SPL needs to set up the machine ready for loading 64-bit U-Boot and jumping
to it. Call the existing init routines in order to accomplish this.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Addresses should not be cast to size_t. Use uintptr_t instead.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Add a 64-bit relocation function. SPL loads U-Boot into RAM at a fixed
address and runs it. U-Boot then relocates itself to the top of RAM using
this relocation function.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Move the core relocation code into a separate function so that the checking
code can be used for 64-bit relocation also.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Use this new option to control the location of 32-bit init. This will allow
us to place this in SPL if needed.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
At present this uses u32 to store an address. We should use unsigned long
and avoid special types in function return values and parameters unless
necessary. This makes the code more portable.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
We should use unsigned long rather than u32 for addresses. Update this so
that the table-generation code builds correctly on 64-bit machines.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Commit e2f88dfd2d ("libfdt: Introduce new ARCH_FIXUP_FDT option")
allows us to skip memory setup of DTB, but a problem for ARM is that
spin_table_update_dt() and psci_update_dt() are skipped as well if
CONFIG_ARCH_FIXUP_FDT is disabled.
This commit allows us to skip only fdt_fixup_memory_banks() instead
of the whole of arch_fixup_fdt(). It will be useful when we want to
use a memory node from a kernel DTB as is, but need some fixups for
Spin-Table/PSCI.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: Alexey Brodkin <abrodkin@synopsys.com>
Acked-by: Simon Glass <sjg@chromium.org>
Fixed build error for x86:
Signed-off-by: Simon Glass <sjg@chromium.org>
Today we can compile a self-contained hello world efi test binary that
allows us to quickly verify whether the EFI loader framwork works.
We can use that binary outside of the self-contained test case though,
by providing it to a to-be-tested system via tftp.
This patch separates compilation of the helloworld.efi file from
including it in the u-boot binary for "bootefi hello". It also modifies
the efi_loader test case to enable travis to pick up the compiled file.
Because we're now no longer bloating the resulting u-boot binary, we
can enable compilation always, giving us good travis test coverage.
Signed-off-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Tom Rini <trini@konsulko.com>
On ls2080 we have a separate network fabric component which we need to
shut down before we enter Linux (or any other OS). Along with that also
comes configuration of the fabric using a description file.
Today we always stop and configure the fabric in the boot script and
(again) exit it on device tree generation. This works ok for the normal
booti case, but with bootefi the payload we're running may still want to
access the network.
So let's add a new fsl_mc command that defers configuration and stopping
the hardware to when we actually exit U-Boot, so that we can still use
the fabric from an EFI payload.
For existing boot scripts, nothing should change with this patch.
Signed-off-by: Alexander Graf <agraf@suse.de>
Reviewed-by: York Sun <york.sun@nxp.com>
[agraf: Fix x86 build]
Add compiler flags and make a few minor adjustments to support the efi
loader.
Signed-off-by: Simon Glass <sjg@chromium.org>
[agraf: Add Kconfig dep]
Signed-off-by: Alexander Graf <agraf@suse.de>
These files now need to be in a standard place so that they can be located
by generic Makefile rules. Move them to the 'lib' directory.
Signed-off-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
These files now need to be in a standard place so that they can be located
by generic Makefile rules. Move them to the 'lib' directory.
Signed-off-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
There is a build warning for three x86 boards since
write_smbios_table_wrapper() is not used. Fix it.
Fixes: e824cf3f (smbios: Allow compilation on 64bit systems)
Signed-off-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Add the required pieces to support the EFI loader on x86.
Since U-Boot only builds for 32-bit on x86, only a 32-bit EFI application
is supported. If a 64-bit kernel must be booted, U-Boot supports this
directly using FIT (see doc/uImage.FIT/kernel.its). U-Boot can act as a
payload for both 32-bit and 64-bit EFI.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
The SMBIOS generation code passes pointers as u32. That causes the compiler
to warn on casts to pointers. This patch moves all address pointers to
uintptr_t instead.
Technically u32 would be enough for the current SMBIOS2 style tables, but
we may want to extend the code to SMBIOS3 in the future which is 64bit
address capable.
