Hi Stefano,
On 11/05/2014 02:32 PM, Stefano Babic wrote:
Hi Nikita,
On 29/10/2014 16:56, Nikita Kiryanov wrote:
Introduce board specific function board_preboot_os() to allow for board specific config before we boot.
Signed-off-by: Nikita Kiryanov nikita@compulab.co.il Cc: Igor Grinberg grinberg@compulab.co.il Cc: Stefano Babic sbabic@denx.de Cc: Tom Rini trini@ti.com Cc: Jeroen Hofstee jeroen@myspectrum.nl Cc: Otavio Salvador otavio@ossystems.com.br
Changes in V2:
- Added board_preboot_os to bootm.h
- Split cm_fx6 stuff into a separate patch
common/bootm_os.c | 7 +++++++ include/bootm.h | 1 + 2 files changed, 8 insertions(+)
There is something that does not convince me. Really, the general statement should be to turn all devices off before booting the kernel, because this is what Linux expects. That said, this could be later solved with dm, because we will can iterate through all drivers and call a shutdown function.
Currently, "preboot" functions are turning off the hardware, so the name is already misleading. And I see that arch_preboot_os() was also convinced (because it is weak) to become a board_preboot, for example here: board/renesas/koelsch/koelsch.c (and in other renesas implementation, too).
I have some concerns adding a new weak function, that enables some further hooks inside board code. I find it not well scalable. In your case, you need such as sata_shutdown(), and it should be responsibility of the general sata code to call something specific for the board, if necessary.
In current code, bootm_disable_interrupts() calls also usb_stop() and eth_halt(), - nothing to do with the name of the function. IMHO it should be better to move the code to shutdown interface inside boot_selected_os(), or having a common function "disable_hardware()" that calls usb_stop(), eth_halt() and then, why not, a sata_stop().
There is also in arch/arm/imx-common/cpu.c a derived implementation for arch_preboot_os(). Really it has nothing to do with arch, as I can see: it shuts down only the IPU. But it is another hook, and IMHO it is better stopping sata here as adding a hidden callback.
You make some good points. I'll switch to using arch_preboot_os until there's a better way of doing this.