dev_request_mem_region returns a void * which shall be checked with
IS_ERR(), but in some cases the valid pointer returned clashes with
error values. This is especially the case on some Atmel SoCs.
This introduces dev_request_mem_resource which returns a struct resource
instead which in any case can be checked with IS_ERR(). It's the drivers
responsibility then to get the IOMEM pointer from the resource.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
When a device probe fails the device is removed from the active list.
If then the device is unregistered afterwards it is removed from the
active list again resulting in a crash. To fix this initialize the
devices active list entry when removing it from the active list.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
This adds initial regmap support. Function prototypes are from
the Kernel, the implemention is mostly rewritten. Currently the
regmap support is limited to the bare minimum to get started.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Matching against driver's name before looking throught its id_table
can lead to a somewhat strange and unintuitive behaviour where a
device whose driver's name matches one of the lines in id_table (which
is not unheard of in Linux kernel) will be probed against said driver
with 'id_table' field set to NULL which in turn will result in
dev_get_drvdata erroring out with -ENODEV.
This patch changes the behaviour such that device_match() only tries
to match against driver's name only if id_table of that driver is not
present.
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Use IOMEM() macro in all dev_request_mem_region_* routines to allow
the proper mapping on all platforms.
Signed-off-by: Yegor Yefremov <yegorslists@googlemail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Let device_detect_by_name detect parent devices aswell. We
separate the devname strings by colons and call device_detect()
for each component. This makes it possible for example to
detect nand0.root.ubi.root
With the above detect will be called for nand0, nand0.root, nand0.root.ubi
and nand0.root.ubi.root. The nand0.root detection step will detect the
UBI volume and attach it.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Since 2011 barebox' of_device_id struct uses unsigned long type for data field:
struct of_device_id {
char *compatible;
unsigned long data;
};
Almost always struct of_device_id.data field are used as pointer
and need 'unsigned long' casting.
E.g. see 'git grep -A 4 of_device_id drivers/' output:
drivers/ata/sata-imx.c:static __maybe_unused struct of_device_id imx_sata_dt_ids[] = {
drivers/ata/sata-imx.c- {
drivers/ata/sata-imx.c- .compatible = "fsl,imx6q-ahci",
drivers/ata/sata-imx.c- .data = (unsigned long)&data_imx6,
drivers/ata/sata-imx.c- }, {
Here is of_device_id struct in linux kernel v4.0:
struct of_device_id {
char name[32];
char type[32];
char compatible[128];
const void *data;
};
Changing of_device_id.data type to 'const void *data' will increase
barebox' linux kernel compatibility and decrease number of 'unsigned
long' casts.
Part of the patch was done using the 'coccinelle' tool with the
following semantic patch:
@rule1@
identifier dev;
identifier type;
identifier func;
@@
func(...) {
<...
- dev_get_drvdata(dev, (unsigned long *)&type)
+ dev_get_drvdata(dev, (const void **)&type)
...>
}
@rule2@
identifier dev;
identifier type;
identifier func;
identifier data;
@@
func(...) {
<...
- dev_get_drvdata(dev, (unsigned long *)&type->data)
+ dev_get_drvdata(dev, (const void **)&type->data)
...>
}
Signed-off-by: Antony Pavlov <antonynpavlov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
There should be no reason to ever change the data pointed to by id_table
or of_compatible. So make this offical by adding some 'const's.
In Linux the corresponding pointers are marked as const, too.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
As expected, we would need deferred probing sooner or later. This is
a first approach to allow devices to return -EPROBE_DEFER and get
sorted into a list of deferred devices that will be re-probed later.
Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
In devices_shutdown we should call the busses remove function
which in turn calls the drivers remove function. Otherwise for
example PCI devices never get removed since they do not have
a remove function but a pcidev->remove function instead.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
The bus implementations currently call the drivers remove
hook unconditionally, but this hook is seldomly populated. Only call
it when it's actually populated.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
The active list is supposed to collect active devices in the
opposite order they are probed. This is used to remove the
devices in the correct order in devices_shutdown. The order
is wrong though when in a drivers probe function other devices
are registered. To get the order right we have to add the new
device to the active list before it is probed.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
The commit
commit 0d7a213345
Author: Antony Pavlov <antonynpavlov at gmail.com>
Date: Wed Sep 10 11:42:19 2014 +0400
MIPS: dts: use physical addresses (as Linux does)
With IOMEM() adapted for MIPS we can use physical addresses
in device tree reg property.
has switched MIPS dts files to physical addresses usage.
The patch was tested on qemu, but qemu malta board is tolerant
of using physical addresses for accesing to device
(Ingenic JZ4755 and JZ4780 processors are tolerant too!).
But other CPUs (e.g. Loongson LS1B) can throw an exception
in this situation.
Additional physical address to virtual address translation
on MIPS is needed.
We already have this in include/common.h
#if defined(CONFIG_MIPS)
#include <asm/addrspace.h>
#define IOMEM(addr) ((void __force __iomem *)CKSEG1ADDR(addr))
#else
#define IOMEM(addr) ((void __force __iomem *)(addr))
#endif
So use IOMEM() to fix dev_request_mem_region() return value.
Signed-off-by: Antony Pavlov <antonynpavlov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
broken since
commit ed6e965824
Author: Sascha Hauer <s.hauer@pengutronix.de>
resource: Let dev_request_mem_region return an error pointer
Introduce dev_request_mem_region_err_null
only used on platform like at91 where the resource address conflicts
with errno PTR.
Signed-off-by: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
This adds a buffer for log messages and a 'dmesg' command to
print the messages. The log buffer is implemented as log objects
rather than a string buffer. This makes it easy to implement
limiting the messages, cleaning the buffer and timestamping
the messages.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Rework the current framework so that I/O mapped I/O resources are
also possible.