Signed-off-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
We will need the SMBIOS generation function on ARM as well going forward,
so let's move it into a non arch specific location.
Signed-off-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
We need the checksum function without all the other table functionality
soon, so let's split it out into its own C file.
Signed-off-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
This should return normal errors, not device-tree errors. Fix it.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Drop init_bd_struct_r() which is no-longer used. Also drop the declaration
for init_func_spi() since this is now handled by generic board init.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Bring in a faster memmove() from Linux 4.7. This speeds up scrolling on the
display.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Now, arch/${ARCH}/include/asm/errno.h and include/linux/errno.h have
the same content. (both just wrap <asm-generic/errno.h>)
Replace all include directives for <asm/errno.h> with <linux/errno.h>.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
[trini: Fixup include/clk.]
Signed-off-by: Tom Rini <trini@konsulko.com>
Quite a few places have a bind() method which just calls dm_scan_fdt_dev().
We may as well call dm_scan_fdt_dev() directly. Update the code to do this.
Signed-off-by: Simon Glass <sjg@chromium.org>
Now that platform-specific ACPI global NVS is added, pack it into
ACPI table and get its address fixed up.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: George McCollister <george.mccollister@gmail.com>
Tested-by: George McCollister <george.mccollister@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
For any FSP-enabled boards that want to enable debug UART support,
setup_internal_uart() will be called, but this API is only available
on BayTrail platform. Change to wrap it with CONFIG_INTERNAL_UART.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
An accumulated length was incorrectly added to current each pass
through the loop. On system with more than 2 cores this caused a
corrupt MADT to be generated.
Signed-off-by: George McCollister <george.mccollister@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Make use of the newly added Kconfig options of board manufacturer
and product name to write SMBIOS tables.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Before moving 'current' pointer during ACPI table writing, we always
check the table length to see if it is larger than the table header.
Since our purpose is to generate valid tables, the check logic is
always true, which can be avoided.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
The generated AmlCode[] from IASL already has the calculated DSDT
table checksum in place. No need for us to calculate it again.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Per ACPI spec, during ACPI OS initialization, OSPM can determine
that the ACPI hardware registers are owned by SMI (by way of the
SCI_EN bit in the PM1_CNT register), in which case the ACPI OS
issues the ACPI_ENABLE command to the SMI_CMD port. The SCI_EN bit
effectively tracks the ownership of the ACPI hardware registers.
However since U-Boot does not support SMI, we report all 3 fields
in FADT (SMI_CMD, ACPI_ENABLE, ACPI_DISABLE) as zero, by following
the spec who says: these fields are reserved and must be zero on
system that does not support System Management mode.
U-Boot seems to behave in a correct way that the ACPI spec allows,
at least Linux does not complain, but apparently Windows does not
think so. During Windows bring up debugging, it is observed that
even these 3 fields are zero, Windows are still trying to issue SMI
with hardcoded SMI port address and commands, and expecting SCI_EN
to be changed by the firmware. Eventually Windows gives us a BSOD
(Blue Screen of Death) saying ACPI_BIOS_ERROR and refuses to start.
To fix this, turn on the SCI_EN bit by ourselves. With this patch,
now U-Boot can install and boot Windows 8.1/10 successfully with
the help of SeaBIOS using legacy interface (non-UEFI mode).
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Now that we already reserved high memory for configuration tables,
call high_table_malloc() to allocate tables from the region.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Currently when CONFIG_SEABIOS is on, U-Boot allocates configuration
tables via normal malloc(). To simplify, use a dedicated memory
region which is reserved on the stack before relocation for this
purpose. Add functions for reserve and malloc.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
PIRQ routing table checksum is fixed up in copy_pirq_routing_table(),
which is fine if we only write the configuration table once. But with
the SeaBIOS case, when we write the table for the second time, the
checksum will be fixed up to zero per the checksum algorithm, which
is caused by the checksum field not being zero before fix up, since
the checksum has already been calculated in the first run.
To fix this, move the checksum fixup to create_pirq_routing_table(),
so that copy_pirq_routing_table() only does what its function name
suggests: copy the table to somewhere else.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
At present board_final_cleanup() is called before booting a Linux
kernel. This actually needs to be done before booting anything,
like SeaBIOS, VxWorks or Windows.
Move the call to last_stage_init() instead.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
CONFIG_GENENRATE_ACPI_TABLE controls the generation of ACPI table which
uses U-Boot's built-in methods and CONFIG_QEMU_ACPI_TABLE controls whether
to load ACPI table from QEMU's fw_cfg interface.
But with commit "697ec431469ce0a4c2fc2c02d8685d907491af84 x86: qemu: Drop
our own ACPI implementation", there is only one way to support ACPI table
for QEMU targets which is the fw_cfg interface. Having two Kconfig options
for this purpose is not necessary any more, so this patch consolidates
the two.
Signed-off-by: Miao Yan <yanmiaobest@gmail.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
- Move the command portion of arch/x86/cpu/qemu/fw_cfg.c into
cmd/qemu_fw_cfg.c
- Move arch/x86/include/asm/fw_cfg.h to include/qemu_fw_cfg.h
- Rename ACPI table portion to arch/x86/cpu/qemu/acpi_table.c
Signed-off-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Like other MADT table write routines, make acpi_create_madt_lapics()
return how many bytes it has written instead of the table end addr.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
The comment of initializing table header revision says:
/* ACPI 1.0/2.0: 1, ACPI 3.0: 2, ACPI 4.0: 3 */
which might mislead it may increase per ACPI spec revision.
However this is not the case. It's actually a fixed number
as defined in ACPI spec, and in the laest ACPI spec 6.1,
some table header revisions are still 1. Clean these up.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Per ACPI spec, the FACS table address must be aligned to a 64 byte
boundary (Windows checks this, but Linux does not).
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Tested-by: Stefan Roese <sr@denx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
Use u32 instead of unsigned long in the table write routines, as
other routines do.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Rearrange the routine order a little bit, to follow the order
in which ACPI table is defined in acpi_table.h.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Tested-by: Stefan Roese <sr@denx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
Rename fill_header() to acpi_fill_header() for consistency.
Change its signature to remove the 'length' parameter and
make it a public API.
Also remove the unnecessary include files, and improve the
AmlCode[] comment a little bit.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Tested-by: Stefan Roese <sr@denx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
This acpi_create_ssdt_generator() currently does nothing.
Remove this for now.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Tested-by: Stefan Roese <sr@denx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
- Use "U-BOOT" and "U-BOOTBL" for the OEM ID and OEM table ID.
- Do not typedef acpi_header_t, instead use struct acpi_table_hader.
- Use a shorter name aslc_id and aslc-revision.
- Change MCFG base address to use 32-bit value pairs (_l and _h).
- Apply ACPI_APIC_ prefix to MADT APIC type macros and make
their names to be more readable.
- Apply __packed to struct acpi_madt_irqoverride and struct
acpi_madt_lapic_nmi tables, as they are not naturally aligned
by the compiler which leads to wrong sizeof(struct).
- Rename model to res1 as it is reserved after ACPI spec 1.0.
- Apply ACPI_ prefix to the PM profile macros and change them
to enum.
- Add ospm_flags to FACS structure which is defined since ACPI 4.0.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Tested-by: Stefan Roese <sr@denx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
Fix the following two build warnings in function 'write_acpi_tables':
warning: format '%lx' expects argument of type 'long unsigned int',
but argument 2 has type 'u32' [-Wformat=]
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Tested-by: Stefan Roese <sr@denx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
The following build warning is seen in tables.c:
warning: implicit declaration of function 'memalign'
Add the missing header file to fix it.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Tested-by: Stefan Roese <sr@denx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
Add a driver which sets up the pin configuration on x86 devices with an ICH6
(or later) Platform Controller Hub.
