[PATCH v2 2/3] watchdog: rti_wdt: Add support for loading firmware

Mon Jun 29 09:44:43 CEST 2020

On 29.06.20 06:54, Lokesh Vutla wrote:
> 
> 
> On 29/06/20 10:20 am, Jan Kiszka wrote:
>> On 29.06.20 04:26, Lokesh Vutla wrote:
>>> +Tom
>>>
>>> On 23/06/20 8:11 pm, Jan Kiszka wrote:
>>>> On 23.06.20 14:37, Jan Kiszka wrote:
>>>>> On 23.06.20 13:50, Lokesh Vutla wrote:
>>>>>>
>>>>>>
>>>>>> On 23/06/20 4:45 pm, Jan Kiszka wrote:
>>>>>>> From: Jan Kiszka <jan.kiszka at siemens.com>
>>>>>>>
>>>>>>> To avoid the need of extra boot scripting on AM65x for loading a
>>>>>>> watchdog firmware, add the required rproc init and loading logic for the
>>>>>>> first R5F core to the watchdog start handler. The firmware itself is
>>>>>>> embedded into U-Boot binary.
>>>>>>>
>>>>>>> One possible firmware source is https://github.com/siemens/k3-rti-wdt.
>>>>>>>
>>>>>>> Signed-off-by: Jan Kiszka <jan.kiszka at siemens.com>
>>>>>>> ---
>>>>>>>     drivers/watchdog/Kconfig      | 20 ++++++++++++++++++++
>>>>>>>     drivers/watchdog/Makefile     |  3 +++
>>>>>>>     drivers/watchdog/rti_wdt.c    | 24 ++++++++++++++++++++++++
>>>>>>>     drivers/watchdog/rti_wdt_fw.S | 20 ++++++++++++++++++++
>>>>>>>     4 files changed, 67 insertions(+)
>>>>>>>     create mode 100644 drivers/watchdog/rti_wdt_fw.S
>>>>>>>
>>>>>>> diff --git a/drivers/watchdog/Kconfig b/drivers/watchdog/Kconfig
>>>>>>> index ee99bd2af1..fd6ab9a85b 100644
>>>>>>> --- a/drivers/watchdog/Kconfig
>>>>>>> +++ b/drivers/watchdog/Kconfig
>>>>>>> @@ -162,6 +162,26 @@ config WDT_K3_RTI
>>>>>>>           Say Y here if you want to include support for the K3 watchdog
>>>>>>>           timer (RTI module) available in the K3 generation of processors.
>>>>>>> +if WDT_K3_RTI
>>>>>>> +
>>>>>>> +config WDT_K3_RTI_LOAD_FW
>>>>>>> +    bool "Load watchdog firmware"
>>>>>>> +    depends on REMOTEPROC
>>>>>>> +    help
>>>>>>> +      Automatically load the specified firmware image into the MCU R5F
>>>>>>> +      core 0. On the AM65x, this firmware is supposed to handle the expiry
>>>>>>> +      of the watchdog timer, typically by resetting the system.
>>>>>>> +
>>>>>>> +config WDT_K3_RTI_FW_FILE
>>>>>>> +    string "Watchdog firmware image file"
>>>>>>> +    default "k3-rti-wdt.fw"
>>>>>>> +    depends on WDT_K3_RTI_LOAD_FW
>>>>>>> +    help
>>>>>>> +      Firmware image to be embedded into U-Boot and loaded on watchdog
>>>>>>> +      start.
>>>>>>> +
>>>>>>> +endif
>>>>>>> +
>>>>>>>     config WDT_SANDBOX
>>>>>>>         bool "Enable Watchdog Timer support for Sandbox"
>>>>>>>         depends on SANDBOX && WDT
>>>>>>> diff --git a/drivers/watchdog/Makefile b/drivers/watchdog/Makefile
>>>>>>> index 16bdbf4970..bf58684875 100644
>>>>>>> --- a/drivers/watchdog/Makefile
>>>>>>> +++ b/drivers/watchdog/Makefile
>>>>>>> @@ -28,7 +28,10 @@ obj-$(CONFIG_WDT_MT7621) += mt7621_wdt.o
>>>>>>>     obj-$(CONFIG_WDT_MTK) += mtk_wdt.o
>>>>>>>     obj-$(CONFIG_WDT_OMAP3) += omap_wdt.o
>>>>>>>     obj-$(CONFIG_WDT_K3_RTI) += rti_wdt.o
>>>>>>> +obj-$(CONFIG_WDT_K3_RTI_LOAD_FW) += rti_wdt_fw.o
>>>>>>>     obj-$(CONFIG_WDT_SP805) += sp805_wdt.o
>>>>>>>     obj-$(CONFIG_WDT_STM32MP) += stm32mp_wdt.o
>>>>>>>     obj-$(CONFIG_WDT_TANGIER) += tangier_wdt.o
>>>>>>>     obj-$(CONFIG_WDT_XILINX) += xilinx_wwdt.o
>>>>>>> +
