. Callbacks defined in
spi_flash structure are expected to be used only by the SPI flash
driver. Any chipset that requires grouping of volatile operations can
select the newly added Kconfig option SPI_FLASH_HAS_VOLATILE_GROUP and
define callbacks for chipset_volatile_group_{begin,end}.
spi_claim_bus/spi_release_bus calls have been removed from the SPI flash
chip drivers which end up calling do_spi_flash_cmd since it already has
required calls for claiming and releasing SPI bus before performing a
read/write operation.
BUG=None
BRANCH=None
TEST=Compiles successfully.
Change-Id: Idfc052e82ec15b6c9fa874cee7a61bd06e923fbf
Signed-off-by: Furquan Shaikh <furquan@chromium.org>
Reviewed-on: https://review.coreboot.org/17462
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Newer AMD families have multiple models within them, each often
requiring unique support. The chip_name files were starting to
have a lot of duplication. Specify the model in the name, as well
as the family.
Change-Id: I236b260e2a565e212c486347c4a633eadcdf0042
Signed-off-by: Marshall Dawson <marshalldawson3rd@gmail.com>
Reviewed-on: https://review.coreboot.org/17187
Tested-by: build bot (Jenkins)
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Martin Roth <martinroth@google.com>
Add implementation to use actual requirements of ramstage size
for S3 resume backup in CBMEM. The backup covers complete pages of 4 KiB.
Only the required amount of low memory is backed up when ACPI_TINY_LOWMEM_BACKUP
is selected for the platform. Enable this option for AGESA and binaryPI, other
platforms (without RELOCATABLE_RAMSTAGE) currently keep their romstage ramstack
in low memory for s3 resume path.
Change-Id: Ide7ce013f3727c2928cdb00fbcc7e7e84e859ff1
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/15255
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Tested-by: build bot (Jenkins)
Tested-by: Raptor Engineering Automated Test Stand <noreply@raptorengineeringinc.com>
Replace the use of the old device_t definition inside
cpu/amd/model_fxx.
Change-Id: Iac7571956ed2fb927a6b8cc88514e533f40490d0
Signed-off-by: Antonello Dettori <dev@dettori.io>
Reviewed-on: https://review.coreboot.org/16437
Tested-by: build bot (Jenkins)
Reviewed-by: Martin Roth <martinroth@google.com>
Replace the use of the old device_t definition inside
cpu/amd/family_10h-family_15h.
Change-Id: Ia1b155eeb7b67d94cf7aaa7789843a3e4ed3497a
Signed-off-by: Antonello Dettori <dev@dettori.io>
Reviewed-on: https://review.coreboot.org/16436
Tested-by: build bot (Jenkins)
Reviewed-by: Martin Roth <martinroth@google.com>
Zero-filling memory below 1 MiB resets car_migrated variable so
any CAR GLOBALs are not addressed correctly for the remaining
time in romstage. Also there is no actual need to do this as
ramstage loader handles BSS.
This fixes regression with commit 70cd54310 that broke fam10 boards
with romstage spinlocks enabled.
Change-Id: I7418821997a980ae5b818bd57e8a1b6507a543af
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/15754
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Tested-by: Raptor Engineering Automated Test Stand <noreply@raptorengineeringinc.com>