The driver is not in the pinctrl uclass due to some oddities of the way x86
devices work:
- The GPIO controller is not present in I/O space until it is set up
- This is done by writing a register in the PCH
- The PCH has a driver which itself uses PCI, another driver
- The pinctrl uclass requires that a pinctrl device be available before any
other device can be probed
It would be possible to work around the limitations by:
- Hard-coding the GPIO address rather than reading it from the PCH
- Using special x86 PCI access to set the GPIO address in the PCH
However it is not clear that this is better, since the pin configuration
driver does not actually provide normal pin configuration services - it
simply sets up all the pins statically when probed. While this remains the
case, it seems better to use a syscon uclass instead. This can be probed
whenever it is needed, without any limitations.
Also add an 'invert' property to support inverting the input.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Each CPU needs to have its microcode loaded. Add support for this so that
all CPUs will have the same version.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
SeaBIOS is an open source implementation of a 16-bit x86 BIOS.
It can run in an emulator or natively on x86 hardware with the
use of coreboot. With SeaBIOS's help, we can boot some OSes
that require 16-bit BIOS services like Windows/DOS.
As U-Boot, we have to manually create a table where SeaBIOS gets
system information (eg: E820) from. The table unfortunately has
to follow the coreboot table format as SeaBIOS currently supports
booting as a coreboot payload.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
To prepare generating coreboot table from U-Boot, implement functions
to handle the writing.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
For those secondary bootloaders like SeaBIOS who want to live in
the F segment, which conflicts the configuration table address,
now we allow write_tables() to write the configuration tables in
high area (malloc'ed memory).
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Given all table write routines have the same signature, we can
simplify the codes by using a function table.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Change the parameter and return value of write_acpi_tables() to u32
to conform with other table write routines.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Add a new variable rom_table_start and pass it to ROM table write
routines. This reads better than previous single rom_table_end.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Intel IvyBridge FSP seems to be buggy that it does not report memory
used by FSP itself as reserved in the resource descriptor HOB. The
FSP specification does not describe how resource descriptor HOBs are
generated by the FSP to describe what memory regions. It looks newer
FSPs like Queensbay and BayTrail do not have such issue. This causes
U-Boot relocation overwrites the important boot service data which is
used by FSP, and the subsequent call to fsp_notify() will fail.
To resolve this, we find out the lowest memory base address allocated
by FSP for the boot service data when walking through the HOB list in
fsp_get_usable_lowmem_top(). Check whether the memory top address is
below the FSP HOB list, and if not, use the lowest memory base address
allocated by FSP as the memory top address.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested on link (ivybridge non-FSP)
Tested-by: Simon Glass <sjg@chromium.org>
Correct spelling of "U-Boot" shall be used in all written text
(documentation, comments in source files etc.).
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Heiko Schocher <hs@denx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Minkyu Kang <mk7.kang@samsung.com>
Now that we have converted all x86 codes to DM PCI, drop pci_type1.c
which is only built for legacy PCI. Also per checkpatch.pl warning,
DEFINE_PCI_DEVICE_TABLE is now deprecated so drop that too.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
At present irq_router is declared as a static struct irq_router in
arch/x86/cpu/irq.c. Since it's a driver control block, it makes sense
to move it to a per driver priv. Adjust existing APIs to accept an
additional parameter of irq_router's udevice.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
This patch adds the ability to load and link ACPI tables provided by QEMU.
QEMU tells guests how to load and patch ACPI tables through its fw_cfg
interface, by adding a firmware file 'etc/table-loader'. Guests are
supposed to parse this file and execute corresponding QEMU commands.
Signed-off-by: Miao Yan <yanmiaobest@gmail.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
This patch adds a parameter to the function setup_early_uart() to either
enable or disable the internal BayTrail legacy UART. Since the name
setup_early_uart() does not match its functionality any more, lets
rename it to setup_internal_uart() as well in this patch.
Signed-off-by: Stefan Roese <sr@denx.de>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
When the final MRC cache record is the same as the one we want to write, we
skip writing since there is no point. This is normal behaviour.
Avoiding printing an error when this happens.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Add a uclass for the northbridge / SDRAM controller found on some older
Intel chipsets.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Instead of searching for the device tree node, use the IRQ device which has
a record of it.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
A Platform Controller Hub is an Intel concept - it is like the peripherals
on an SoC and is often in a separate chip from the CPU. The chip is typically
found on the first PCI bus and integrates multiple devices.
