This code is pretty old and we want to support only 32-bit systems now.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Graeme Russ <graeme.russ@gmail.com>
Invert the polarity of this option to simplify the Makefile logic.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Gabe Black <gabeblack@chromium.org>
This x86 CPU variant is no longer required as the boards that use it have
been removed.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Graeme Russ <graeme.russ@gmail.com>
The libfdt.h file is the definition file for libfdt. It is unnecessary
to include other fdt header files (the necessary ones are pulled in
by libfdt.h).
Signed-off-by: Gerald Van Baren <gvb@unssw.com>
Acked-by: Simon Glass <sjg@chromium.org>
Acked-by: Stefan Roese <sr@denx.de>
and, if including libfdt.h which includes libfdt_env.h in
the correct order, don't include fdt.h before libfdt.h.
this is needed to get the fdt type definitions set from
the project environment before fdt.h uses them.
Signed-off-by: Kim Phillips <kim.phillips@freescale.com>
Cc: Jerry Van Baren <gvb.uboot@gmail.com>
Move this field into arch_global_data and tidy up.
Signed-off-by: Simon Glass <sjg@chromium.org>
[trini: Add arch/x86/cpu/cpu.c changes after Graeme's comments]
Signed-off-by: Tom Rini <trini@ti.com>
We currently assume that the global data pointer is at the start of
struct global_data. We want to remove this restriction, and it is
easiest to do this in C.
Remove the asm code and add equivalent code in C.
This idea was proposed by Graeme Russ here:
http://patchwork.ozlabs.org/patch/199741/
Signed-off-by: Simon Glass <sjg@chromium.org>
[trini: Apply Graeme Russ' comments
http://patchwork.ozlabs.org/patch/206305/ here, re-order]
Signed-off-by: Tom Rini <trini@ti.com>
We plan to move architecture-specific data into a separate structure so
that we can make the rest of it common.
As a first step, create struct arch_global_data to hold these fields.
Initially it is empty.
This patch applies to all archs at once. I can split it if this is really
a pain.
Signed-off-by: Simon Glass <sjg@chromium.org>
This function is not intended to be exported from the video drivers, so
remove the prototype. This fixes an error:
cfb_console.c:1793:12: error: static declaration of 'video_init' follows non-static declaration
Signed-off-by: Simon Glass <sjg@chromium.org>
The function setup_pcat_compatibility() is weak and implemented as empty
function in board.c hence we don't have to override that with another
empty function.
monitor_flash_len is unused, drop it.
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
... because that information is already "encoded" in the directory name.
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Some systems (like Google Link device) provide the ability to keep a
history of the target CPU port80 accesses, which is extremely handy
for debugging. The problem is that the EC handling port 80 access is
orders of magnitude slower than the AP. This causes random loss of
trace data.
This change allows to throttle port 80 accesses such that in case the
AP is trying to post faster than the EC can handle, a delay is
introduced to make sure that the post rate is throttled. Experiments
have shown that on Link the delay should be at least 350,000 of tsc
clocks.
Throttling is not being enabled by default: to enable it one would
have to set MIN_PORT80_KCLOCKS_DELAY to something like 400 and rebuild
the u-boot image. With upcoming EC code optimizations this number
could be decreased (new new value should be established
experimentally).
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Some u-boot modules rely on availability of get_ticks() and
get_tbclk() functions, reporting a free running clock and its
frequency respectively. Traditionally these functions return number
and frequency of timer interrupts.
Intel's core architecture processors however are known to run the
rdtsc instruction at a constant rate of the so called 'Max Non Turbo
ratio' times the external clock frequency which is 100MHz. This is
just as good for the timer tick functions in question.
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This will write magic value to APMC command port which
will trigger an SMI and cause coreboot to lock down
the ME, chipset, and CPU.
Signed-off-by: Duncan Laurie <dlaurie@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Coreboot was always using MTRR 7 for the write-protect
cache entry that covers the ROM and U-boot was removing it.
However with 4GB configs we need more MTRRs for the BIOS
and so the WP MTRR needs to move. Instead coreboot will
always use the last available MTRR that is normally set
aside for OS use and U-boot can clear it before the OS.
Signed-off-by: Duncan Laurie <dlaurie@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This helps us monitor boot progress and determine where U-Boot dies if
there are any problems.
Signed-off-by: Stefan Reinauer <reinauer@google.com>
Signed-off-by: Simon Glass <sjg@chromium.org>
This contains just the minimum information for a coreboot-based board.
