[OpenBIOS] [PATCH] ppc: don't use virtual image stack for startup context

[Mark Cave-Ayland]

On 22/10/16 11:05, Mark Cave-Ayland wrote:

> Part of commit 070ea60 created a new stack within the virtual image for use by
> the startup context. Unfortunately the current implementations of virt_to_phys()
> as used by some drivers are incorrect which causes issues with some parts of
> OpenBIOS, notably the USB stack.
> 
> Revert back to using the initial stack in start.S which ensures a 1:1 mapping
> between virtual and physical addresses to avoid such problems.
> 
> Signed-off-by: Mark Cave-Ayland <mark.cave-ayland at ilande.co.uk>
> ---
>  arch/ppc/qemu/context.c |    1 -
>  arch/ppc/qemu/ldscript  |    6 +-----
>  arch/ppc/qemu/start.S   |    6 ++++++
>  include/arch/ppc/io.h   |    2 +-
>  4 files changed, 8 insertions(+), 7 deletions(-)
> 
> diff --git a/arch/ppc/qemu/context.c b/arch/ppc/qemu/context.c
> index 18459b8..65cd4e4 100644
> --- a/arch/ppc/qemu/context.c
> +++ b/arch/ppc/qemu/context.c
> @@ -41,7 +41,6 @@ void of_client_callback(void);
>   * to start us up.
>   */
>  static struct context main_ctx = {
> -    .sp = (unsigned long) &_estack - SAVE_SPACE,
>      .pc = (unsigned long) start_main,
>      .return_addr = (unsigned long) __exit_context,
>  };
> diff --git a/arch/ppc/qemu/ldscript b/arch/ppc/qemu/ldscript
> index 11ebf4b..8027b39 100644
> --- a/arch/ppc/qemu/ldscript
> +++ b/arch/ppc/qemu/ldscript
> @@ -51,11 +51,7 @@ SECTIONS
>  	*(.bss)
>  	*(.bss.*)
>  	*(COMMON)
> -	
> -	_stack = .;
> -	. += CSTACK_SIZE;
> -	. = ALIGN(16);
> -	_estack = .;
> +	_ebss = .;
>      }
>  
>      . = HRESET_ADDR;
> diff --git a/arch/ppc/qemu/start.S b/arch/ppc/qemu/start.S
> index 33ca1e0..fed1b55 100644
> --- a/arch/ppc/qemu/start.S
> +++ b/arch/ppc/qemu/start.S
> @@ -482,6 +482,12 @@ real_entry:
>  #endif
>  
>  	bl	BRANCH_LABEL(setup_mmu)
> +
> +	/* load stack pointer into context */
> +	LOAD_REG_IMMEDIATE(r4, __context)
> +	PPC_LL  r4, 0(r4)
> +	PPC_STL r1, (2 * ULONG_SIZE)(r4)
> +
>  	bl	BRANCH_LABEL(__switch_context_nosave)
>  1:	nop
>  	b	1b
> diff --git a/include/arch/ppc/io.h b/include/arch/ppc/io.h
> index 39c60d7..3449c5b 100644
> --- a/include/arch/ppc/io.h
> +++ b/include/arch/ppc/io.h
> @@ -6,7 +6,7 @@
>  #define NO_QEMU_PROTOS
>  #include "arch/common/fw_cfg.h"
>  
> -extern char _start, _end, _estack;
> +extern char _start, _end;
>  extern unsigned long virt_offset;
>  
>  #define phys_to_virt(phys) ((void *) ((unsigned long) (phys) - virt_offset))

I've given this a good test today, and it fixes the USB issue without
introducing any regressions on my PPC boot tests so I've applied this to
master.


ATB,

Mark.