Add a __always_inline macro that wraps __attribute__((always_inline))
and replace current users with the macro, excluding files under
src/vendorcode.
Change-Id: Ic57e474c1d2ca7cc0405ac677869f78a28d3e529
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: https://review.coreboot.org/28587
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Julius Werner <jwerner@google.com>
Support for more situations: floating point, compressed instructions,
etc. Add support for redirect exception to S-Mode.
Change-Id: I9983d56245eab1d458a84cb1432aeb805df7a49f
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/27972
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Add a interface, which is implemented by SoC.
Change-Id: I5524732f6eb3841e43afd176644119b03b5e5e27
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/28372
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
RAMSTAGE will revoke CAR/scratchpad, so stack and exception handling
needs to be moved to ddr memory. So add a assembly file to do this.
Change-Id: I58aa6ff911f385180bad6e026d3c3eace846e37d
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/28384
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Highest two bits of misa can be used to check machine length. Add code
to support this.
Change-Id: I3bab301d38ea8aabf2c70437e179287814298b25
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/27770
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Must to set MXR, when needs to read the page which is execution-only.
So make this change.
Change-Id: I19519782fe791982a8fbd48ef33b5a92a3c48bfc
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/28394
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
BOOTBLOCK/ROMSTAGE run in CAR/scratchpad. When RAMSTAGE begins
execution will enable cache, then CAR will disappear. So the
Stack will be separated.
Change-Id: I37a0c1928052cabf61ba5c25b440363b75726782
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/28383
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Julius Werner <jwerner@chromium.org>
These RISC-V ABIs defined by GCC : ilp32 ilp32d ilp32f lp64 lp64d lp64f.
Through this we know that the length of the long's bit is equal to pointer.
So update this code. This's more flexible.
Change-Id: I16e1a2c12c6034df75dc360b65acb1b6affec49b
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/27768
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
cache_sync_instructions() has been superseded by
arch_program_segment_loaded() and friends for a while. There are no uses
in common code anymore, so let's remove it from <arch/cache.h> for all
architectures.
arm64 still has an implementation and one reference, but they are not
really needed since arch_program_segment_loaded() does the same thing
already. Remove them.
Leave it in arm(32) since there are several references (including in SoC
code) that I don't feel like tracking down and testing right now.
Change-Id: I6b776ad49782d981d6f1ef0a0e013812cf408524
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://review.coreboot.org/27879
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
src/arch/riscv/stages.c is an entry of romstage/ramstage, and does not
needs to be bootblock.
src/arch/riscv/id.S src/arch/riscv/id.ld is used to generate some
compile/board/time information, which is repeated with src/lib/version.c
Change-Id: Ic736b378e24df387584c5f86a2b04078fc55723d
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/27557
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
When I tried to compile the RISC-V code (202e7d4f3c), I found some errors:
`PRIu64` is undefined
src/arch/riscv/timestamp.c does not exist
Currently RISC-V does not have the implementation and use of timestamp,
so I temporarily delete the code related to timestamp in the Makefile.
And define PRIu64.
Change-Id: I7f1a0793113bce7c1411e39f102cf20dbadda5d6
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/27543
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
This code was copied from x86. It is not needed for RISC-V.
Change-Id: If6c3bfdc4090e45d171e68a28d27c38dabe91687
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/27544
Reviewed-by: Philipp Hug <philipp@hug.cx>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
This was updating flags for ALL architectures, not just riscv.
That was bad, and gave us errors, although they weren't fatal for
some reason:
i386-elf-gcc: error: missing argument to '-mcmodel='
i386-elf-gcc: error: missing argument to '-march='
i386-elf-gcc: error: missing argument to '-mabi='
This issue started from commit 5fed693a (riscv: add support for
modifying compiler options)
Add comments to the other 'endif' statements since they're now
surrounded by a global ifeq
Change-Id: Ifa12ad98b04a5ac36148609ccdf46ca427fc5a27
Signed-off-by: Martin Roth <martin@coreboot.org>
Reviewed-on: https://review.coreboot.org/27535
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Georgi <pgeorgi@google.com>
Add an interface to support cache as ram.
Initialize stack pointer for each hart.
Change-Id: Ic3920e01dd1a7f047a53de57250589000a111409
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/27430
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Each HART of a SoC like fu540 supports a different ISA. In order for the
coreboot's code can run on each core, need to modify the compile options.
So add this code.