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Allow a device tree to be provided through the standard mechanisms.
Signed-off-by: Gabe Black <gabeblack@google.com>
Signed-off-by: Simon Glass <sjg@chromium.org>
This option delays loading of the environment until later, so that only the
default environment will be available to U-Boot.
This can address the security risk of untrusted data being used during boot.
When CONFIG_DELAY_ENVIRONMENT is defined, it is convenient to have a
run-time way of enabling loadinlg of the environment. Add this to the
fdt as /config/delay-environment.
Note: This patch depends on http://patchwork.ozlabs.org/patch/194342/
Signed-off-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
These were removed, but actually are useful.
Cold means that we started from a reset/power on.
Warm means that we started from another U-Boot.
We determine whether u-boot on x86 was warm or cold booted (really if
it started at the beginning of the text segment or at the ELF entry point).
We plumb the result through to the global data structure.
Signed-off-by: Simon Glass <sjg@chromium.org>
Because calculate_relocation_address now uses the e820 map, it will be able
to avoid addresses over 32 bits and regions that are at high addresses but
not big enough for U-Boot. It also means we can remove the hack which
limitted U-Boot's idea of the size of memory to less than 4GB.
Also take into account the space needed for the heap and stack, so we avoid
picking a very small region those areas might overlap with something it
shouldn't.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Different systems may have different mechanisms for picking a suitable place
to relocate U-Boot to.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This changes the layout in decreasing addresses from:
1. Stack
2. Sections in the image
3. Heap
to
1. Sections in the image
2. Heap
3. Stack
This allows the stack to grow significantly more since it isn't constrained by
the other u-boot areas. More importantly, the generic memory wipe code assumes
that the stack is the lowest addressed area used by the main part of u-boot.
In the original layout, that means that u-boot tramples all over itself. In
the new layout, it works.
Signed-off-by: Gabe Black <gabeblack@google.com>
Signed-off-by: Simon Glass <sjg@chromium.org>
If we have SPI support, make sure that we init it.
Signed-off-by: Gabe Black <gabeblack@google.com>
Signed-off-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Vic Yang <victoryang@chromium.org>
Implement arch_phys_memset so that it can set memory at physical addresses
above 4GB using PAE paging. Because there are only 5 page tables in PAE mode,
1 PDPT and 4 PDTs, those tables are statically allocated in the BSS. The
tables must be 4K page aligned and are declared that way, and because U-Boot
starts as 4K aligned and the relocation code relocates it to a 4K aligned
address, the tables work as intended.
While paging is turned on, all 4GB are identity mapped except for one 2MB
page which is used as the window into high memory. This way, U-Boot will
continue to work as expected when running code that expects to access memory
freely, but the code can still get at high memory through its window.
The window is put at 2MB so that it's 2MB page aligned, low in memory to be
out of the way of things U-Boot is likely to care about, and above the lowest
1MB where lots of random things live.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
These types should be 64 bits long to reflect the fact that physical
addresses and the size of physical areas of memory are more than 32 bits
long.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
The use of post-increment with a do-while loop results in
the loop going one step too far when handling relocation fixups.
In about 1/100 cases this would cause it to hang.
Signed-off-by: Duncan Laurie <dlaurie@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
U-boot is unable to actually use that memory and it can
cause problems with relocation if it tries to.
Signed-off-by: Duncan Laurie <dlaurie@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This fixes the following warning:
zimage.c:312: Warning: indirect jmp without `*'
Also fixed these warnings to keep checkpatch quiet:
warning: arch/x86/lib/zimage.c,311: unnecessary whitespace before a quoted newline
warning: arch/x86/lib/zimage.c,312: unnecessary whitespace before a quoted newline
warning: arch/x86/lib/zimage.c,313: unnecessary whitespace before a quoted newline
Signed-off-by: Simon Glass <sjg@chromium.org>
This cleans up the rom caching optimization implemented in coreboot (and
needed throughout U-Boot runtime).
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This function can be used by boards which want to do some clean-up
before booting a zImage.
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This way when that dram "banks" are displayed, there's some useful information
there. The number of "banks" we claim to have needs to be adjusted so that it
covers the number of RAM e820 regions we expect to have/care about.
This needs to be done after "RAM" initialization even though we always run
from RAM. The bd pointer in the global data structure doesn't automatically
point to anything, and it isn't set up until "RAM" is available since, I
assume, it would take too much space in the very constrained pre-RAM
environment.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
To maintain the initialization state of the timestamp facility, thesq
pointer to the CBMEM section containing the timestamp table should be
kept in the .data section (so that it is maintained across u-boot
relocation).