We have a very simple uclass to support PCHs. Add a few operations, such as
setting up the devices on the PCH and finding the SPI controller base
address. Also move it into drivers/pch/ since we will be adding a few PCH
drivers.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
In a number of places we had wordings of the GPL (or LGPL in a few
cases) license text that were split in such a way that it wasn't caught
previously. Convert all of these to the correct SPDX-License-Identifier
tag.
Signed-off-by: Tom Rini <trini@konsulko.com>
board_init_f_mem() alters the C runtime environment's
stack it is actually already using. This is not a valid
behaviour within a C runtime environment.
Split board_init_f_mem into C functions which do not alter
their own stack and always behave properly with respect to
their C runtime environment.
Signed-off-by: Albert ARIBAUD <albert.u.boot@aribaud.net>
Acked-by: Thomas Chou <thomas@wytron.com.tw>
In the 'fsp hob' command output, decimal numbers and hexadecimal
numbers are used mixedly. Now change to always use hex numbers
to keep consistency.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
fsp_init() runtime buffer parameter might be different across
different platforms. Move this to update_fsp_configs().
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
Not every FSP supports UPD, thus we introduce a Kconfig option
CONFIG_FSP_USE_UPD and use it to wrap these common UPD handling
codes in fsp_support.c.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
To support platform-specific configurations (might not always be
UPD on some platform), use a better name update_fsp_configs() and
accepct struct fsp_config_data as its parameter so that platform
codes can handle whatever configuration data for that FSP.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
FSP has several config data like UPD, HDA verb table which can be
overridden or provided by bootloader. Currently in U-Boot only UPD
is handled via struct shared_data. To accommodate any platform, we
rename shared_data to fsp_config_data and move the definition from
common place fsp_support.h to platform-specific place fsp_configs.h.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
Save boot_mode in struct shared_data for future refactoring.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
Declare stack_top as u32 in struct shared_data and struct common_buf
so that we can avoid casting in fsp_init().
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
There is no need to pass shared_data to fsp_continue() so we can
remove unnecessary codes that simplifies the function a lot.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
This function should take a struct udevice rather than pci_dev_t. Update it.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Adjust this code to use driver model for devices where possible. Since
existing users have not been converted the old code must remain.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
To group all dm timer drivers together, move tsc timer to
drivers/timer directory.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Now that we have converted all x86 boards to use driver model timer,
remove these legacy timer codes in the tsc driver.
Note this also removes the TSC_CALIBRATION_BYPASS Kconfig option,
as it is not needed with driver model.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Replace __attribute__((no_instrument_function)) with notrace from
<linux/compiler.h>.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Now that we have converted all x86 boards to use driver model pci,
remove these legacy pci codes.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Rename pcat_timer.c to i8254.c and pcat_interrupts.c to i8259.c,
to match their header file names (i8254.h and i8259.h).
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Initialize counter 1, used to refresh request signal. This is
required for legacy purpose as some codes like vgabios utilizes
counter 1 to provide delay functionality.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
This cleans up i8254 and i8259 codes to fix several cosmetic
issues, like coding convention and some comments improvement.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
PARANOID_IRQ_TRIGGERS is not referenced anywhere in U-Boot.
Remove these dead codes wrapped by it.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
CONFIG_SYS_NUM_IRQS is actually not something we can configure,
but an architecture defined number of ISA IRQs. Move it from
x86-common.h to asm/interrupt.h and rename it to SYS_NUM_IRQS.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
fsp_init() call has a parameter nvs_buf which is used by FSP as the
MRC cache but currently is blindly set to NULL. Retreive the MRC
cache from SPI flash and pass it to fsp_init() call. After the call,
save FSP produced MRC cache to SPI flash too.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Currently struct fmap_entry is used to describe a mrc region.
However this structure contains some other fields that are not
related to mrc cache and causes confusion. Besides, it does not
include a base address field to store SPI flash's base address.
