When building for a custom board, it is quite common to maintain a
private branch which include some defconfig and .dts files. But to
hook up those .dts files requires modifying a file "belonging" to
upstream U-Boot, the arch/*/dts/Makefile. Forward-porting that branch
to a newer upstream then often results in a conflict which, while it
is trivial to resolve by hand, makes it harder to have a CI do "try to
build our board against latest upstream".
The .config usually includes information on precisely what .dtb(s) are
needed, so to avoid having to modify the Makefile, simply add the
files in (SPL_)OF_LIST to dtb-y.
A technicality is that (SPL_)OF_LIST is not always defined, so rework
the Kconfig symbols so that (SPL_)OF_LIST is always defined (when
(SPL_)OF_CONTROL), but only prompted for in the cases which used to be
their "depends on".
nios2 and microblaze already have something like this in their
dts/Makefile, and the rationale in commit 41f59f6853 is similar to
the above. So this simply generalizes existing practice. Followup
patches could remove the logic in those two makefiles, just as there's
potential for moving some common boilerplate from all the
arch/*/dts/Makefile files to the new scripts/Makefile.dts.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
The build system already automatically looks for and includes an
in-tree *-u-boot.dtsi when building the control .dtb. However, there
are some things that are awkward to maintain in such an in-tree file,
most notably the metadata associated to public keys used for verified
boot.
The only "official" API to get that metadata into the .dtb is via
mkimage, as a side effect of building an actual signed image. But
there are multiple problems with that. First of all, the final U-Boot
(be it U-Boot proper or an SPL) image is built based on a binary
image, the .dtb, and possibly some other binary artifacts. So
modifying the .dtb after the build requires the meta-buildsystem
(Yocto, buildroot, whatnot) to know about and repeat some of the steps
that are already known to and handled by U-Boot's build system,
resulting in needless duplication of code. It's also somewhat annoying
and inconsistent to have a .dtb file in the build folder which is not
generated by the command listed in the corresponding .cmd file (that
of course applies to any generated file).
So the contents of the /signature node really needs to be baked into
the .dtb file when it is first created, which means providing the
relevant data in the form of a .dtsi file. One could in theory put
that data into the *-u-boot.dtsi file, but it's more convenient to be
able to provide it externally: For example, when developing for a
customer, it's common to use a set of dummy keys for development,
while the consultants do not (and should not) have access to the
actual keys used in production. For such a setup, it's easier if the
keys used are chosen via the meta-buildsystem and the path(s) patched
in during the configure step. And of course, nothing prevents anybody
from having DEVICE_TREE_INCLUDES point at files maintained in git, or
for that matter from including the public key metadata in the
*-u-boot.dtsi directly and ignore this feature.
There are other uses for this, e.g. in combination with ENV_IMPORT_FDT
it can be used for providing the contents of the /config/environment
node, so I don't want to tie this exclusively to use for verified
boot.
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Fix doc formatting error (make htmldocs)
Signed-off-by: Simon Glass <sjg@chromium.org>
Commit 0d52bab46 (mx7dsabre: Enable DM_ETH) changed these flags from 0
(aka GPIO_ACTIVE_HIGH) to GPIO_ACTIVE_LOW. It claimed to "Also sync
device tree with v5.5-rc1", but in the linux tree, these gpios have
always been GPIO_ACTIVE_HIGH ever since this node was introduced
around v4.13 (linux commit 184f39b5).
I'm guessing that the reason for the GPIO_ACTIVE_LOW was to work
around the behaviour of the soft-spi driver back then, which
effectively defaulted to spi-mode 3 and not 0. That was arguably a bug
in the soft-spi driver, which then got fixed in 0e146993bb (spi: add
support for all spi modes with soft spi), but that commit then broke
ethernet on this board.
Fix it by setting the gpios as active high, which as a bonus actually
brings us in sync with the .dts in the linux source tree.
Without this, one gets
Net: Could not get PHY for FEC0: addr 0
No ethernet found.
With this, ethernet (at least ping and tftp) works as expected from
the U-Boot shell.
Cc: Fabio Estevam <festevam@gmail.com>
Cc: Joris Offouga <offougajoris@gmail.com>
Cc: "Christian Bräuner Sørensen" <yocto@bsorensen.net>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Check that the watchdog_reset() implementation in wdt-uclass behaves
as expected:
- resets all activated watchdog devices
- leaves unactivated/stopped devices alone
- that the rate-limiting works, with a per-device threshold
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
It seems that no other test has claimed gpio_a:7 yet, so use that.
The only small wrinkle is modifying the existing wdt test to use
uclass_get_device_by_driver() since we now have two UCLASS_WDT
instances in play, so it's a little more robust to fetch the device by
driver and not merely uclass+index.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
A rather common kind of external watchdog circuit is one that is kept
alive by toggling a gpio. Add a driver for handling such a watchdog.
The corresponding linux driver apparently has support for some
watchdog circuits which can be disabled by tri-stating the gpio, but I
have never actually encountered such a chip in the wild; the whole
point of adding an external watchdog is usually that it is not in any
way under software control. For forward-compatibility, and to make DT
describe the hardware, the current driver only supports devices that
have the always-running property. I went a little back and forth on
whether I should fail ->probe or only ->start, and ended up deciding
->start was the right place.
The compatible string is probably a little odd as it has nothing to do
with linux per se - however, I chose that to make .dts snippets
reusable between device trees used with U-Boot and linux, and this is
the (only) compatible string that linux' corresponding driver and DT
binding accepts. I have asked whether one should/could add "wdt-gpio"
to that binding, but the answer was no:
https://lore.kernel.org/lkml/CAL_JsqKEGaFpiFV_oAtE+S_bnHkg4qry+bhx2EDs=NSbVf_giA@mail.gmail.com/
If someone feels strongly about this, I can certainly remove the
"linux," part from the string - it probably wouldn't the only place where
one can't reuse a DT snippet as-is between linux and U-Boot.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
A board can have and make use of more than one watchdog device, say
one built into the SOC and an external gpio-petted one. Having
wdt-uclass only handle the first is both a little arbitrary and
unexpected.
So change initr_watchdog() so we visit (probe) all DM watchdog
devices, and call the init_watchdog_dev helper for each.
Similarly let watchdog_reset() loop over the whole uclass - each
having their own ratelimiting metadata, and a separate "is this device
running" flag.
This gets rid of the watchdog_dev member of struct global_data. We
do, however, still need the GD_FLG_WDT_READY set in
initr_watchdog(). This is because watchdog_reset() can get called
before DM is ready, and I don't think we can call uclass_get() that
early.
The current code just returns 0 if "getting" the first device fails -
that can of course happen because there are no devices, but it could
also happen if its ->probe call failed. In keeping with that, continue
with the handling of the remaining devices even if one fails to
probe. This is also why we cannot use uclass_probe_all().
If desired, it's possible to later add a per-device "u-boot,autostart"
boolean property, so that one can do CONFIG_WATCHDOG_AUTOSTART
per-device.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Since the gd->watchdog_dev member is going away, switch to using the
new wdt_stop_all() helper.
While here, clean up the preprocessor conditional: The ->watchdog_dev
member is actually guarded by CONFIG_WDT [disabling that in
x530_defconfig while keeping CONFIG_WATCHDOG breaks the build], and in
the new world order so is the existence of the wdt_stop_all()
function.
Actually, existence of wdt_stop_all() depends on CONFIG_${SPL_}WDT, so
really spell the condition using CONFIG_IS_ENABLED, and make it a C
rather than cpp if.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Since the watchdog_dev member of struct global_data is going away in
favor of the wdt-uclass handling all watchdog devices, prepare for
that by adding a helper to call wdt_stop() on all known devices.
If an error is encountered, still do wdt_stop() on remaining devices,
but remember and return the first error seen.
Initially, this will only be used in one single
place (board/alliedtelesis/x530/x530.c).
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
For the unit tests, it is more convenient if the tests are in charge
of when the watchdog devices are started and stopped, so prevent
wdt-uclass from doing it automatically.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
As a step towards handling all DM watchdogs in watchdog_reset(), use a
per-device flag to keep track of whether the device has been started
instead of a bit in gd->flags.
We will still need that bit to know whether we are past
initr_watchdog() and hence have populated gd->watchdog_dev -
incidentally, that is how it was used prior to commit 9c44ff1c5f.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
In preparation for handling all DM watchdogs in watchdog_reset(), pull
out the code which handles starting (or not) the gd->watchdog_dev
device.
Include the device name in various printfs.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
The addition of .pre_probe and .per_device_auto made this look
bad. Fix it.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
As preparation for having the wdt-uclass provided watchdog_reset()
function handle all DM watchdog devices, and not just the first such,
introduce a uclass-owned struct to hold the reset_period and
next_reset, so these become per-device instead of being static
variables.
No functional change intended.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
wdt_start() does the "no ->start? return -ENOSYS" check, don't
open-code that in wdt_expire_now().
Also, wdt_start() maintains some global (and later some per-device)
state, which would get out of sync with this direct method call - not
that it matters much here since the board is supposed to reset very
soon.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
There's currently no user of %p[iI]6, so including ip6_addr_string()
in the image is a waste of bytes. It's easy enough to have the
compiler elide it without removing the code completely.
The closest I can find to anybody "handling" ipv6 in U-Boot currently
is in efi_net.c which does
if (ipv6) {
ret = EFI_UNSUPPORTED;
As indicated in the comment, it can easily be put back, but preferably
under a config knob.
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
U-Boot doesn't support %pS/%pF or any other kind of kallsyms-like
lookups. Remove the comment.
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
This saves some code, both in terms of #LOC and .text size, and it is
also the normal convention that foo(...) is implemented in terms of
vfoo().
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
I removed CONFIG_IS_BUILTIN and CONFIG_IS_MODULE in commit
7d78a4547d ("linux/kconfig.h: remove unused helper macros"), but
fixdep.c still looks for those. It's harmless, but also pointless and
possibly confusing to a future reader.
Fixes: 7d78a4547d ("linux/kconfig.h: remove unused helper macros")
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Each _device_ belonging to a given uclass of course has its own ->ops,
of a type determined by and known to the uclass.
However, no instance of a uclass_driver seems to populate ->ops, and
the only reference to it in code is this relocation.
Moreover, it's not really clear what could sensibly be assigned; it
would have to be some "struct uclass_ops *" providing a set of methods
for the core to call on that particular uclass, but should the need
for that ever arise, it would be better to have a member of that
particular type instead of void*.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
The layout and contents of struct global_data depends on a lot of
CONFIG_* preprocessor macros, not all of which are entirely converted
to Kconfig - not to mention weird games played here and there. This
can result in one translation unit using one definition of struct
global_data while the actual layout is another.
That can be very hard to debug. But we already have a mechanism that
can help catch such bugs at build time, namely the asm-offsets
machinery which is necessary anyway to provide assembly code with the
necessary constants. So make sure that every C translation unit that
include global_data.h actually sees the same size of struct
global_data as that which was seen by the asm-offsets.c TU.
It is likely that this patch will break the build of some boards. For
example, without the patch from Matt Merhar
(https://lists.denx.de/pipermail/u-boot/2021-May/450135.html) or some
other fix, this breaks P2041RDB_defconfig:
CC arch/powerpc/lib/traps.o
AS arch/powerpc/cpu/mpc85xx/start.o
In file included from include/asm-generic/global_data.h:26,
from ./arch/powerpc/include/asm/global_data.h:109,
from include/init.h:21,
from arch/powerpc/lib/traps.c:7:
include/linux/build_bug.h:99:41: error: static assertion failed: "sizeof(struct global_data) == GD_SIZE"
99 | #define __static_assert(expr, msg, ...) _Static_assert(expr, msg)
| ^~~~~~~~~~~~~~
include/linux/build_bug.h:98:34: note: in expansion of macro ‘__static_assert’
98 | #define static_assert(expr, ...) __static_assert(expr, ##__VA_ARGS__, #expr)
| ^~~~~~~~~~~~~~~
include/asm-generic/global_data.h:470:1: note: in expansion of macro ‘static_assert’
470 | static_assert(sizeof(struct global_data) == GD_SIZE);
| ^~~~~~~~~~~~~
make[1]: *** [scripts/Makefile.build:266: arch/powerpc/lib/traps.o] Error 1
make: *** [Makefile:1753: arch/powerpc/lib] Error 2
make: *** Waiting for unfinished jobs....
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
[Linux commit 6bab69c65013bed5fce9f101a64a84d0385b3946]
BUILD_BUG_ON() is a little annoying, since it cannot be used outside
function scope. So one cannot put assertions about the sizeof() a
struct next to the struct definition, but has to hide that in some more
or less arbitrary function.
Since gcc 4.6 (which is now also the required minimum), there is support
for the C11 _Static_assert in all C modes, including gnu89. So add a
simple wrapper for that.
_Static_assert() requires a message argument, which is usually quite
redundant (and I believe that bug got fixed at least in newer C++
standards), but we can easily work around that with a little macro
magic, making it optional.
For example, adding
static_assert(sizeof(struct printf_spec) == 8);
in vsprintf.c and modifying that struct to violate it, one gets
./include/linux/build_bug.h:78:41: error: static assertion failed: "sizeof(struct printf_spec) == 8"
#define __static_assert(expr, msg, ...) _Static_assert(expr, "" msg "")
godbolt.org suggests that _Static_assert() has been support by clang
since at least 3.0.0.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
Check that a variable defined in /config/environment is found in the
run-time environment, and that clearing fdt_env_path from within that
node works.
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
[trini: Conditionalize the test being linked in]
Signed-off-by: Tom Rini <trini@konsulko.com>
It can be useful to use the same U-Boot binary for multiple purposes,
say the normal one, one for developers that allow breaking into the
U-Boot shell, and one for use during bootstrapping which runs a
special-purpose bootcmd. Or one can have several board variants that
can share almost all boot logic, but just needs a few tweaks in the
variables used by the boot script.
To that end, allow the control dtb to contain a /config/enviroment
node (or whatever one puts in fdt_env_path variable), whose
property/value pairs are used to update the run-time environment after
it has been loaded from its persistent location.
The indirection via fdt_env_path is for maximum flexibility - for
example, should the user wish (or board logic dictate) that the values
in the DTB should no longer be applied, one simply needs to delete the
fdt_env_path variable; that can even be done automatically by
including a
fdt_env_path = "";
property in the DTB node.
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
When CONFIG_USE_DEFAULT_ENV_FILE=y and the file
CONFIG_DEFAULT_ENV_FILE is empty (or at least doesn't contain any
non-comment, non-empty lines), we end up feeding nothing into xxd,
which in turn then outputs nothing. Then blindly appending ", 0x00"
means that we end up trying to compile (roughly)
const char defaultenv[] = { , 0x00 }
which is of course broken.
To fix that, change the frobbing of the text file so that we always
end up printing an extra empty line (which gets turned into that extra
nul byte we need) - that corresponds better to the binary format
consisting of a series of key=val nul terminated strings, terminated
by an empty string.
Reported-by: Oleksandr Suvorov <oleksandr.suvorov@toradex.com>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Oleksandr Suvorov <oleksandr.suvorov@toradex.com>
When flush_cache() is called during boot on our ~7M kernel image, the
hundreds of thousands of WATCHDOG_RESET calls end up adding
significantly to boottime. Flushing a single cache line doesn't take
many microseconds, so doing these calls for every cache line is
complete overkill.