>>>>>>> +$(obj)/rti_wdt_fw.o: $(shell readlink -f $(CONFIG_WDT_K3_RTI_FW_FILE)) FORCE
>>>>>>> diff --git a/drivers/watchdog/rti_wdt.c b/drivers/watchdog/rti_wdt.c
>>>>>>> index ebe29c7409..38e82a6b6b 100644
>>>>>>> --- a/drivers/watchdog/rti_wdt.c
>>>>>>> +++ b/drivers/watchdog/rti_wdt.c
>>>>>>> @@ -14,6 +14,7 @@
>>>>>>>     #include <power-domain.h>
>>>>>>>     #include <wdt.h>
>>>>>>>     #include <asm/io.h>
>>>>>>> +#include <remoteproc.h>
>>>>>>>     /* Timer register set definition */
>>>>>>>     #define RTIDWDCTRL        0x90
>>>>>>> @@ -37,11 +38,16 @@
>>>>>>>     #define WDT_PRELOAD_MAX        0xfff
>>>>>>> +#define RTI_PROC_ID        0
>>>>>>
>>>>>> Can we get the rproc id from DT?
>>>>>
>>>>> You mean, resolve "mcu_r5fss0_core0" to the ID? Doable.
>>>>>
>>>>
>>>> I'm now probing for the first instance of "ti,am654-r5f" compatible. That
>>>> excludes usage for the J721E for now, but that one is fine without firmware -
>>>> provided there is way to prevent power-down for RTI watchdog otherwise...
>>>
>>> I was more of thinking to have a DT property to reference R5-core. But anyways,
>>> the property is not present in kernel. I am okay with this for now.
>>>
>>>>
>>>>>>
>>>>>>> +
>>>>>>>     struct rti_wdt_priv {
>>>>>>>         phys_addr_t regs;
>>>>>>>         unsigned int clk_khz;
>>>>>>>     };
>>>>>>> +extern const u32 rti_wdt_fw[];
>>>>>>> +extern const int rti_wdt_fw_size;
>>>>>>
>>>>>> FIT is the preferred way of packing images in U-Boot. Can you try using it?
>>>>>
>>>>> How does that work? Some example for me?
>>>>>
>>>>
>>>> If you happen to refer to fs-loader: That does not target OSPI, our primary use
>>>> case.
>>>
>>> No. I was thinking to pack the image in FIT along with ATF and OPTEE like in
>>> k3_fit_atf.sh and register a U_BOOT_FIT_LOADABLE_HANDLER for installing the
>>> firmware.
>>
>> In case of our board, that image is not processed by U-Boot (only generated).
>> Also, processing it from the R5 loader would imply putting remoteproc support
>> into that level, duplicating what we have in A53 U-Boot.
>>
>>>
>>>>
>>>>> What benefit would that bring? There are other users of this pattern, e.g.
>>>>> board/xilinx/zynqmp/pm_cfg_obj.S.
>>>
>>> I didn't know U-Boot is accepting this. Tom, is this the preferred way for
>>> including firmware images?
>>>
>>>>>
>>>>
>>>> The only benefit of an alternative loading mechanism seems to be handling of
>>>> larger images from different sources. But that's what I would tackle via boot
>>>> scripting and, thus, without this feature. If you only have a few hundred bytes
>>>> to embed, like for k3-rti-wdt, you quickly have a lot of overhead with other
>>>> approaches.
>>>
>>> hmm.. Does the build break if firmware is not present?
>>
>> Yes, if you configure to include the firmware but specify an invalid path, the
>> build breaks.
> 
> So, how do we handle the travis CI/gitlab build without this firmware. I am not
> sure how xilinx is handling this. It should be the same case in the above
> scenario you pointed out.

You mean some all-yes-config? We could create a dummy image prior to 
running that build.

The Xilinx has its logic disabled when the file name is empty, which is 
the default. The linker should kick unused bits out then. I can try that 
as well.

Jan

-- 
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux