Hi Tom,
On Tue, Mar 12, 2019 at 03:42:24PM -0400, Tom Rini wrote:
On Tue, Mar 12, 2019 at 07:59:40PM +0100, Eugeniu Rosca wrote:
Hi Marek,
On Tue, Mar 5, 2019 at 4:41 AM Marek Vasut marek.vasut@gmail.com wrote: [..]
+.align 8 +.globl rcar_atf_boot_args +rcar_atf_boot_args:
.dword 0.dword 0.dword 0.dword 0+ENTRY(save_boot_params)
adr x8, rcar_atf_boot_argsstp x0, x1, [x8], #16stp x2, x3, [x8], #16b save_boot_params_ret+ENDPROC(save_boot_params)
What about relocating the function to C like in [1] and passing the 4 arguments to it from ASM like in [2]? This would allow to:
- reduce asm operations to minimum (mov and bl) and make code transparent.
- avoid custom globals and store the ATF information in some C-defined struct.
[1] https://github.com/ARM-software/arm-trusted-firmware/blob/c48d02bade88b07fa7... [2] https://github.com/ARM-software/arm-trusted-firmware/blob/c48d02bade88b07fa7...
This is super early in the boot process and I'm really not a fan in general of writing and calling C before we have things setup for C calls to actually work normally, even more so for small things that can be commented to be obvious as to what's going on.
I don't have a strong preference about that. It is just my experience that Renesas once rewrote the elaborate Bosch-contributed BL2 routine of enabling the cntfrq_el0 system counter from ASM [1] to C [2]. This is at the BL1-BL2 boundary, while we are still in EL3, so much earlier in the boot process. This implementation currently runs on the targets of our customers.
I am fine with any working solution. Thanks for commenting!
[1] https://github.com/renesas-rcar/arm-trusted-firmware/commit/e23524689abb637b... [2] https://github.com/renesas-rcar/arm-trusted-firmware/commit/867d84191319#dif...
-- Tom
Best regards, Eugeniu.