The generic watchdog_reset() provided by wdt-uclass.c actually
contains some rate-limiting logic that should in theory mitigate this,
but alas, that rate-limiting must be disabled on powerpc because of
its get_timer() implementation - get_timer() works just fine until
interrupts are disabled, but it just so happens that the "big"
flush_cache() call happens in the part of bootm where interrupts are
indeed disabled. [1] [2] [3]
I have checked with objdump that the generated code doesn't change
when this option is left at its default value of 0: gcc is smart
enough to see that the ">=" comparison is tautologically true, hence
all assignments to "flushed" are eliminated as dead stores.
On our board, setting the option to something like 65536 ends up
reducing total boottime by about 0.8 seconds.
[1] https://patchwork.ozlabs.org/project/uboot/patch/20200605111657.28773-1-rasmus.villemoes@prevas.dk/
[2] https://lists.denx.de/pipermail/u-boot/2021-April/446906.html
[3] https://lists.denx.de/pipermail/u-boot/2021-April/447280.html
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
CONFIG_5xx hasn't existed since commit 5025897774 (powerpc, 5xx:
remove support for 5xx). Remove this last mention of it.
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Having WATCHDOG_RESET() called automatically from the timer interrupt
runs counter to the idea of a watchdog device - if the board runs into
an infinite loops with interrupts still enabled, the watchdog will
never fire.
When using CONFIG_(SPL_)WDT, the watchdog_reset function is a lot more
complicated than just poking a few SOC-specific registers - it
involves accessing all kinds of global data, and if the interrupt
happens at the wrong time (say, in the middle of an WATCHDOG_RESET()
call from ordinary code), that can end up corrupting said global data.
Allow the board to opt out of calling WATCHDOG_RESET() from the timer
interrupt handler by setting CONFIG_SYS_WATCHDOG_FREQ to 0 - as that
setting is currently nonsensical (it would be compile-time
divide-by-zero), it cannot affect any existing boards.
Add documentation for both the existing and extended meaning of
CONFIG_SYS_WATCHDOG_FREQ.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
The code, which is likely copied from arch/powerpc/lib/interrupts.c,
lacks a fallback definition of CONFIG_SYS_WATCHDOG_FREQ and refers to
a non-existing timestamp variable - obviously priv->timestamp is
meant.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Some boards don't work with the rate-limiting done in the generic
watchdog_reset() provided by wdt-uclass.
For example, on powerpc, get_timer() ceases working during bootm since
interrupts are disabled before the kernel image gets decompressed, and
when the decompression takes longer than the watchdog device
allows (or enough of the budget that the kernel doesn't get far enough
to assume responsibility for petting the watchdog), the result is a
non-booting board.
As a somewhat hacky workaround (because DT is supposed to describe
hardware), allow specifying hw_margin_ms=0 in device tree to
effectively disable the ratelimiting and actually ping the watchdog
every time watchdog_reset() is called. For that to work, the "has
enough time passed" check just needs to be tweaked a little to allow
the now==next_reset case as well.
Suggested-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Stefan Roese <sr@denx.de>
When CONFIG_USE_DEFAULT_ENV_FILE=y and the file
CONFIG_DEFAULT_ENV_FILE is empty (or at least doesn't contain any
non-comment, non-empty lines), we end up feeding nothing into xxd,
which in turn then outputs nothing. Then blindly appending ", 0x00"
means that we end up trying to compile (roughly)
const char defaultenv[] = { , 0x00 }
which is of course broken.
To fix that, change the frobbing of the text file so that we always
end up printing an extra empty line (which gets turned into that extra
nul byte we need) - that corresponds better to the binary format
consisting of a series of key=val nul terminated strings, terminated
by an empty string.
Reported-by: Oleksandr Suvorov <oleksandr.suvorov@toradex.com>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Oleksandr Suvorov <oleksandr.suvorov@toradex.com>
When flush_cache() is called during boot on our ~7M kernel image, the
hundreds of thousands of WATCHDOG_RESET calls end up adding
significantly to boottime. Flushing a single cache line doesn't take
many microseconds, so doing these calls for every cache line is
complete overkill.
The generic watchdog_reset() provided by wdt-uclass.c actually
contains some rate-limiting logic that should in theory mitigate this,
but alas, that rate-limiting must be disabled on powerpc because of
its get_timer() implementation - get_timer() works just fine until
interrupts are disabled, but it just so happens that the "big"
flush_cache() call happens in the part of bootm where interrupts are
indeed disabled. [1] [2] [3]
I have checked with objdump that the generated code doesn't change
when this option is left at its default value of 0: gcc is smart
enough to see that the ">=" comparison is tautologically true, hence
all assignments to "flushed" are eliminated as dead stores.
On our board, setting the option to something like 65536 ends up
reducing total boottime by about 0.8 seconds.
[1] https://patchwork.ozlabs.org/project/uboot/patch/20200605111657.28773-1-rasmus.villemoes@prevas.dk/
[2] https://lists.denx.de/pipermail/u-boot/2021-April/446906.html
[3] https://lists.denx.de/pipermail/u-boot/2021-April/447280.html
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
CONFIG_5xx hasn't existed since commit 5025897774 (powerpc, 5xx:
remove support for 5xx). Remove this last mention of it.
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Having WATCHDOG_RESET() called automatically from the timer interrupt
runs counter to the idea of a watchdog device - if the board runs into
an infinite loops with interrupts still enabled, the watchdog will
never fire.
When using CONFIG_(SPL_)WDT, the watchdog_reset function is a lot more
complicated than just poking a few SOC-specific registers - it
involves accessing all kinds of global data, and if the interrupt
happens at the wrong time (say, in the middle of an WATCHDOG_RESET()
call from ordinary code), that can end up corrupting said global data.
Allow the board to opt out of calling WATCHDOG_RESET() from the timer
interrupt handler by setting CONFIG_SYS_WATCHDOG_FREQ to 0 - as that
setting is currently nonsensical (it would be compile-time
divide-by-zero), it cannot affect any existing boards.
Add documentation for both the existing and extended meaning of
CONFIG_SYS_WATCHDOG_FREQ.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
The code, which is likely copied from arch/powerpc/lib/interrupts.c,
lacks a fallback definition of CONFIG_SYS_WATCHDOG_FREQ and refers to
a non-existing timestamp variable - obviously priv->timestamp is
meant.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Some boards don't work with the rate-limiting done in the generic
watchdog_reset() provided by wdt-uclass.
For example, on powerpc, get_timer() ceases working during bootm since
interrupts are disabled before the kernel image gets decompressed, and
when the decompression takes longer than the watchdog device
allows (or enough of the budget that the kernel doesn't get far enough
to assume responsibility for petting the watchdog), the result is a
non-booting board.
As a somewhat hacky workaround (because DT is supposed to describe
hardware), allow specifying hw_margin_ms=0 in device tree to
effectively disable the ratelimiting and actually ping the watchdog
every time watchdog_reset() is called. For that to work, the "has
enough time passed" check just needs to be tweaked a little to allow
the now==next_reset case as well.
Suggested-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Stefan Roese <sr@denx.de>
This is roughly the U-Boot side equivalent to commit
e282c422e0 (tools: fw_env: use erasesize from MEMGETINFO ioctl). The
motivation is the case where one has a board with several revisions,
where the SPI flashes have different erase sizes.
In our case, we have an 8K environment, and the flashes have erase
sizes of 4K (newer boards) and 64K (older boards). Currently, we must
set CONFIG_ENV_SECT_SIZE to 64K to make the code work on the older
boards, but for the newer ones, that ends up wasting quite a bit of
time reading/erasing/restoring the last 56K.