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This change turns on the code which allows u-boot to add
timestamps to the timestamp table created by coreboot.
Since u-boot does not use the tsc_t like structure to represent
HW counter readings, this structure is being replaced by 64 bit
integer.
The timestamp_init() function is now initializing the base timer
value used by u-boot to calculate the HW counter increments.
Timestamp facility is initialized as soon as the timestamp table
pointer is found in the coreboot table. The u-boot generated
timer events' ID will start at 1000 to clearly separate u-boot
events from coreboot events in the timer trace.
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Put this function in the u-boot-x86.h header file. We could instead create
timer.h perhaps.
We support setting a base time, and reading the time relative to this base.
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Implement <asm-generic/gpio.h> functions for Intel ICH6 and later.
Only GPIOs 0-31 are handled by this code.
Signed-off-by: Bill Richardson <wfrichar@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This patch prevents u-boot from "spamming" random progress codes on
a port 80 "post card".
The previous version of this patch just removed the delays in the "slow"
IO functions, as they do not need to be slow, however, this patch is
less intrusive.
It uses another unused port that is often used by BIOSes (and the Linux
Kernel) for small delay timing purposes.
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
The i386 version of io.h depends on the phys_addr_t type which is defined in
types.h. It wasn't including that explicitly, and was working presumably
because the other files including it had already included types.h themselves
directly or indirectly. This change fixes that.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
The microsecond timer is not currently implemented, but add a dummy
implementation for now.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This function provides an opportunity for some last minute cleanup and
reconfiguration before control is handed over to Linux. It's possible this
may need to do something in the future, but for now it's left empty. It's set
up as a weak symbol so it can be overridden if necessary on a case by case
basis.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This change adds a pointer to the global data structure in x86 to point to
the device tree. This mirrors an identical pointer in ARM.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
We don't want this for coreboot, so provide a way of compiling it out.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Add support for decoding tags for GPIOs, compile/build info, cbmem and
other features.
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
sysinfo.c only contains the lib_sysinfo data structure which
is used/filled by tables.c. This split was introduced by importing
code from libpayload originally, but to keep the code simple, add
the single line of actual code to tables.c
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
A hook is installed to configure PCI bus bridges as they encountered by u-boot.
The hook extracts the secondary bus number from the bridge's config space and
then recursively scans that bus.
On Coreboot, the PCI bus address space has identity mapping with the
physical address space, so declare it as such to ensure that the "pci_map_bar"
function used by some PCI drivers is behaving properly. This fixes the
EHCI PCI driver initialization on Stumpy.
This was tested as follows:
Ran the PCI command on Alex, saw devices on bus 0, the OXPCIe 952 on
bus 1, and empty busses 2 through 5. This matches the bridges
reported on bus 0 and the PCI configuration output from coreboot.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Vincent Palatin <vpalatin@chromium.org>
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
U-boot needs a host controller or "hose" to interact with the PCI busses
behind them. This change installs a host controller during initialization of
the coreboot "board" which implements some of X86's basic PCI semantics. This
relies on some existing generic code, but also duplicates a little bit of code
from the sc520 implementation. Ideally we'd eliminate that duplication at some
point.
It looks like in order to scan buses beyond bus 0, we'll need to tell u-boot's
generic PCI configuration code what to do if it encounters a bridge,
specifically to scan the bus on the other side of it.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Graeme Russ <graeme.russ@gmail.com>
coreboot.c and coreboot_pci.c don't contain board specific but only
coreboot specific code. Hence move it to the coreboot directory in
arch/x86/cpu (which should probably be moved out of cpu/ in another
commit)
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
I suspect these includes were usually available because something else
included them earlier or because they were brought in transitively.
Change-Id: I6aae2ac94dc792eac6febb4345e8125f69f70988
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
When running from coreboot we don't want this code.
This version works by ifdef-ing out all of the code that would go
into those sections and all the code that refers to it. The sections are
then empty, and the linker will either leave them empty for the loader
to ignore or remove them entirely.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Putting global data on the stack simplifies the init process (and makes it
slightly quicker). During the 'flash' stage of the init sequence, global
data is in the CAR stack. After SDRAM is initialised, global data is copied
from CAR to the SDRAM stack
Signed-off-by: Graeme Russ <graeme.russ@gmail.com>
Signed-off-by: Simon Glass <sjg@chromium.org>
So it can be used as a type in struct global_data and remove an ugly typecast
Signed-off-by: Graeme Russ <graeme.russ@gmail.com>
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Marek Vasut <marex@denx.de>
The command declaration now uses the new LG-array method to generate
list of commands. Thus the __u_boot_cmd section is now superseded and
redundant and therefore can be removed. Also, remove externed symbols
associated with this section from include/command.h .
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Joe Hershberger <joe.hershberger@gmail.com>
Cc: Mike Frysinger <vapier@gentoo.org>
Add section for the linker-generated lists into all possible linker
files, so that everyone can easily use these lists. This is mostly
a mechanical adjustment.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Joe Hershberger <joe.hershberger@gmail.com>
Cc: Mike Frysinger <vapier@gentoo.org>
Remove the parts depending either on disabled CONFIG_SERIAL_MULTI
or ifdefs around CONFIG_SERIAL_MULTI parts since CONFIG_SERIAL_MULTI
is now enabled by default.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Marek Vasut <marek.vasut@gmail.com>
Cc: Tom Rini <trini@ti.com>
Cc: Anatolij Gustschin <agust@denx.de>
Cc: Stefan Roese <sr@denx.de>
All the global flag defines are the same across all arches. So unify them
in one place, and add a simple way for arches to extend for their needs.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
All arches init this the same way, so move the logic into the core
net code to avoid duplicating it everywhere else.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
This field gets read in one place (by "bdinfo"), and we can replace
that with getenv("ipaddr"). After all, the bi_ip_addr field is kept
up-to-date implicitly with the value of the ipaddr env var.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
Building the eNET_SRAM board fails for me:
sc520_timer.c: In function 'sc520_udelay':
sc520_timer.c:81:7: error: variable 'temp' set but not used
[-Werror=unused-but-set-variable]
cc1: all warnings being treated as errors
make[1]: *** [sc520_timer.o] Error 1
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Create an init function array for board_init_f_r - This finalises the
migration to a purely array based initialisation mechanism
Also tweak a few comments while we are at it so everything is 'correct'
--
Changes for v2:
- Renamed to a more apt name
- Fix bug in set_reloc_flag_r
- Re-instate gd->flags = boot_flags; in board_init_f
- Added commit message
This patch moves towards reducing board.c to simply a set of init cores for
the three initialisation phases (Flash, Flash/RAM, and RAM), a set of three
init function arrays and a init function array processing function
Use the base address of the 'F' segment as a pointer to the global data
structure. By adding the linear address (i.e. the 'D' segment address) as
the first word of the global data structure, the address of the global data
relative to the 'D' segment can be found simply, for example, by:
fs movl 0, %eax
This makes the gd 'pointer' writable prior to relocation (by reloading the
Global Desctriptor Table) which brings x86 into line with all other arches
NOTE: Writing to the gd 'pointer' is expensive (but we only do it
twice) but using it to access global data members (read and write) is
still fairly cheap
--
Changes for v2:
- Rebased against changes made to patch #3
- Removed extra indent
- Tweaked commit message
This commit introduces no functional changes - It simply re-arranges the
calculations so that adding to them in future commits will be cleaner
--
Changes for v2:
- Fixed typo in title
- Added commit message
Move the relocation offset calculation out of assembler and into C. This
also paves the way for the upcoming init sequence simplification by adding
the board_init_f_r flash to RAM transitional function
--
Changes for v2:
- Added commit message
- Minor adjustment to new stack address comment
The inline assembler is ugly and uses hard coded magic numbers. Make it more
elegant to allow cleaner implementation of future GDT related patches. The
compiler seems smart enough to generate the same code anyway
--
Changes for v2:
- Rebased against revised patch #3
- Use GDT size define instead of magic number
- Added commit message
The implementations of memcpy and memset are now the optimised versions
from glibc, so use them instead of simple copy loops
--
Changes for v2:
- Removed unneeded brackets
This change finishes plumbing the initrd support built into the zboot
mechanism out to the command interface.
It also fixes a bug in the command declaration where the kernel size could
be passed as an optional second parameter but not enough arguments were
allowed.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
If vboot successfully verifies a kernel, it will leave it in place and
basically ready to boot. The zeropage table which is part of the x86 boot
protocol is at the end of the kernel, though, instead of the beginning, and
because the image is already in place there's no need to copy it around.