Signed-off-by: Michel Stam <michel@reverze.net>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Otherwise we loose memory on each device_unregister. The ethernet
code used to do this before calling unregister_device. This can
now be removed.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
bus_list is only initialized when the device has a bus, but it
needs to be initialized in unregister_device, so initialize the
list unconditionally.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
i2c/spi devices in the devicetree often come with a "vendor,device"
comaptible string. These do not match in barebox, so add a
device_match_of_modalias function which does exactly that.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
The dm9000 needs a resource for an index register and one for
the data register. Both should have a size of the access width,
and not two times the access width. The current code is probably
a leftover when the dm9000 had only one resource.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
We currently have functions to add a device based on function parameters.
This adds the corresponding functions to add resources to a device without
registering the device itself. This is useful to manipulate devices before
registering them.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
A lot of files rely on include/driver.h including include/of.h (and
this including include/errno.h. include the files explicitly so we can
eventually get rid of including of.h from driver.h
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
With this the verbosity of barebox can be controlled during runtime
using the 'loglevel' globalvar.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
This change will allow reuse this function for other buses.
Signed-off-by: Alexander Shiyan <shc_work@mail.ru>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
It becomes a common pattern for boards to find a device and
call device_detect on it. Add a convenience wrapper for it.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
We often encounter the situation where slow devices should not be
probed during startup since probing is slow and maybe unnecessary
for unused devices. With MMC we have the 'probe' device parameter,
for ata we have the same, for USB we have the 'usb' command. Overall
this is not very consistent.
With MMC there is the additional problem that the probe parameter
is attached to the logical device when we often have the information
which physical device we want to probe.
This patch adds a 'detect' callback for devices and adds a command
to detect devices and to list the devices which are actually detecable.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Since the info is device specific and not driver specific, attach
the callback to the device. This makes it possible to have a info
callback for a device which does not have a driver attached.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Some parameters may wish to provide some information about their
meaning or possible values. Provide an info callback for parameters.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
This is a massively stripped down pinctrl support. The upper API
consists of only of:
int pinctrl_select_state(struct device_d *dev, const char *state);
This is used to setup the pinmux for a device to a certain state.
This function normally does not need to be called manually. The
device core will setup the default state before probing a device.
The pinctrl core has the job of handling the devicetree. It parses
the pinctrl phandles for a device from devicetree, finds the correct
pinctrl device and calls its set_state callback with the pinctrl
setup device node.
The simplicity of this pinctrl framework comes from the fact that
we:
- Limit usage to devicetree only for now. For non devicetree use the
old legacy SoC specific APIs still can be used.
- Do not parse the devicetree into internal data structures which
are used by the drivers later. This adds the overhead that we
may parse the devicetree multiple times for more dynamic setups,
but on the other hand we do not need to parse devices from the
devicetree we don't use in barebox
- Do not detect resource conflicts. Since the framework mainly is
a devicetree parser this would be hard to implement. It should
be easy for board maintainers to avoid resource conflicts though.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
struct bus_type contains an embedded struct device_d which is quite
a big structure. Dynamically allocate this instead to save the space
in the binary.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
dev_add_child is a very unsafe function. If called multiple times
it allows setting the same device to different parents thus corrupting
the siblings list. This happens regularly since:
| commit c2e568d19c
| Author: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
| Date: Sat Nov 3 16:11:05 2012 +0100
|
| bus: add bus device
|
| automatically add it as parent of any bus device if none already specified
|
| we have now a nice output per bus
If for example a FATfs is mounted this nice output per bus often ends with:
> `---- fat0
> `---- 0
> `---- 0x86f0000087020031-0x86f000410df27124: /dev/<NULL>
> `---- sram00
> `---- 0x00000000-0xffffffffffffffff: /dev/<NULL>
> `---- 0x00000000-0xffffffffffffffff: /dev/<NULL>
> unable to handle NULL pointer dereference at address 0x0000000c
> pc : [<87f08a20>] lr : [<87f08a04>]
> sp : 86eff8c0 ip : 87f3fbde fp : ffffffff
> r10: ffffffff r9 : 00000000 r8 : 00000003
> r7 : 86f075b8 r6 : 00000002 r5 : ffffffec r4 : 86f07544
> r3 : 00000000 r2 : 43f900b4 r1 : 00000020 r0 : 00000005
> Flags: Nzcv IRQs off FIQs off Mode SVC_32
> [<87f08a20>] (do_devinfo_subtree+0x90/0x130) from [<87f08a90>] (do_devinfo_subtree+0x100/0x130)
>
> [<87f3e070>] (unwind_backtrace+0x0/0x90) from [<87f28514>] (panic+0x28/0x3c)
> [<87f28514>] (panic+0x28/0x3c) from [<87f3e4b8>] (do_exception+0x10/0x14)
> [<87f3e4b8>] (do_exception+0x10/0x14) from [<87f3e544>] (do_data_abort+0x2c/0x38)
> [<87f3e544>] (do_data_abort+0x2c/0x38) from [<87f3e268>] (data_abort+0x48/0x60)
This patch fixes this by adding a device to its parents children list in
register_device so that dev_add_child is no longer needed. This function
is removed from the tree. Now callers of register_device have to clearly
set the parent *before* registering a device.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Reported-by: Jan Lübbe <jlu@pengutronix.de>
Andrey Smirnov
Sascha Hauer
Yegor Yefremov
Herve Codina
Antony Pavlov
Uwe Kleine-König
Sebastian Hesselbarth
Marc Kleine-Budde
Jean-Christophe PLAGNIOL-VILLARD
Michel Stam
Alexander Shiyan