Change-Id: Ie33edc175e612846d4a74f3cbf7520d4145cb68b
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/27442
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-by: Philipp Hug <philipp@hug.cx>
Replicate directory layout from x86 for SMP.
Change-Id: I27aee55f24d96ba9e7d8f2e6653f6c9c5e85c66a
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/27355
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Add support to check ISA extension for RISC-V.
Change-Id: I5982fb32ed1dd435059edc6aa0373bffa899e160
Signed-off-by: Xiang Wang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/27410
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-by: Philipp Hug <philipp@hug.cx>
GCC pre-defined some macros for detecting ISA extensions.
We should use these macros to detect ISA features.
Change-Id: I5782cdd1bf64b0161c58d789f46389dccfe44475
Signed-off-by: XiangWang <wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/27300
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
RISC-V doesn't set up page tables anymore, since commit b26759d703
("arch/riscv: Don't set up virtual memory").
Change-Id: Id1e759b63fb0bc88ab256994d3849d16814affa0
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/25701
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
CSRs are XLEN bits wide (i.e. the same width as general purpose
registers), so size_t seems a little more correct than int.
This change doesn't affect functionality because MSTATUS_MPRV already
fits in 31 bits.
Change-Id: I003c1b88b4493681dc9b6178ac785be330203ef5
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/25625
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
RISC-V does not have the kind of I/O space that x86 has. Other
architectures tend to leave out these definitions as well.
Change-Id: I7328dae1f1fa4ef8772750244a0b11a3fa5aa88f
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/25566
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
These exceptions were new in the Privileged Architecture spec 1.10.
We need to delegate them to S-mode.
Change-Id: Iec15afe9656107b9aeea1677c5b8dc7d654fa746
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/23774
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Update encoding.h to the version shipped with spike commit
0185d36 ("Merge pull request #165 from riscv/small_progbuf"),
and copy the license header from the LICENSE file.
Change-Id: I517042e5865986e88a589dc8623745f8d584d6b8
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/23773
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
The RISC-V boot protocol foresees that at every stage boundary (bootrom
to boot loader, boot loader -> OS), register a0 contains the Hart ID and
a1 contains the physical address of the Flattened Device Tree that the
stage shall use.
As a first step, pass the bootrom-provided FDT to the payload,
unmodified.
Change-Id: I468bc64a47153d564087235f1c7e2d10e3d7a658
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/23797
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Due to changes in the RISC-V Privileged Architecture specification,
Linux can now be started in physical memory and it will setup its own
page tables.
Thus we can delete most of virtual_memory.c.
Change-Id: I4e69d15f8ee540d2f98c342bc4ec0c00fb48def0
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/23772
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
In order to support RISC-V processors with and without the RVC
extension, configure the architecture variant (-march=...) explicitly.
NOTE: Spike does support RVC, but currently doesn't select
ARCH_RISCV_COMPRESSED, because coreboot's trap handler doesn't
support RVC.
Change-Id: Id4f69fa6b33604a5aa60fd6f6da8bd966494112f
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/23733
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
The RISC-V Privileged Architecture spec 1.10 requires that the address part of
mtvec is four-byte aligned. The lower two bits encode a "mode" flag and should
be zero for now.
Add the necessary alignment directive before trap_entry.
Change-Id: I83ea23e2c8f984775985ae7d61f80ad75286baaa
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/23173
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
coreboot only maintains a single trap entry, because it only runs in
machine mode.
Change-Id: I7324d9c8897d5c4e9d4784e7bc2a055890eab698
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/22595
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
The code dealing with the old config string isn't needed anymore,
because the config string has been deprecated in favor of
OpenFirmware-derived devicetrees.
Change-Id: I71398fb4861dbaf7eefc6e6f222bb7159798fafa
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/22594
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
This Supervisor Binary Interface, which is based on a page of code
that's provided to operating systems by the M-mode software, has been
superseded by a different (currently not really documented) SBI, which
is based on directly executing ECALLs instructions. Thus some of our
code becomes obsolete. Just rip it out until we implement the new SBI.
Change-Id: Iec9c20b750f39a2b8f1553e25865bbf150605a6d
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/22593
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
There is currently no case where a struct cpu_device_id instance needs
to be modified. Thus, declare all instances as const.
Change-Id: I5ec7460b56d75d255b3451d76a46df76a51d6365
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/22526
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
This is the lazy solution, as explained in the comment, but it works for
now.
Change-Id: I46e18b6d633280d6409e42462500fbe7c6823b4d
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/21689
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Accessing the config string doesn't work anymore on current versions of
spike. Thus return dummy pointers until we have a better solution.
Change-Id: I684fc51dc0916f2235e57e36b913d363e1cb02b1
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/21687
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Our toolchain can compile mret now, and once the encoding changes, we'll
have to adjust the code anyway.
Change-Id: Ic37a849f65195006fa15d74f651a8aa9a9da5b5c
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/21762
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
This fixes a case of mstatus corruption, where GCC generated code that
used the same register for the mprv bit and the result.
GCC inline assembly register modifiers are documented here:
https://gcc.gnu.org/onlinedocs/gcc/Modifiers.html
Change-Id: I2c563d171892c2e22ac96b34663aa3965553ceb3
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/21688
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
In current versions of spike, the config string is not available
anymore, because RISC-V is moving toward OpenFirmware-derived device
trees (either in FDT or text format). Using query_config_string leads to
a crash in these versions of spike.
With this commit and If0bea4bf52d ("riscv: Update register address"),
coreboot reaches the romstage again, on spike.
Change-Id: Ib1e6565145f0b2252deb1f4658221a4f816e2af4
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/21657
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
This is easier to read than the raw shift amount that's extracted from
load/store instructions.
Change-Id: Ia16ab9fbaf55345b654ea65e65267ed900eb29e1
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/21654
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
I triggered a bug, when I try to debug riscv code by spike.
This bug is caused by an instruction exception[csrwi 0x320,7].
This is operate for mcounteren. This address is error. 0x306
is right. scounteren is not need to be set, because S-mode
code controls it.
Change-Id: If0bea4bf52d8ad2fb2598724d6feb59dc1b3084a
Signed-off-by: wxjstz<wxjstz@126.com>
Reviewed-on: https://review.coreboot.org/20043
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
- Update all symbols to use IS_ENABLED()
- Update non-romcc usage to use 'if' instead of '#if' where it
makes sense.
Change-Id: I5a84414d2d1631e35ac91efb67a0d4c1f673bf85
Signed-off-by: Martin Roth <martinroth@google.com>
Reviewed-on: https://review.coreboot.org/20005
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Georgi <pgeorgi@google.com>
The word 'coreboot' should always be written in lowercase, even at the
start of a sentence.
Change-Id: I7945ddb988262e7483da4e623cedf972380e65a2
Signed-off-by: Martin Roth <martinroth@google.com>
Reviewed-on: https://review.coreboot.org/20029
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Philippe Mathieu-Daudé <philippe.mathieu.daude@gmail.com>
Reviewed-by: Patrick Georgi <pgeorgi@google.com>
Caching is a very architecture-specific thing, but most architectures
have a cache in general. Therefore it can be useful to have a generic
architecture-independent API to perform simple cache management tasks
from common code.
We have already standardized on the dcache_clean/invalidate naming
scheme that originally comes from ARM in libpayload, so let's just do
the same for coreboot. Unlike libpayload, there are other things than
just DMA coherency we may want to achieve with those functions, so
actually implement them for real even on architectures with
cache-snooping DMA like x86. (In the future, we may find applications
like this in libpayload as well and should probably rethink the API
there... maybe move the current functionality to a separate
dma_map/unmap API instead. But that's beyond scope of this patch.)
Change-Id: I2c1723a287f76cd4118ef38a445339840601aeea
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://review.coreboot.org/19788
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-by: Philippe Mathieu-Daudé <philippe.mathieu.daude@gmail.com>
Coverity is detecting 'sp' as a variable which has not been initialized.
This is obviously not correct, so this patch *TRIES* to mark it as false
I'm not positive that this will work because the annotation needs to go
on the line above the error, but this error is inside of a # define.
Does the whole #define count as one line? Can it go on the line
above the #define in the .h file? Does it have to precede every line
where the #define is used? The documentation doesn't make this clear.
Should suppress coverity issues: 1368525 & 1368527
uninit_use: Using uninitialized value sp.
Change-Id: Ibae5e206c4ff47991ea8a11b6b59972b24b71796
Signed-off-by: Martin Roth <martinroth@google.com>
Reviewed-on: https://review.coreboot.org/18247
Tested-by: build bot (Jenkins)
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
The new name and location make more sense:
- The instruction used to call into machine mode isn't called "ecall"
anymore; it's mcall now.
- Having SBI_ in the name is slightly wrong, too: these numbers are not
part of the Supervisor Binary Interface, they are just used to
forward SBI calls (they could be renumbered arbitrarily without
breaking an OS that's run under coreboot).
Also remove mcall_dev_{req,resp} and the corresponding mcall numbers,
which are no longer used.
Change-Id: I76a8cb04e4ace51964b1cb4f67d49cfee9850da7
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/18146
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
SBI calls, as it turned out, were never right.
They did not set the stack correctly on traps.
They were not correctly setting the MIP instead of the SIP
(although this was not really well documented).
On Harvey, we were trying to avoid using them,
and due to a bug in SPIKE, our avoidance worked.
Once SPIKE was fixed, our avoidance broke.
This set of changes is tested and working with Harvey
which, for the first time, is making SBI calls.
It's not pretty and we're going to want to rework
trap_util.S in coming days.
Change-Id: Ibef530adcc58d33e2c44ff758e0b7d2acbdc5e99
Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/18097
Tested-by: build bot (Jenkins)
The user and supervisor counters could not be safely enabled
before as the register numbers were not finalized. Now that
everyone agrees, we can enable them. Until we are sure the
toolchains are caught up, we use the hardcode name with
the register names in comments. As soon as toolchains
settle down we'll do one more pass and convert to
the symbolic names.
Tested on lowrisc bitstream and SPIKE simulator.
Change-Id: I21fe5cac44fafe4b7806e004c179aa27541be4b6
Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-on: https://review.coreboot.org/17920
Tested-by: build bot (Jenkins)
Reviewed-by: Alex Bradbury <asb@lowrisc.org>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-by: Andrew Waterman <aswaterman@gmail.com>
RISCV requires that timer interrupts be handled in machine
mode and delegated as necessary. Also you can only reset the
timer interrupt by writing to mtimecmp. Further, you must
write a number > mtime, not just != mtime. This rather clumsy
situation requires that we write some value into the future
into mtimecmp lest we never be able to leave machine mode as
the interrupt either is not cleared or instantly reoccurs.
This current code is tested and works for harvey (Plan 9)
timer interrupts.
Change-Id: I8538d5fd8d80d9347773c638f5cbf0da18dc1cae
Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-on: https://review.coreboot.org/17807
Tested-by: build bot (Jenkins)
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Linux needs these SBI calls, but so far it seems to work when they don't
do anything.
Change-Id: I2cd0bb3ab91e89805fed84ec87e4a48ce70c3a46
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/17593
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
o The first 4G of physical address space is now mapped at 0.
o The first 4G of physical address space is now mapped at 1 << 38.
o The first 2G of DRAM (2 - 4 GiB of physical address space)
is now mapped at the top of memory save for the last 4K
i.e. at 0xffffffff80000000, with SBI page at the very top.
Of these, we hope to remove the *most* of the
last one once the gcc toolchain
can handle linking programs that can run at "top 33 bits
of address not all ones (but bit 63 set)". The 4K mapping
of the top of the 64 bit address space will always remain,
however, for SBI calls.
Change-Id: I77b151720001bddad5563b0f8e1279abcea056fa
Reviewed-on: https://review.coreboot.org/17403
Tested-by: build bot (Jenkins)
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
This variable can be set in a debugger (e.g. Spike)
to finely control which traps go to coreboot and
which go to the supervisor.
Change-Id: I292264c15f002c41cf8d278354d8f4c0efbd0895
Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-on: https://review.coreboot.org/17404
Tested-by: build bot (Jenkins)
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
The riscv 1.9 standard defines a textual config string to be passed
to kernels and hypervisors. Change the payload function to pass
this string in a0.
Change-Id: I3be7f1712accf2d726704e4c970f22749d3c3f36
Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-on: https://review.coreboot.org/17254
Tested-by: build bot (Jenkins)
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
These functions will allow us to remove hardcodes,
as long as we can verify the qemu and lowrisc targets
implement the configstring correctly. Hence, for the
most part, we'll start with mainboard changes first.
Define a new config variable, CONFIG_RISCV_CONFIGSTRING,
which has a default value that works on all existing
systems but which can be changed
as needed for a new SOC or mainboard.