At first, I wanted to allow setting CONFIG_ENV_SECT_SIZE to 0 to mean
"use the erase size the chip reports", but that config
option is used in a number of preprocessor conditionals, and shared
between ENV_IS_IN_FLASH and ENV_IS_IN_SPI_FLASH.
So instead, introduce a new boolean config option, which for now can
only be used with ENV_IS_IN_SPI_FLASH. If left off, there's no change
in behaviour.
The only slightly annoying detail is that, when selected, the compiler
is apparently not smart enough to see that the the saved_size and
saved_offset variables are only used under the same "if (sect_size >
CONFIG_ENV_SIZE)" condition as where they are computed, so we need to
initialize them to 0 to avoid "may be used uninitialized" warnings.
On our newer boards with the 4K erase size, saving the environment now
takes 0.080 seconds instead of 0.53 seconds, which directly translates
to that much faster boot time since our logic always causes the
environment to be written during boot.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
As preparation for the next patch, use a local variable to represent
the sector size. No functional change.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
When trying to port our mpc8309-based board to DM_ETH, on top of
Heiko's patches, I found that nothing in mdio-uclass.c seems to
support the use of a fixed-link subnode of the ethernet DT node. That
is, the ethernet node looks like
enet0: ethernet@2000 {
device_type = "network";
compatible = "ucc_geth";
...
fixed-link {
reg = <0xffffffff>;
speed = <100>;
full-duplex;
};
but the current code expects there to be phy-handle property. Adding
that, i.e.
phy-handle = <&enet0phy>;
enet0phy: fixed-link {
just makes the code break a few lines later since a fixed-link node
doesn't have a reg property. Ignoring the dtc complaint and adding a
dummy reg property, we of course hit "can't find MDIO bus for node
ethernet@2000" since indeed, the parent node of the phy node does not
represent an MDIO bus. So that's obviously the wrong path.
Now, in linux, it seems that the fixed link case is treated specially;
in the of_phy_get_and_connect() which roughly corresponds to
dm_eth_connect_phy_handle() we have
if (of_phy_is_fixed_link(np)) {
ret = of_phy_register_fixed_link(np);
...
} else {
phy_np = of_parse_phandle(np, "phy-handle", 0);
...
}
phy = of_phy_connect(dev, phy_np, hndlr, 0, iface);
And U-Boot's phy_connect() does have support for fixed-link
subnodes. Calling phy_connect() directly with NULL bus and a dummy
address does seem to make the ethernet work.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Heiko Schocher <hs@denx.de>
The previous patch made the table look bad. Fix it, and leave some
space for a future element being a bit longer than the current
maximum.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
When setting aside a GPT partition for holding the U-Boot environment,
having a partition type GUID [1] indicating "Linux filesystem" (as
most tools default to) is somewhat misleading - and there's no other
well-known type GUID that is better suited. So to have a canonical
value to put into the type field, define
3de21764-95bd-54bd-a5c3-4abe786f38a8
to mean a partition holding a U-Boot environment.
This is a v5 namespace-name GUID [2], generated [3] from a namespace
of "25cbcde0-8642-47c6-a298-1a3a57cd256b" and name "U-Boot
environment".
Should future type GUIDs be defined in the context of U-Boot, it's
sensible to use that same namespace GUID.
[1] https://en.wikipedia.org/wiki/GUID_Partition_Table#Partition_type_GUIDs
[2] https://en.wikipedia.org/wiki/Universally_unique_identifier#Versions_3_and_5_(namespace_name-based)
[3] https://www.uuidtools.com/v5
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
There's no reason to require an appropriately sized output parameter
for the string, that's error-prone should the table ever grow an
element with a longer string. We can just return the const char*
pointer directly.
Update the only caller accordingly, and get rid of pointless ifdeffery
in the header so that the compiler always sees a declaration and can
thus do type-checking, whether or not PARTITION_TYPE_GUID is enabled
or not.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
This code is intended do prevent one from setting a shell variable abc
by doing
abc=123
if an environment variable named abc already exists. However, the
check is broken, since the env_get is done before the split on =, so
we look up whether an environment variable "abc=123" exists, which is
obviously never the case.
One could move the code to below the split on =, but instead, just
remove it, saving a little .text: The check has never worked as
intended (it goes all the way back to the initial git commit), and it
would anyway not guard against one first setting the shell variable,
then doing 'env set abc xyz'.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Commit fdf0819afb (rsa: fix alignment issue when getting public
exponent) changed the logic to avoid doing an 8-byte access to a
possibly-not-8-byte-aligned address.
However, using rsa_convert_big_endian is wrong: That function converts
an array of big-endian (32-bit) words with the most significant word
first (aka a BE byte array) to an array of cpu-endian words with the
least significant word first. While the exponent is indeed _stored_ as
a big-endian 64-bit word (two BE words with MSW first), we want to
extract it as a cpu-endian 64 bit word. On a little-endian host,
swapping the words and byte-swapping each 32-bit word works, because
that's the same as byte-swapping the whole 64 bit word. But on a
big-endian host, the fdt32_to_cpu are no-ops, but
rsa_convert_big_endian() still does the word-swapping, breaking
verified boot.
To fix that, while still ensuring we don't do unaligned accesses, add
a little helper that first memcpy's the bytes to a local fdt64_t, then
applies fdt64_to_cpu(). [The name is chosen based on the
[bl]eXX_to_cpup in linux/byteorder/generic.h].
Fixes: fdf0819afb ("rsa: fix alignment issue when getting public exponent")
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
Fix build failure, it used to get this implicitly through common.h
until f7ae49fc4f (common: Drop log.h from common header).
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Heiko Schocher <hs@denx.de>
Somewhere between v2020.04 and v2020.07 the mpc8xxx_spi driver broke,
I'm guessing due to this hunk
@@ -559,6 +560,8 @@ int dm_gpio_set_dir_flags(struct gpio_desc *desc, ulong flags)
if (ret)
return ret;
+ /* combine the requested flags (for IN/OUT) and the descriptor flags */
+ flags |= desc->flags;
ret = _dm_gpio_set_dir_flags(desc, flags);
from commit 695e5fd546 ("gpio: update dir_flags management"). But
the blame is mostly on the driver itself which seems rather confused:
The chip select gpios are requested with GPIOD_ACTIVE_LOW, but then in
each activate/deactivate, dm_gpio_set_dir_flags() is called with
merely GPIOD_IS_OUT, and then the driver call set_value(0) for
activate.
That used to work, but with the above hunk, the ACTIVE_LOW setting
from the request becomes persistent, so the gpio driver ends up being
asked to set the value to 1 in mpc8xxx_spi_cs_activate().
So drop the dm_gpio_set_dir_flags() calls in the activate/deactivate
functions, and use a value of 1 to mean "logically enabled".
Ideally, I think we should also drop the GPIOD_ACTIVE_LOW from the
request and make it up to the list of gpio cs in DT to indicate
whether that CS is enabled when driven low (as is of course usually
the case), but that requires changing
arch/powerpc/dts/gdsys/gazerbeam-base.dtsi among others, and I don't
have that hardware to test on. I have, however, tested our
own (mpc8309-based) hardware with this change, and I have also tested
that removing the GPIOD_ACTIVE_LOW from the request and updating our
DT as
- gpios = <&spisel 0 0>;
+ gpios = <&spisel 0 GPIO_ACTIVE_LOW>;
still works.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
The linux changes since v3.16 are
78283edf2c01 kbuild: setlocalversion: print error to STDERR
b24413180f56 License cleanup: add SPDX GPL-2.0 license identifier to files with no license
6147b1cf1965 scripts/setlocalversion: git: Make -dirty check more robust
8ef14c2c41d9 Revert "scripts/setlocalversion: git: Make -dirty check more robust"
ff64dd485730 scripts/setlocalversion: Improve -dirty check with git-status --no-optional-locks
7a82e3fa28f1 scripts/setlocalversion: clear local variable to make it work for sh
991b78fbd223 scripts: setlocalversion: fix a bashism
3c96bdd0ebfa scripts: setlocalversion: replace backquote to dollar parenthesis
and it's ff64dd485730 that is the motivation for this sync. It fixes
false positive "-dirty" added to the version string when building with
Yocto
(https://lists.openembedded.org/g/openembedded-core/message/137702).