Instead in the mrccache.c it tries to use CONFIG_ROM_SIZE to
calculate the SPI flash base address, which unfortunately is
not 100% correct as CONFIG_ROM_SIZE may not match the whole
SPI flash size.
Define a new struct mrc_region and use it instead.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
This adds mrccache_reserve(), mrccache_get_region() and
mrccache_save() APIs to the mrccache codes. They are ported
from the ivybridge implementation, but with some changes.
For example, in the mrccache_reserve(), ivybridge version
only reserves the pure MRC data, which causes additional
malloc() when saving the cache as the save API needs some
meta data. Now we change it to save the whole MRC date plus
the meta data to elinimate the need for the malloc() later.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Fix some nits, improve some comments and reorder some codes
a little bit.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
For the cache record to write in mrccache_update(), we should
perform a sanity test to see if it is a valid one.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
mrccache implementation can be common for all boards. Move it
from ivybridge cpu directory to the common lib directory.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
It would be helpful to have a command to show FSP header. So far
it only supports FSP header which conforms to FSP spec 1.0.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Introduce a new fsp command and make the existing hob command a
sub-command to fsp for future extension. Also move cmd_hob.c to
the dedicated fsp sub-directory in arch/x86/lib.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
When examining a HOB, it's useful to see which GUID this HOB
belongs to. Add GUID output in the hob command to aid this.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Compact hob command output, especially by making hob type string a
little bit shorter so that we can leave room for future extension.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
System Management BIOS (SMBIOS) is a specification for how
motherboard and system vendors present management information
about their products in a standard format by extending the BIOS
interface on Intel architecture systems. As of today the latest
spec is 3.0 and can be downloaded from DMTF website. This commit
adds a simple and minimum required implementation.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
install_e820_map() has nothing to do with zimage related codes.
Move it to a dedicated place.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
The logic to calculate the number of E820 table entries is wrong
when walking through the FSP HOB tables. Fix it.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Jian Luo <jian.luo4@boschrexroth.de>
Acked-by: Simon Glass <sjg@chromium.org>
Implement write_acpi_table() to create a minimal working ACPI table.
This includes writing FACS, XSDT, RSDP, FADT, MCFG, MADT, DSDT & SSDT
ACPI table entries.
Use a Kconfig option GENERATE_ACPI_TABLE to tell U-Boot whether we need
actually write the APCI table just like we did for PIRQ routing, MP table
and SFI tables. With ACPI table existence, linux kernel gets control of
power management, thermal management, configuration management and
monitoring in hardware.
Signed-off-by: Saket Sinha <saket.sinha89@gmail.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tidied up whitespace and aligned some tabs:
Signed-off-by: Simon Glass <sjg@chromium.org>
Add some comments in start.S for the fact that with FSP U-Boot
actually enters the code twice. Also change to use fsp_init()
and fsp_continue for accuracy.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
After fsp_init() returns, the stack has already been switched to a
place within system memory as defined by CONFIG_FSP_TEMP_RAM_ADDR.
Enlarge the size of malloc() pool before relocation since we have
plenty of memory now.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Now that we have generic routine to calculate relocation address,
remove the x86 specific one which is now only used by coreboot.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Now that we have an efi.h header we can use that for FSP error defines.
Drop the FSP ones.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
This code may be useful for boards that use driver model for PCI.
Note: It would be better to have driver model automatically call this
function somehow. However for now it is probably safer to have it under
board control.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
When using different release version of Intel FSP, the VPD_IMAGE_REV
is different (ie: BayTrail Gold 3 is 0x0303 while Gold 4 is 0x0304).
Remove the asserting of this so that U-Boot does not hang in a debug
build.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
When U-Boot is running from EFI some of the x86 init is replaced with
EFI-specific init. For example, since DRAM has already been set up, we only
need to find it, not init it. Add these functions so that boards can easily
allow booting from EFI if required.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
When U-Boot runs as an EFI payload it needs to avoid setting up the CPU
again. Also U-Boot currently does not handle interrupts for many devices, so
run with interrupts disabled.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
The EFI stub provides information to U-Boot in a table. This includes the
memory map which is needed to decide where to relocate U-Boot. Collect this
information in the early init code and store it in global_data.
Fix up the BIST code at the same time since we don't have it when booting
from EFI and can assume it is 0.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Most EFI implementations use 64-bit. Add a way to build U-Boot as a 64-bit
EFI payload. The payload unpacks a (32-bit) U-Boot and starts it. This can
be enabled for x86 boards at present.
Signed-off-by: Simon Glass <sjg@chromium.org>
Improvements to how the payload is built:
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Add support for building a 32/64-bit EFI stub for x86. This involves
building the startup and relocation code for either i386 or x86_64.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Add a linker script and relocation code for building 64-bit EFI
applications. This can be used for the EFI stub.
Signed-off-by: Simon Glass <sjg@chromium.org>
Improvements to how the payload is built:
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
This code currently requires CONFIG_SYS_TEXT_BASE but this should be
unnecessary. As a first step, remove the build-time limitation and report an
error instead.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Add the required x86 glue code. This includes the initial start-up,
relocation and jumping to efi_main(). We also need to avoid fiddling with
interrupts.
Signed-off-by: Ben Stoltz <stoltz@google.com>
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
When running as an EFI application we must skip relocation. Add support for
this in the x86 relocation code.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
This is now handled by generic U-Boot code so we do not need an x86 version.
It is no-longer called, so remove it.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
When running SMP configuration on QEMU (tcg mode, no kvm), there is
a busy loop in start_aps(), calling udelay(), that waits for APs to
show up online. However, there is a chance that VCPU1 will be timeout
waiting, IOW the secondary VCPUs haven't started their execution yet.
This patch adds a 'pause' instruction in __udelay() only for QEMU
target, to give other VCPUs a chance to run. When QEMU sees the
'pause' instruction, it will yeild the execution to other CPUs.
Signed-off-by: Miao Yan <yanmiaobest@gmail.com>
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
We should mark PCIe ECAM address range in the E820 table as reserved
otherwise kernel will not attempt to use ECAM.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
On some platforms the I/O APIC interrupt pin#0-15 may be connected
to platform pci devices' interrupt pin. In such cases the legacy ISA
IRQ is not available so we should not write ISA interrupt entry if
it is already occupied.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Currently during writing MP table I/O interrupt assignment entry, we
assume the PIRQ is directly mapped to I/O APIC INTPIN#16-23, which
however is not always the case on some platforms.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
We need walk through all functions within a PCI device and assign
their IRQs accordingly.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Store VESA parameters to Linux setup header so that vesafb driver
in the kernel could work.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Jian Luo <jian.luo4@boschrexroth.de>
Print the meaningful base address and mask of an MTRR range without showing
the memory type encoding or valid bit.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
We should allow pci config read/write to host bridge (b.d.f = 0.0.0)
in the int1a_handler() which is a valid pci device.
Signed-off-by: Jian Luo <jian.luo4@boschrexroth.de>
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
PCI option rom may use different SS during its execution, so it is not
safe to assume esp pointed to the same location in the protected mode.
Signed-off-by: Jian Luo <jian.luo4@boschrexroth.de>
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Implement write_mp_table() to create a minimal working MP table.
This includes an MP floating table, a configuration table header
and all of the 5 base configuration table entries. The I/O interrupt
assignment table entry is created based on the same information used
in the creation of PIRQ routing table from device tree. A check
duplicated entry logic is applied to prevent writing multiple I/O
interrupt entries with the same information.
Use a Kconfig option GENERATE_MP_TABLE to tell U-Boot whether we
need actually write the MP table at the F seg, just like we did for
PIRQ routing and SFI tables. With MP table existence, linux kernel
will switch to I/O APIC and local APIC to process all the peripheral
interrupts instead of 8259 PICs. This takes full advantage of the
multicore hardware and the SMP kernel.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
The MP table provides a way for the operating system to support
for symmetric multiprocessing as well as symmetric I/O interrupt
handling with the local APIC and I/O APIC. We provide a bunch of
APIs for U-Boot to write the floating table, configuration table
header as well as base and extended table entries.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
The call to FspInitEntry is done in arch/x86/lib/fsp/fsp_car.S so far.