Change-Id: I7dd3f553d3e61f1c49752fb04402b134fdfdf979
Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-on: https://review.coreboot.org/17256
Tested-by: build bot (Jenkins)
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Note that currently, traps are only handled by the trap handler
installed in the bootblock. The romstage and ramstage don't override it.
TEST=Booted emulation/spike-qemu and lowrisc/nexys4ddr with a linux
payload. It worked as much as before (Linux didn't boot, but it
made some successful SBI calls)
Change-Id: Icce96ab3f41ae0f34bd86e30f9ff17c30317854e
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/17057
Tested-by: build bot (Jenkins)
Reviewed-by: Martin Roth <martinroth@google.com>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
After I did a new toolchain build, I found the
the mhartid register value is wrong for Spike.
The docs seem to agree with Spike, not the
code the toolchain produces?
Until such time as the bitstreams and toolchain can find
a way to agree, just hardcode it. We've been playing this game
for two years now so this is hardly a new approach.
This is intentionally ugly because we really need the
toolchains and emulators and bitstreams to sync up,
and that's not happening yet. Lowrisc
allegedly implements the v1.9 spec but it's PTEs are clearly
1.7. Once it all settles down we can just use constants
supplied by the toolchain.
I hope the syncup will have happened by the workshop in November.
This gets spike running again.
Change-Id: If259bcb6b6320ef01ed29a20ce3d2dcfd0bc7326
Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-on: https://review.coreboot.org/17183
Tested-by: build bot (Jenkins)
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
This version of coreboot successfully starts a Harvey (Plan 9) kernel as a payload,
entering main() with no supporting assembly code for startup. The Harvey port
is not complete so it just panics but ... it gets started.
We provide a standard payload function that takes a pointer argument
and makes the jump from machine to supervisor mode;
the days of kernels running in machine mode are over.
We do some small tweaks to the virtual memory code. We temporarily
disable two functions that won't work on some targets as register
numbers changed between 1.7 and 1.9. Once lowrisc catches up
we'll reenable them.
We add the PAGETABLES to the memlayout.ld and use _pagetables in the virtual
memory setup code.
We now use the _stack and _estack from memlayout so we know where things are.
As time goes on maybe we can kill all the magic numbers.
Change-Id: I6caadfa9627fa35e31580492be01d4af908d31d9
Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-on: https://review.coreboot.org/17058
Tested-by: build bot (Jenkins)
Reviewed-by: Furquan Shaikh <furquan@google.com>
The stack pointer (SP) is already printed in print_trap_information.
Don't print it again in handle_misaligned_{load,store}.
Change-Id: I156cf5734a16605decc2280e54e6db3089e094a2
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/16996
Tested-by: build bot (Jenkins)
Reviewed-by: Martin Roth <martinroth@google.com>
The pointers printed on unaligned memory accesses are now aligned to
those printed at the end of print_trap_information.
Change-Id: Ifec1cb639036ce61b81fe8d0a9b14c00d5b2781a
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/16983
Tested-by: build bot (Jenkins)
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
TEST=Compiled for and ran on spike; it booted as before.
Change-Id: Id173643a3571962406f9191db248b206235dca35
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/16995
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
spike_util.h:
- (LOG_)REGBYTES and STORE are already defined in
arch/riscv/include/bits.h.
- TOHOST_CMD, FROMHOST_* are helper macros for the deprecated
Host-Target Interface (HTIF).
qemu_util.c:
- mcall_query_memory now uses mprv_write_ulong instead of first
translating the address and then accessing it normally. Thus,
translate_address isn't used anymore.
- Several functions used the deprecated HTIF CSRs mtohost/mfromhost.
They have mostly been replaced by stub implementations.
- htif_interrupt and testPrint were unused and have been deleted.
spike_util.c:
- translate_address and testPrint were unused and have been deleted.
After this commit, spike_util.c and qemu_util.c are exactly the same and
can be moved to a common location.
Change-Id: I1789bad8bbab964c3f2f0480de8d97588c68ceaf
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/16985
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Normally machine-mode code operates completely within physical address
space. When emulating less privileged memory accesses (e.g. when the
hardware doesn't support unaligned read/write), it is useful to access
memory through the MMU (and with virtual addresses); this patch
implements this functionality using the MPRV bit.
Change-Id: Ic3b3301f348769faf3ee3ef2a78935dfbcbd15fd
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/16260
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Not all SBI calls are implemented, but it's enough to see a couple dozen
lines of Linux boot output.
It should also be noted that the SBI is still in flux:
https://groups.google.com/a/groups.riscv.org/forum/#!topic/sw-dev/6oNhlW0OFKM
Change-Id: I80e4fe508336d6428ca7136bc388fbc3cda4f1e4
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/16119
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
A new Kconfig option, DEBUG_PRINT_PAGE_TABLES, is added to control this
behaviour. It is currently only available on RISC-V, but other
architectures can use it, too, should the need arise.
Change-Id: I52a863d8bc814ab3ed3a1f141d0a77edc6e4044d
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/16015
Tested-by: build bot (Jenkins)
Reviewed-by: Martin Roth <martinroth@google.com>
In particular:
- Fix the condition of the loop that fills the mid-level page table
- Adhere to the format of sptbr
Change-Id: I575093445edfdf5a8f54b0f8622ff0e89f77ccec
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/16120
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
I copied it from commit e10d2def7d of spike and made sure the copyright
header is still there.
Change-Id: Ie8b56cd2f4855b97d36a112a195866f4ff0feec5
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15832
Tested-by: build bot (Jenkins)
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
mb() is used in src/arch/riscv/ and src/mainboard/emulation/*-riscv/.
It is currently provided by atomic.h, but I think it fits better into
barrier.h.
The "fence" instruction represents a full memory fence, as opposed to
variants such as "fence r, rw" which represent a partial fence. An
operating system might want to use precisely the right fence, but
coreboot doesn't need this level of performance at the cost of
simplicity.
Change-Id: I8d33ef32ad31e8fda38f6a5183210e7bd6c65815
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15830
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
A few things are currently missing:
- The trap handler doesn't set the stack pointer, which can easily
result in trap loops or memory corruptions.
- The SBI trampolin page (as described in version 1.9 of the RISC-V
Privileged Architecture Specification), has been removed for now.
Change-Id: Id89c859fab354501c94a0e82d349349c29fa4cc6
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15591
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
And do the detection just before the initialization.
Change-Id: I9a52430262f799baa298dc4f4ea459880abe250e
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15831
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
These functions are not used anywhere.
Change-Id: Ica1f4650e8774dd796be0aff00054f3698087816
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15829
Tested-by: build bot (Jenkins)
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
This function is unused since coreboot starts payloads in machine mode,
and it uses the obsolete eret instruction.
Change-Id: I98d7d0de5a3959821c21a0ba4319efb610fdefde
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15729
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Using the opcode directly is necessary for the transition to the GCC
6.1.0 based toolchain, because the old toolchain only supports eret and
the new toolchain only supports mret.
Change-Id: I17e14d4793ae5259f7ce3ce0211cbb27305506cc
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15290
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Ron Minnich writes: "we'll change cbfstool to put a header on the
payload to jump to supervisor if that is desired. The principal here is
that payloads are always started in machine mode, but we want to set the
page tables up for them."
Change-Id: I5cbfc90afd3febab33835935f08005136a3f47e9
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15510
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
The different entry points (0x100, 0x140, ...), which were defined in
the RISC-V Privileged Specification 1.7, aren't used anymore. Instead
the Spike bootrom jumps at the start of our image, and traps are handled
through mtvec.
Change-Id: I865adec5e7a752a25bac93a45654ac06e27d5a8e
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15283
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
This is where the RAM is (now), on RISC-V.
We need to put coreboot.rom in RAM because Spike (at the moment) only
supports loading code into the RAM, not into the boot ROM.
Change-Id: I6c9b7cffe5fa414825491ee4ac0d2dad59a2d75c
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15149
Tested-by: build bot (Jenkins)
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
In the default (medlow) code model, pointers are loaded with a lui, addi
instruction sequence:
lui a0, 0xNNNNN
addi a0, a0, 0xNNN
Since lui sign-extends bits 32-63 from bit 31 on RV64, lui/addi can't
load pointers just above 0x80000000, where RISC-V's RAM now lives.
The medany code model gets around this restriction by loading pointers
trough auipc and addi:
auipc a0, 0xNNNNN
addi a0, a0, 0xNNN
This way, any pointer within the current pc ±2G can be loaded, which is
by far sufficient for coreboot.
Change-Id: I77350d9218a687284c1337d987765553cf915a22
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15148
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
The uart8250mem driver needs it.
Change-Id: I09e6a17cedf8a4045f008f5a0d225055d745e8db
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-on: https://review.coreboot.org/15147
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Aaron Durbin