There has been one U-Boot specific patch since this was synced with
linux 3.16: 81630a3b38 (scripts: setlocalversion: safely extract
variables from auto.conf using awk). Keep these changes applied.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
[trini: Re-add 81630a3b38 and reword message]
Signed-off-by: Tom Rini <trini@konsulko.com>
Since commit 4b0bcfa7c4 (Kconfig: Migrate CONFIG_BOOTM_* options),
the config_defaults.h file has been void of any actual content - and
these days, "sane defaults for everyone" is achieved by appropriate
default values in Kconfig. Remove it, and thus make every translation
unit process one less header file.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Booting an lz4-compressed kernel image fails on our powerpc board with
-EPROTONOSUPPORT. Adding a bit of debug prints, we get
magic: 0x184d2204
flags: 0x64
reserved0: 1
has_content_checksum: 1
has_content_size: 0
has_block_checksum: 0
independent_blocks: 1
version: 0
block_descriptor: 70
reserved1: 7
max_block_size: 0
reserved2: 0
So the magic is ok, but the version check fails, also some reserved
bits are apparently set. But that's because the code interprets the
"flags" and "block_descriptor" bytes wrongly:
Using bit-fields to access individual bits of an "on the wire" format
is not portable, not even when restricted to the C flavour implemented
by gcc. Quoting the gcc manual:
* 'The order of allocation of bit-fields within a unit (C90 6.5.2.1,
C99 and C11 6.7.2.1).'
Determined by ABI.
and indeed, the PPC Processor ABI supplement says
* Bit-fields are allocated from right to left (least to most
significant) on Little-Endian implementations and from left to
right (most to least significant) on Big-Endian implementations.
The upstream code (github.com/lz4/lz4) uses explicit shifts and masks
for encoding/decoding:
/* FLG Byte */
*dstPtr++ = (BYTE)(((1 & _2BITS) << 6) /* Version('01') */
+ ((cctxPtr->prefs.frameInfo.blockMode & _1BIT ) << 5)
+ ((cctxPtr->prefs.frameInfo.blockChecksumFlag & _1BIT ) << 4)
+ ((unsigned)(cctxPtr->prefs.frameInfo.contentSize > 0) << 3)
+ ((cctxPtr->prefs.frameInfo.contentChecksumFlag & _1BIT ) << 2)
+ (cctxPtr->prefs.frameInfo.dictID > 0) );
/* Flags */
{ U32 const FLG = srcPtr[4];
U32 const version = (FLG>>6) & _2BITS;
blockChecksumFlag = (FLG>>4) & _1BIT;
blockMode = (FLG>>5) & _1BIT;
contentSizeFlag = (FLG>>3) & _1BIT;
contentChecksumFlag = (FLG>>2) & _1BIT;
dictIDFlag = FLG & _1BIT;
/* validate */
if (((FLG>>1)&_1BIT) != 0) return err0r(LZ4F_ERROR_reservedFlag_set); /* Reserved bit */
if (version != 1) return err0r(LZ4F_ERROR_headerVersion_wrong); /* Version Number, only supported value */
}
Do the same here, and while at it, be more careful to use unaligned
accessors to what is most likely unaligned. Also update the comment to
make it clear that it only refers to the lz4.c file, not the following
code of lz4_wrapper.c.
This has been tested partly, of course, by seeing that my
lz4-compressed kernel now boots, partly by running the (de)compression
test-suite in the (x86_64) sandbox - i.e., it should still work just
fine on little-endian hosts.
Reviewed-by: Julius Werner <jwerner@chromium.org>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
This adds a bunch of preprocessor magic to extend the capabilities of
CONFIG_IS_ENABLED. The existing semantics of
CONFIG_IS_ENABLED(FOO)
expanding to a 1 or 0 (depending on build context and the defined-ness
or not of the appropriate CONFIG_FOO/CONFIG_SPL_FOO/CONFIG_TPL_FOO)
are of course preserved. With this, one is also allowed a two-argument
form
CONFIG_IS_ENABLED(FOO, (something))
which expands to something precisely when CONFIG_IS_ENABLED(FOO) would
expand to 1, and expands to nothing otherwise. It is, in other words,
completely equivalent to the three lines
#if CONFIG_IS_ENABLED(FOO)
something
#endif
The second argument must be parenthesized in order to allow any
tokens, including a trailing comma, to appear - one use case for this
is precisely to make it a bit more ergonomic to build an array and
only include certain items depending on .config. That should increase
both readability and not least "git grep"ability.
A third variant is also introduced,
CONFIG_IS_ENABLED(FOO, (xxx), (yyy))
which corresponds to
#if CONFIG_IS_ENABLED(FOO)
xxx
#else
yyy
#endif
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
U-Boot does not have loadable modules, and nothing currently uses any
of the (CONFIG_)?IS_(BUILTIN|MODULE) macros - only
the (CONFIG_)?IS_ENABLED variants are ever used.
While I understand the desire to keep this somewhat synchronized with
linux, we've already departed by the introduction of the
CONFIG_IS_ENABLED extra logic, and deleting these makes the next patch
much simpler, since I won't have to duplicate a lot of logic for no
real gain (as there are no users).
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Instead of using the arg1_or_junk trick to pick between two choices,
with a bit of duplication between the branches (and most of the
CONFIG_TPL_BUILD case being redundant, as _IS_TPL is known to be
defined to 1 in that case), simply define a prefix that we inject
between CONFIG_ and the given config symbol.
This only requires one level of indirection (to get the
_CONFIG_PREFIX macro expanded before the token concatenation takes
place), and makes it easy to, say, introduce a CONFIG_HOSTTOOL_
prefix. [I would expect most HOSTTOOL_ symbols to just be def_bool y,
but it would allow us to clean up some of the ifdef HOSTCC mess in the
sources shared between U-Boot and host tools.]
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Add tests of the "list", "read" and "write" subcommands of the rtc
shell command.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heiko Schocher <hs@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
In order to allow adding unit tests of the rtc command, add it to the
various sandbox defconfigs.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Define a few aux registers and check that they can be read/written
individually. Also check that one can access the time-keeping
registers directly and get the expected results.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heiko Schocher <hs@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
It's more natural that any write that happens to touch the reset
register should cause a reset, rather than just a write that starts at
that offset.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heiko Schocher <hs@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
The current set method is broken; a simple test case is to first set
the date to something in April, then change the date to 31st May:
=> date 040412122020.34
Date: 2020-04-04 (Saturday) Time: 12:12:34
=> date 053112122020.34
Date: 2020-05-01 (Friday) Time: 12:12:34
or via the amending of the existing rtc_set_get test case similarly:
$ ./u-boot -T -v
=> ut dm rtc_set_get
Test: dm_test_rtc_set_get: rtc.c
expected: 31/08/2004 18:18:00
actual: 01/08/2004 18:18:00
The problem is that after each register write,
sandbox_i2c_rtc_complete_write() gets called and sets the internal
time from the current set of registers. However, when we get to
writing 31 to mday, the registers are in an inconsistent state (mon is
still 4), so the mktime machinery ends up translating April 31st to
May 1st. Upon the next register write, the registers are populated by
sandbox_i2c_rtc_prepare_read(), so the 31 we just wrote to mday gets
overwritten by a 1.
Fix it by writing all registers at once, and for consistency, update
the get method to retrieve them all with one "i2c transfer".