This change refactors the code which implements the zboot command so that
the configuration of the zeropage table and loading the pieces of the
kernel into memory are done separately. Also, because the command line goes
before the zeropage table in vboot which is somewhat incompatible with the
normal protocol, where to put the command line is a now a parameter instead
of being hard coded.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Also approximate the size of RAM using the largest RAM address available
in the tables. There may be areas which are marked as reserved which are
actually at the end of RAM.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
This change conditionally modifies the zboot command so that it can use the
32 bit boot protocol. This is necessary because the 16 bit realmode entry
point assumes that it can call BIOS services which neither coreboot nor
u-boot provide.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
This change cleans up some formatting issues in the zimage handling code, and
converts it from using offsets added to a base pointer to using the available
structure definitions which were already being included.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
This change also forces the lib_sysinfo structure to be in the .data
section. Otherwise it ends up in the .bss section. U-boot assumes that it
doesn't need to copy it over during relocation, and instead fills that
whole section with zeroes. If we really were booting from ROM that would be
appropriate, but we need some information from the coreboot tables (memory
size) before then and have to fill that structure before relocation. We
skirt u-boot's assumption by putting this in .data where it assumes there
is still read only but non-zero data.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Add a target for running u-boot as a coreboot payload in boards.cfg, a
board, CPU and a config. This is a skeleton implementation which always
reports the size of memory as 64 MB.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
When gcc compiles some 64 bit operations on a 32 bit machine, it generates
calls to small functions instead of instructions which do the job directly.
Those functions are defined in libgcc and transparently provide whatever
functionality was necessary. Unfortunately, u-boot can be built with a
non-standard ABI when libgcc isn't. More specifically, u-boot uses
-mregparm. When the u-boot and libgcc are linked together, very confusing
bugs can crop up, for instance seemingly normal integer division or modulus
getting the wrong answer or even raising a spurious divide by zero
exception.
This change borrows (steals) a technique and some code from coreboot which
solves this problem by creating wrappers which translate the calling
convention when calling the functions in libgcc. Unfortunately that means
that these instructions which had already been turned into functions have
even more overhead, but more importantly it makes them work properly.
To find all of the functions that needed wrapping, u-boot was compiled
without linking in libgcc. All the symbols the linker complained were
undefined were presumed to be the symbols that are needed from libgcc.
These were a subset of the symbols covered by the coreboot code, so it was
used unmodified.
To prevent symbols which are provided by libgcc but not currently wrapped
(or even known about) from being silently linked against by code generated
by libgcc, a new copy of libgcc is created where all the symbols are
prefixed with __normal_. Without being purposefully wrapped, these symbols
will cause linker errors instead of silently introducing very subtle,
confusing bugs.
Another approach would be to whitelist symbols from libgcc and strip out
all the others. The problem with this approach is that it requires the
white listed symbols to be specified three times, once for objcopy, once so
the linker inserts the wrapped, and once to generate the wrapper itself,
while this implementation needs it to be listed only twice. There isn't
much tangible difference in what each approach produces, so this one was
preferred.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
The new implementation is about twice as fast as the old. This is from
glibc-2.14, sysdeps/i386/memset.c.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
This change adds an upper bound for symbols which are fixed up after u-boot
is relocated into RAM. This way portions that are left at their original
location can be referred to without having to manually fix up any pointers.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
There are two blobs embedded into the u-boot image which are linked to run
at an address which is different from where they actually end up in the
ROM, one called "realmode" and one called "bios". There are realmode_setup
and bios_setup functions which prepare those blobs by copying them into the
location they're supposed to run from, among other things.
During u-boot relocation from ROM to RAM, the text and a few data segments
are copied over. The realmode and bios sections are not copied, and so the
only place they can be read from is their original location in the ROM.
Looking specifically at the bios blob, there are symbols defined in the
linker script called __bios_start and __bios_size which are defined to be
the start and size of the blob in the ROM.
In the bios_setup function, there seem to be two mistakes happening. First,
the offset from ROM to RAM is being added to __bios_start which implies that
this code expects to use the copy moved to RAM. No such copy is made, so
that's wrong. More subtly, when u-boot relocates itself, it goes through
all of the relocations stored in .rel.dyn and fixes them up. This has the
effect of transforming the __bios_start reference in bios_setup so that it
refers to the version in RAM (if one existed) instead of the one in ROM. To
correct for that, the offset actually needs to be subtracted out again to
translate the address back into the ROM.
The net effect is that for both blobs, a + needs to be changed to a -.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