It worked pretty well but looks not that good. Apart from doing too
much work than just enabling CAR, it cannot read the configuration
data from device tree at that time. Now we want to move it a little
bit later as part of init_sequence_f[] being called by board_init_f().
This way it looks and works better in the U-Boot initialization path.
Due to FSP's design, after calling FspInitEntry it will not return to
its caller, instead it jumps to a continuation function which is given
by bootloader with a new stack in system memory. The original stack in
the CAR is gone, but its content is perserved by FSP and described by
a bootloader temporary memory HOB. Technically we can recover anything
we had before in the previous stack, but that is way too complicated.
To make life much easier, in the FSP continuation routine we just
simply call fsp_init_done() and jump back to car_init_ret() to redo
the whole board_init_f() initialization, but this time with a non-zero
HOB list pointer saved in U-Boot's global data so that we can bypass
the FspInitEntry for the second time.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Andrew Bradford <andrew.bradford@kodakalaris.com>
Tested-by: Simon Glass <sjg@chromium.org>
Currently the FSP execution environment GDT is setup by U-Boot in
arch/x86/cpu/start16.S, which works pretty well. But if we try to
move the FspInitEntry call a little bit later to better fit into
U-Boot's initialization sequence, FSP will fail to bring up the AP
due to #GP fault as AP's GDT is duplicated from BSP whose GDT is
now moved into CAR, and unfortunately FSP calls AP initialization
after it disables the CAR. So basically the BSP's GDT still refers
to the one in the CAR, whose content is no longer available, so
when AP starts up and loads its segment register, it blows up.
To resolve this, we load GDT before calling into FspInitEntry.
The GDT is the same one used in arch/x86/cpu/start16.S, which is
in the ROM and exists forever.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Andrew Bradford <andrew.bradford@kodakalaris.com>
Tested-by: Simon Glass <sjg@chromium.org>
Acked-by: Simon Glass <sjg@chromium.org>
High mem starts at 4 GiB.
Signed-off-by: Andrew Bradford <andrew.bradford@kodakalaris.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
This provides a way of passing information to Linux without requiring the
full ACPI horror. Provide a rudimentary implementation sufficient to be
recognised and parsed by Linux.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Now that reset_cpu() functions correctly, use it instead of directly
accessing the port on boards that use a Firmware Support Package (FSP).
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Previously the PIRQ routing table sanity check was performed against
the original table provided by the platform codes. Now we switch to
check its sanity on the final table in the F segment as this one is
the one seen by the OS.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
On x86 boards, platform chipset receives up to four different
interrupt signals from PCI devices (INTA/B/C/D), which in turn
will be routed to chipset internal PIRQ lines then routed to
8259 PIC finally if configuring the whole system to work under
the so-called PIC mode (in contrast to symmetric IO mode which
uses IOAPIC).
We add two major APIs to aid this, one for routing PIRQ and the
other one for generating a PIRQ routing table.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
We can write the configuration table in last_stage_init() for all x86
boards, but not with coreboot since coreboot already has them.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Create a default e820 table with 3 entries which is enough to boot
a Linux kernel.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
On x86 systems this device is commonly used to provide legacy port access.
It is sort-of a replacement for the old ISA bus.
Add a uclass for this, and allow it to have child devices.
Signed-off-by: Simon Glass <sjg@chromium.org>
Add a simple uclass for this chip which is often found in x86 systems
where the CPU is a separate device.
The device can have children, so make it scan the device tree for these.
Signed-off-by: Simon Glass <sjg@chromium.org>
Convert this driver over to use driver model. Since all x86 platforms use
it, move x86 to use driver model for SPI and SPI flash. Adjust all dependent
code and remove the old x86 spi_init() function.
Note that this does not make full use of the new PCI uclass as yet. We still
scan the bus looking for the device. It should move to finding its details
in the device tree.
Signed-off-by: Simon Glass <sjg@chromium.org>