powerpc SPL framework: Avoiding relocate_code
Hi,
SPL framework is used to support multi-stage booting. Where first
level boot loader is created via SPL having relocate_code() function. I am working on a Freescale's SoC which has less internal SRAM. I don't want to use relocate_code() as to support this function, I need to reduce SPL bin to SRAM/2 size.
is there way to avoid relocate_code function ?
I tried with below sequence, but it is not working for me :(
.globl relocate_coderelocate_code:
mr r1,r3 /* Set new stack pointer */
mr r9,r4 /* Save copy of Init Data pointer */
mr r10,r5 /* Save copy of Destination Address */ GET_GOT
#ifndef CONFIG_SPL_BUILD
--
--
#endif
.globl in_ramin_ram:
The reason is bss "variables" which are mapped to 0x0000_0000 onwards because "bss"section are mapped after 0xfffffffc in lds. They are not available during SPL execution. is there way to relocate bss section in the execution range of SPL?
Please advice.
Regards,
Prabhakar
Quoted 18 lines · Prabhakar Kushwaha
Hi,
SPL framework is used to support multi-stage booting. Where first
level boot loader is created via SPL havingrelocate_code()function. I am working on a Freescale's SoC which has less internal SRAM. I don't want to userelocate_code()as to support this function, I need to reduce SPL bin to SRAM/2 size.is there way to avoid
relocate_codefunction ?I tried with below sequence, but it is not working for me :(
.globl
relocate_coderelocate_code:mr r1,r3 /* Set new stack pointer */ mr r9,r4 /* Save copy of Init Data pointer */ mr r10,r5 /* Save copy of Destination Address */
Quoted 8 lines
GET_GOT
#ifndefCONFIG_SPL_BUILD--
--
#endif
.globlin_ramin_ram:
Well, you certainly don't want to disable it for all SPLs...
Quoted 4 lines
The reason is bss "variables" which are mapped to 0x0000_0000 onwards because "bss"section are mapped after
0xfffffffcin lds. They are not available during SPL execution. is there way to relocate bss section in the execution range of SPL?
Are you talking about a scenario in which the SPL is loaded into SRAM rather than (e.g.) the NAND buffer? In that case, why is U-Boot not linked at the actual SRAM address? No copy should be needed in that case, and the BSS will not be at zero.
If you are talking about a NAND buffer style boot (i.e. not a PBL-using chip), then you must relocate, so the NAND buffer can be reused for I/O (and in any case the SRAM is much larger than the NAND buffer on Freescale PPC chips that I'm familiar with).
-Scott
Quoted 22 lines · Scott Wood
Quoted 21 lines · Prabhakar Kushwaha
Hi,
SPL framework is used to support multi-stage booting. Where first
level boot loader is created via SPL havingrelocate_code()function. I am working on a Freescale's SoC which has less internal SRAM. I don't want to userelocate_code()as to support this function, I need to reduce SPL bin to SRAM/2 size.is there way to avoid
relocate_codefunction ?I tried with below sequence, but it is not working for me :(
.globl
relocate_coderelocate_code:mr r1,r3 /* Set new stack pointer */ mr r9,r4 /* Save copy of Init Data pointer */ mr r10,r5 /* Save copy of Destination Address */
GET_GOT
#ifndefCONFIG_SPL_BUILD
Quoted 6 lines
Quoted 5 lines
--
--
#endif
.globlin_ramin_ram:Well, you certainly don't want to disable it for all SPLs...
it should be disabled for those SPL which relocate itself in SRAM, for other no
Quoted 8 lines
Quoted 4 lines
The reason is bss "variables" which are mapped to 0x0000_0000 onwards because "bss"section are mapped after
0xfffffffcin lds. They are not available during SPL execution. is there way to relocate bss section in the execution range of SPL?Are you talking about a scenario in which the SPL is loaded into SRAM rather than (e.g.) the NAND buffer? In that case, why is U-Boot not linked at the actual SRAM address? No copy should be needed in that case, and the BSS will not be at zero.
I am linking SPL with the address of SRAM as I want resetvec at 0xfffffffc but still I am finding bss at 0x0
in u-boot-spl.lds we have :-
#ifdef CONFIG_SYS_MPC85XX_NO_RESETVEC
.bootpg ADDR(.text) - 0x1000 :
{
KEEP(*(.bootpg))
} :text = 0xffff
#else
.resetvec ADDR(.text) + RESET_VECTOR_OFFSET : {
KEEP(*(.resetvec))
} = 0xffff#endif
/*
* Make sure that the bss segment isn't linked at 0x0, otherwise its
* address won't be updated during relocation fixups.
*/
. |= 0x10;
. = ALIGN(4);
__bss_start = .;
.bss : {
*(.sbss*)
*(.bss*)
}
. = ALIGN(4);
__bss_end = .;from above lds, we put resetvec at 0xfffffffc And place BSS just after it. Hence at 0x0.
Ideally __bss_start should not be put after 0xfffffffc as it is making relocation a requirement.
Regards,
Prabhakar
Quoted 28 lines · Prabhakar Kushwaha
Quoted 27 lines · Scott Wood
Quoted 26 lines · Prabhakar Kushwaha
Hi,
SPL framework is used to support multi-stage booting. Where first
level boot loader is created via SPL havingrelocate_code()function. I am working on a Freescale's SoC which has less internal SRAM. I don't want to userelocate_code()as to support this function, I need to reduce SPL bin to SRAM/2 size.is there way to avoid
relocate_codefunction ?I tried with below sequence, but it is not working for me :(
.globl
relocate_coderelocate_code:mr r1,r3 /* Set new stack pointer */ mr r9,r4 /* Save copy of Init Data pointer */ mr r10,r5 /* Save copy of Destination Address */
GET_GOT
#ifndefCONFIG_SPL_BUILD--
--
#endif
.globlin_ramin_ram:
Quoted 4 lines
Quoted 1 line
Well, you certainly don't want to disable it for all SPLs...
it should be disabled for those SPL which relocate itself in SRAM, for other no
Hmm? Any SPL that does relocate itself would need this.
Quoted 11 lines
Quoted 8 lines
Quoted 4 lines
The reason is bss "variables" which are mapped to 0x0000_0000 onwards because "bss"section are mapped after
0xfffffffcin lds. They are not available during SPL execution. is there way to relocate bss section in the execution range of SPL?Are you talking about a scenario in which the SPL is loaded into SRAM rather than (e.g.) the NAND buffer? In that case, why is U-Boot not linked at the actual SRAM address? No copy should be needed in that case, and the BSS will not be at zero.
I am linking SPL with the address of SRAM as I want resetvec at
0xfffffffcbut still I am finding bss at0x0
What address do we start at when PBI loads something into SRAM?
-Scott