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heiko Schocher <hs@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Mostly as an aid for debugging RTC drivers, provide a command that can
be used to read/write arbitrary registers (assuming the driver
provides the read/write methods or their single-register-at-a-time
variants).
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heiko Schocher <hs@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
This simply consists of renaming the existing pcf2127_read_reg()
helper to follow the naming of the other
methods (i.e. pcf2127_rtc_<method name>) and changing the type of its
"len" parameter.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heiko Schocher <hs@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Similar to how the dm_rtc_{read,write} functions fall back to using
the {read,write}8 methods, do the opposite in the rtc_{read,write}8
functions.
This way, each driver only needs to provide either ->read8 or ->read
to make both rtc_read8() and dm_rtc_read() work - without this, a
driver that provides ->read() would most likely just duplicate the
logic here for implementing a ->read8() method in term of its ->read()
method. The same remarks of course apply to the write case.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heiko Schocher <hs@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Similar to dm_rtc_read(), introduce a helper that allows the caller to
write multiple consecutive 8-bit registers with one call. If the
driver provides the ->write method, use that, otherwise loop using
->write8.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heiko Schocher <hs@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Some users may want to read multiple consecutive 8-bit
registers. Instead of each caller having to implement the loop,
provide a dm_rtc_read() helper. Also, allow a driver to provide a
->read method, which can be more efficient than reading one register
at a time.
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heiko Schocher <hs@denx.de>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
This macro is only used (tested for existence) in mpc83xx.c, which
unconditionally includes mpc83xx.h where it is unconditionally
defined. Removing it makes the remaining code easier to read.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Deciding whether to compile the env_sf_save() function based solely on
CONFIG_SPL_BUILD is wrong: For U-Boot proper, it leads to a build
warning in case CONFIG_CMD_SAVEENV=n (because the initialization of
the .save member is guarded by CONFIG_CMD_SAVEENV, while the
env_sf_save() function is built if !CONFIG_SPL_BUILD - and even
without the CONFIG_CMD_SAVEENV guard, the env_save_ptr() macro would
just expand to NULL, with no reference to env_sf_save visible to the
compiler). And for SPL, when one selects CONFIG_SPL_SAVEENV, one
obviously expects to actually be able to save the environment.
The compiler warning can be fixed by using a "<something> ?
env_sf_save : NULL" construction instead of a macro that just eats its
argument and expands to NULL. That way, if <something> is false,
env_sf_save gets eliminated as dead code, but the compiler still sees
the reference to it.
For <something>, we can use CONFIG_IS_ENABLED(SAVEENV), which is true
precisely:
- For U-Boot proper, when CONFIG_CMD_SAVEENV is set (because
CONFIG_SAVEENV is a hidden config symbol that gets set if and only
if CONFIG_CMD_SAVEENV is set).
- For SPL, when CONFIG_SPL_SAVEENV is set.
As a bonus, this also removes quite a few preprocessor conditionals.
This has been run-time tested on a mpc8309-derived board to verify
that saving the environment does indeed work in SPL with these patches
applied.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
CMD_EEPROM and ENV_IS_IN_EEPROM can be selected independently, and
cmd/eeprom.o gets built in either case, so whether to declare the real
prototypes needs to follow the same logic as whether cmd/eeprom.c is
built. Otherwise a ENV_IS_IN_EEPROM=y, CMD_EEPROM=n build fails
cmd/eeprom.c:73:1: error: expected identifier or ‘(’ before ‘{’ token
{
While at it, fix the dummy replacements (at least assuming they are
meant to allow the code to compile) - they need to have the same type
as the expression they replace, or one gets errors such as
env/eeprom.c: In function ‘eeprom_bus_read’:
env/eeprom.c:37:8: error: void value not ignored as it ought to be
rcode = eeprom_read(dev_addr, offset, buffer, cnt);
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Tom Rini <trini@konsulko.com>
It's possible that the default_environment[] array contains multiple
entries for the same variable, e.g. a setting from env_default.h based
on some CONFIG_* variable, and another from
CONFIG_EXTRA_ENV_SETTINGS. In such a case, the last setting takes
effect.
Hence, in order to be able to use the output from this script as an
CONFIG_DEFAULT_ENV_FILE and get the same default environment as one
currently has, we need to preserve the order. So only sort by the
variable name, and disable the last-resort comparison.
We could pipe the result through uniq to remove duplicate lines, but I
think there's some value in seeing that certain variables are defined
multiple times.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
As noted in rtc_def.h, the tm_mon field in struct rtc_time is 1-12,
unlike in struct tm where it is 0-11. Currently, running "date" prints
the wrong
Date: 2020-04-01 (Friday) Time: 13:05:30
and setting the RTC via the date command is also broken.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
hweight32 is a somewhat expensive way to check for power-of-2. Use the
is_power_of_2 helper, which does the standard and cheap idiom
foo&(foo-1)==0.
add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-96 (-96)
Function old new delta
spi_nor_write 388 292 -96
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Acked-by: Vignesh Raghavendra <vigneshr@ti.com>
Acked-by: Jagan Teki <jagan@amarulasolutions.com>
We have a board with several revisions. The older ones use a nor flash
with 64k erase size, while the newer have a flash with 4k sectors. The
environment size is 8k.
Currently, we have to put a column containing 0x10000 (64k) in
fw_env.config in order for it to work on the older boards. But that
ends up wasting quite a lot of time on the newer boards that could
just erase the 8k occupied by the environment - strace says the 64k
erase takes 0.405 seconds. With this patch, as expected, that's about
an 8-fold better, at 0.043 seconds.
Having different fw_env.config files for the different revisions is
highly impractical, and the correct information is already available
right at our fingertips. So use the erasesize returned by the
MEMGETINFO ioctl when the fourth and fifth columns (sector size and
#sectors, respectively) are absent or contain 0, a case where the
logic previously used to use the environment size as erase size (and
consequently computed ENVSECTORS(dev) as 1).
As I'm only testing this on a NOR flash, I'm only changing the logic
for that case, though I think it should be possible for the other
types as well.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
The build currently fails with
drivers/spi/mpc8xxx_spi.c:64:3: warning: implicit declaration of function ‘dev_err’ [-Wimplicit-function-declaration]
...
drivers/spi/built-in.o: In function `mpc8xxx_spi_set_speed':
drivers/spi/mpc8xxx_spi.c:227: undefined reference to `dev_err'
Fixes: 4856cc7a97 (mpc8xxx_spi: implement real ->set_speed)
Fixes: 1a7b462dee (mpc8xxx_spi: put max_cs to use)
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Deciding whether to compile the env_sf_save() function based solely on
CONFIG_SPL_BUILD is wrong: For U-Boot proper, it leads to a build
warning in case CONFIG_CMD_SAVEENV=n (because the env_save_ptr() macro
causes the function to indeed not be referenced anywhere). And for
SPL, when one selects CONFIG_SPL_SAVEENV, one obviously expects to
actually be able to save the environment.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Removing this ifdef/endif pair yields a "defined but unused warning"
for CONFIG_CMD_SAVEENV=n, but that vanishes if we use the ENV_SAVE_PTR
macro instead. This gives slightly better compile testing, and
moreover, it's possible to have
CONFIG_CMD_SAVEENV=n
CONFIG_SPL_SAVEENV=y
SPL_ENV_IS_IN_EXT4=y
in which case env_ext4_save would erroneously not be compiled in.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Always compile the env_fat_save() function, and let
CONFIG_IS_ENABLED(SAVEENV) (via the ENV_SAVE_PTR macro) decide whether
it actually ends up being compiled in.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
The current definition of the env_save_ptr does not take SPL_SAVEENV
into account. Moreover, the way it is implemented means that drivers
need to guard the definitions of their _save methods with ifdefs to
avoid "defined but unused" warnings in case CMD_SAVEENV=n.
The ifdeffery can be avoided by using a "something ? x : NULL"
construction instead and still have the compiler elide the _save
method when it is not referenced. Unfortunately we can't just switch
the existing env_save_ptr macro, since that would give a lot of build
errors unless all the ifdeffery is removed at the same time.
Conversely, removing that ifdeffery first would merely lead to the
"defined but unused" warnings temporarily, but for some storage
drivers it requires a bit more work than just removing their private
CMD_SAVEENV logic.
So introduce an alternative to env_save_ptr, which for lack of a
better name is simply uppercased, allowing one to update storage
drivers piecemeal to both reduce their ifdeffery and honour
CONFIG_SPL_SAVEENV.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Currently, testing whether to compile in support for saving the
environment is a bit awkward when one needs to take SPL_SAVEENV into
account, and quite a few storage drivers currently do not honour
SPL_SAVEENV.
To make it a bit easier to decide whether environment saving should be
enabled, introduce SAVEENV as an alias for the CMD_SAVEENV
symbol. Then one can simply use
CONFIG_IS_ENABLED(SAVEENV)
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
The callback member of struct env_entry is always NULL for an SPL
build. Removing it thus saves a bit of run-time memory in the
SPL (when CONFIG_SPL_ENV_SUPPORT=y) since struct env_entry is embedded
in struct env_entry_node - i.e. about 2KB for the normal case of
512+change hash table entries.
Two small fixups are needed for this, all other references to the
callback member are already under !CONFIG_SPL_BUILD: Don't initialize
.callback in set_flags() - hsearch_r doesn't use that value
anyway. And make env_callback_init() initialize ->callback to NULL for
a new entry instead of relying on an unused or deleted entry having
NULL in ->callback.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
In SPL, environment callbacks are not supported, so e->callback is
always NULL. Removing this makes the SPL a little smaller (about 400
bytes in my ppc build) with no functional change.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
This is preparation for compiling out the "call the callback" code and
associated error handling for SPL, where ->callback is always NULL.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
env.h says this about about callback declarations (U_BOOT_ENV_CALLBACK):
* For SPL these are silently dropped to reduce code size, since environment
* callbacks are not supported with SPL.
So env_callback_init() does a lot of work to not find anything in the
guaranteed empty env_clbk list. Drop callback.o entirely from the link
and stub out the only public function defined in callback.o. This cuts
about 600 bytes from the SPL on my ppc build.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
The combination ENV_IS_IN_NVRAM=y, ENV_IS_IN_REMOTE=y fails to build:
env/remote.o:/mnt/ext4/devel/u-boot/env/remote.c:17: multiple definition of `env_ptr'
env/nvram.o:/mnt/ext4/devel/u-boot/env/nvram.c:41: first defined here
It's not necessarily a meaningful combination, but for build-testing
it's nice to be able to enable most ENV_IS_IN_* at the same time, and
since these env_ptr are not declared anywhere, they really have no
reason to have external linkage.
nand.c and flash.c similarly already define file-scope static env_ptr
variables.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Hitting Ctrl-C is a documented way to exit the sandbox, but it is not
actually equivalent to the reset command. The latter, since it follows
normal process exit, takes care to reset terminal settings and
restoring the O_NONBLOCK behaviour of stdin (and, in a terminal, that
is usually the same file description as stdout and stderr, i.e. some
/dev/pts/NN).
Failure to restore (remove) O_NONBLOCK from stdout/stderr can cause
very surprising and hard to debug problems back in the terminal. For
example, I had "make -j8" consistently failing without much
information about just exactly what went wrong, but sometimes I did
get a "echo: write error". I was at first afraid my disk was getting
bad, but then a simple "dmesg" _also_ failed with write error - so it
was writing to the terminal that was buggered. And both "make -j8" and
dmesg in another terminal window worked just fine.
So install a SIGINT handler so that if the chosen terminal
mode (cooked or raw-with-sigs) means Ctrl-C sends a SIGINT, we will
still call os_fd_restore(), then reraise the signal and die as usual
from SIGINT.
Before:
$ grep flags /proc/$$/fdinfo/1
flags: 0102002
$ ./u-boot
# hit Ctrl-C
$ grep flags /proc/$$/fdinfo/1
flags: 0106002
After:
$ grep flags /proc/$$/fdinfo/1
flags: 0102002
$ ./u-boot
# hit Ctrl-C
$ grep flags /proc/$$/fdinfo/1
flags: 0102002
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
Some watchdog devices, e.g. external gpio-triggered ones, must be
reset more often than once per second, which means that the current
rate-limiting logic in watchdog_reset() fails to keep the board alive.
gpio-wdt.txt in the linux source tree defines a "hw_margin_ms"
property used to specifiy the maximum time allowed between resetting
the device. Allow any watchdog device to specify such a property, and
then use a reset period of one quarter of that. We keep the current
default of resetting once every 1000ms.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Stefan Roese <sr@denx.de>
This function is a bit large for an inline function, and for U-Boot
proper, it is called via a function pointer anyway (in board_r.c), so
cannot be inlined.
It will shortly set a global variable to be used by the
watchdog_reset() function in wdt-uclass.c, so this also allows making
that variable local to wdt-uclass.c.
The WATCHDOG_TIMEOUT_SECS define is not used elsewhere.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Stefan Roese <sr@denx.de>
Since WATCHDOG_TIMEOUT_MSECS was converted to Kconfig (commit
ca51ef7c0c), CONFIG_WATCHDOG_TIMEOUT_MSECS has been guaranteed to be
defined. So remove the dead fallback ifdeffery.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Stefan Roese <sr@denx.de>
Not all boards have the same CSB frequency, nor do every SPI slave
necessarily support running at 16.7 MHz. So implement ->set_speed;
that also allows using a smaller PM (i.e., 0) for slaves that do
support a higher speed.
Based on work by Klaus H. Sørensen.
Cc: Klaus H. Sorensen <khso@prevas.dk>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
There are a few problems with the current driver.
First, it unconditionally reads from dout/writes to din whether or not
those pointers are NULL. So for example a simple "sf probe" ends up
writing four bytes at address 0:
=> md.l 0x0 8
00000000: 45454545 45454545 05050505 05050505 EEEEEEEE........
00000010: 00000000 00000000 07070707 07070707 ................
=> sf probe 0
mpc8xxx_spi_xfer: slave spi@7000:0 dout 0FB53618 din 00000000 bitlen 8
mpc8xxx_spi_xfer: slave spi@7000:0 dout 00000000 din 0FB536B8 bitlen 48
SF: Detected s25sl032p with page size 256 Bytes, erase size 64 KiB, total 4 MiB
=> md.l 0x0 8
00000000: ff000000 45454545 05050505 05050505 ....EEEE........
00000010: 00000000 00000000 07070707 07070707 ................
(here I've change the first debug statement to a printf, and made it
print the din/dout pointers rather than the uints they point at).
Second, as we can also see above, it always writes a full 32 bits,
even if a smaller amount was requested. So for example
=> mw.l $loadaddr 0xaabbccdd 8
=> md.l $loadaddr 8
02000000: aabbccdd aabbccdd aabbccdd aabbccdd ................
02000010: aabbccdd aabbccdd aabbccdd aabbccdd ................
=> sf read $loadaddr 0x400 6
device 0 offset 0x400, size 0x6
mpc8xxx_spi_xfer: slave spi@7000:0 dout 0FB536E8 din 00000000 bitlen 40
mpc8xxx_spi_xfer: slave spi@7000:0 dout 00000000 din 02000000 bitlen 48
SF: 6 bytes @ 0x400 Read: OK
=> sf read 0x02000010 0x400 8
device 0 offset 0x400, size 0x8
mpc8xxx_spi_xfer: slave spi@7000:0 dout 0FB53848 din 00000000 bitlen 40
mpc8xxx_spi_xfer: slave spi@7000:0 dout 00000000 din 02000010 bitlen 64
SF: 8 bytes @ 0x400 Read: OK
=> md.l $loadaddr 8
02000000: 45454545 45450000 aabbccdd aabbccdd EEEEEE..........
02000010: 45454545 45454545 aabbccdd aabbccdd EEEEEEEE........
Finally, when the bitlen is 24 mod 32 (e.g. requesting to read 3 or 7
bytes), the last three bytes and up being the wrong ones, since the
driver does a full 32 bit read and then shifts the wrong byte out:
=> mw.l $loadaddr 0xaabbccdd 4
=> md.l $loadaddr 4
02000000: aabbccdd aabbccdd aabbccdd aabbccdd ................
=> sf read $loadaddr 0x444 10
device 0 offset 0x444, size 0x10
mpc8xxx_spi_xfer: slave spi@7000:0 dout 0FB536E8 din 00000000 bitlen 40
mpc8xxx_spi_xfer: slave spi@7000:0 dout 00000000 din 02000000 bitlen 128
SF: 16 bytes @ 0x444 Read: OK
=> md.l $loadaddr 4
02000000: 552d426f 6f742032 3031392e 30342d30 U-Boot 2019.04-0
=> mw.l $loadaddr 0xaabbccdd 4
=> sf read $loadaddr 0x444 0xb
device 0 offset 0x444, size 0xb
mpc8xxx_spi_xfer: slave spi@7000:0 dout 0FB536E8 din 00000000 bitlen 40
mpc8xxx_spi_xfer: slave spi@7000:0 dout 00000000 din 02000000 bitlen 88
SF: 11 bytes @ 0x444 Read: OK
=> md.l $loadaddr 4
02000000: 552d426f 6f742032 31392e00 aabbccdd U-Boot 219......
Fix all of that by always using a character size of 8, and reject
transfers that are not a whole number of bytes. While it ends being
more work for the CPU, we're mostly bounded by the speed of the SPI
bus, and we avoid writing to the mode register in every loop.
Based on work by Klaus H. Sørensen.
Cc: Klaus H. Sorensen <khso@prevas.dk>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Currently, max_cs is write-only; it's just set in
mpc8xxx_spi_ofdata_to_platdata and not otherwise used.
My mpc8309 was always resetting during an "sf probe 0". It turns out
dm_gpio_set_dir_flags() was being called with garbage, since nothing
had initialized priv->gpios[0] - our device tree used "cs-gpios"
rather than "gpios", so gpio_request_list_by_name() had returned 0.
That would have been a lot easier to figure out if the chip select
index was sanity checked, so rename max_cs to cs_count, and reject a
xfer with a too large cs index.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
The driver correctly handles reading back the value of an output gpio
by reading from the shadow register for output, and from gpdat for
inputs.
Unfortunately, when setting the value of some gpio, we do a RMW cycle
on the gpdat register without taking the shadow register into account,
thus accidentally setting other output gpios (at least those whose
value cannot be read back) to 0 at the same time.
When changing a gpio from input to output, we still need to make sure
it initially has the requested value. So, the procedure is
- update the shadow register
- compute the new gpdir register
- write the bitwise and of the shadow and new gpdir register to gpdat
- write the new gpdir register
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Since some chips don't support reading back the value of output gpios
from the gpdat register, we should not do a RMW cycle (i.e., the
clrbits_be32) on the gpdat register when setting a gpio as input, as
that might accidentally change the value of some other (still
configured as output) gpio.
The extra indirection through mpc8xxx_gpio_set_in() does not help
readability, so just fold the gpdir update into
mpc8xxx_gpio_direction_input().
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
The reference manuals for MPC8308 and MPC8309 both say that the
esdhcctl aka DMA Control Register "is implemented as SDHCCR" in the
System configuration registers. Unfortunately, that doesn't mean that
the registers are just mirrors of each other - any write to esdhcctl
is simply ignored. So to actually enable cache snooping, we
unfortunately have to add a little ifdeffery.
There is, naturally, no description of the bit fields of esdhcctl in
the MPC8309 manual, but comparing the description of esdhcctl from the
LS1021A reference manual to the description of the sdhccr in MPC8309,
one also finds that the fields are bit-reversed, so the bit to set is
0x02000000 rather than 0x00000040 - this is also what board_mmc_init()
uses in the two gdsys/mpc8308/ boards.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Yangbo Lu <yangbo.lu@nxp.com>
I'm also seeing the build failure that commit
7d4776545b env: solve compilation error in SPL
tried to fix, namely that the reference to env_flags_validate from
env_htab cannot be satisfied when flags.o is not built in. However,
that commit got reverted by
d90fc9c3de Revert "env: solve compilation error in SPL"
Necessary, but not sufficient conditions to see this are
CONFIG_SPL=y (obviously)
CONFIG_SPL_ENV_SUPPORT=n (so flags.o does not get compiled)
CONFIG_SPL_LIBCOMMON_SUPPORT=y (so env/built-in.o is part of the SPL link)
Now, these are satisfied for e.g. imx6q_logic_defconfig. But that
builds just fine, and spl/u-boot-spl.map lists .data.env_htab among
the discarded (garbage collected) sections. Yet, on our
mpc8309-derived board, we do see the build failure, so perhaps the
linker works a bit differently on ppc than on ARM, or there's yet some
other configuration option needed to observe the break.
This is another attempt at solving it, which also cleans up
env/Makefile a bit: Introduce a def_bool y symbol CONFIG_ENV_SUPPORT
which complements CONFIG_(SPL/TPL)_SUPPORT. Then use
CONFIG_$(SPL_TPL_)ENV_SUPPORT to decide whether to include the five
basic env/*.o files. For attr.o, flags.o and callback.o, this
shouldn't change anything. Also, common.o and env.o still get
unconditionally built for U-boot proper. But for TPL/SPL, those two
are only included if CONFIG_(SPL/TPL)_SUPPORT is set.
Having that symbol should also allow simplifying conditionals such as
#if !defined(CONFIG_SPL_BUILD) || CONFIG_IS_ENABLED(ENV_SUPPORT)
found in drivers/reset/reset-socfpga.c to just
CONFIG_IS_ENABLED(ENV_SUPPORT).
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
The current mpc83xx_clk driver is broken for any board for which
mpc83xx_has_pci() is true, i.e. anything not MPC8308:
When is_clk_valid() reports that MPC83XX_CLK_PCI is valid,
init_all_clks() proceeds to call init_single_clk(), but that doesn't
know about either MPC83XX_CLK_PCI or has any handling of the
TYPE_SCCR_ONOFF mode correctly returned by retrieve_mode(). Hence
init_single_clk() ends up returning -EINVAL, and the whole board hangs
in serial_init().
The quickest fix is to simply pretend that clock is invalid for
all, since nobody can have been relying on it. Adding proper support
seems to be a bit more involved than just handling TYPE_SCCR_ONOFF:
- The power-on-reset value of SCCR[PCICM] is 0, so
mpc83xx_clk_enable() would probably need to be tought to enable the
clock.
- The frequency of PCI_SYNC_OUT is either SYS_CLK_IN or SYS_CLK_IN/2
depending on the CFG_CLKIN_DIV configuration input, but that can't
be read from software, so to properly fill out
->speed[MPC83XX_CLK_PCI] I think one would need guidance from
Kconfig or dtb.
Partially fixes: 07d538d281 clk: Add MPC83xx clock driver
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Mario Six <mario.six@gdsys.cc>
The MPC8309 has two gpio controllers (which is already correctly
reflected in its struct immap definition).
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Mario Six <mario.six@gdsys.cc>
