[PATCH v2 01/21] mips: add support to restore exception vector base before booting linux

Weijie Gao weijie.gao at mediatek.com
Sun Jan 19 05:18:22 CET 2020


On Fri, 2020-01-17 at 13:15 +0100, Daniel Schwierzeck wrote:
> 
> Am 17.01.20 um 08:45 schrieb Weijie Gao:
> > In U-Boot the exception vector base will be moved to top of memory, to be
> > used to display register dump when exception occurs.
> > 
> > But some old linux kernel does not honor the base set in CP0_EBASE. A
> > modified exception vector base will cause kernel crash.
> > 
> > This patch adds an option to enable reset exception vector base to its
> > previous value, or a user configured value before booting linux kernel.
> > 
> > Signed-off-by: Weijie Gao <weijie.gao at mediatek.com>
> > ---
> > Changes since v1:
> >   Moved core operations to trap_restore() in arch/mips/lib/traps.c.
> >   Save previous ebase instead of using 0x80000000 directly.
> >   Add options to use customized ebase.
> > ---
> >  arch/mips/Kconfig                   | 30 +++++++++++++++++++++++++++++
> >  arch/mips/include/asm/u-boot-mips.h |  2 ++
> >  arch/mips/lib/bootm.c               |  3 +++
> >  arch/mips/lib/traps.c               | 19 ++++++++++++++++++
> >  4 files changed, 54 insertions(+)
> > 
> > diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
> > index a3ae603044..5e20feeefb 100644
> > --- a/arch/mips/Kconfig
> > +++ b/arch/mips/Kconfig
> > @@ -287,6 +287,36 @@ config MIPS_RELOCATION_TABLE_SIZE
> >  
> >  	  If unsure, leave at the default value.
> >  
> > +config RESTORE_EXCEPTION_VECTOR_BASE
> > +	bool "Restore exception vector base before booting linux kernel"
> > +	default n
> > +	help
> > +	  In U-Boot the exception vector base will be moved to top of memory,
> > +	  to be used to display register dump when exception occurs.
> > +	  But some old linux kernel does not honor the base set in CP0_EBASE.
> > +	  A modified exception vector base will cause kernel crash.
> > +
> > +	  This option will restore the exception vector base to its previous
> > +	  value.
> > +
> > +	  If unsure, say N.
> > +
> > +config OVERRIDE_EXCEPTION_VECTOR_BASE
> > +	bool "Override the exception vector base to be restored"
> > +	depends on RESTORE_EXCEPTION_VECTOR_BASE
> > +	default n
> > +	help
> > +	  Enable this option if you want to use a different exception vector
> > +	  base rather than the previously saved one.
> > +
> > +config NEW_EXCEPTION_VECTOR_BASE
> > +	hex "New exception vector base"
> > +	depends on OVERRIDE_EXCEPTION_VECTOR_BASE
> > +	range 0x80000000 0xbffff000
> > +	default 0x80000000
> > +	help
> > +	  The exception vector base to be restored before booting linux kernel
> > +
> >  endmenu
> >  
> >  menu "OS boot interface"
> > diff --git a/arch/mips/include/asm/u-boot-mips.h b/arch/mips/include/asm/u-boot-mips.h
> > index 88438b9576..8b37cc4029 100644
> > --- a/arch/mips/include/asm/u-boot-mips.h
> > +++ b/arch/mips/include/asm/u-boot-mips.h
> > @@ -9,4 +9,6 @@ void except_vec_ejtag_debug(void);
> >  
> >  int arch_misc_init(void);
> >  
> > +void trap_restore(void);
> > +
> >  #endif /* _U_BOOT_MIPS_H_ */
> > diff --git a/arch/mips/lib/bootm.c b/arch/mips/lib/bootm.c
> > index 8c0d7672f2..f1db6d23b8 100644
> > --- a/arch/mips/lib/bootm.c
> > +++ b/arch/mips/lib/bootm.c
> > @@ -294,6 +294,9 @@ static void boot_jump_linux(bootm_headers_t *images)
> >  	bootstage_report();
> >  #endif
> >  
> > +	if (CONFIG_IS_ENABLED(RESTORE_EXCEPTION_VECTOR_BASE))
> > +		trap_restore();
> > +
> >  	if (images->ft_len)
> >  		kernel(-2, (ulong)images->ft_addr, 0, 0);
> >  	else
> > diff --git a/arch/mips/lib/traps.c b/arch/mips/lib/traps.c
> > index 6fe8ebd16b..20f45fc4ed 100644
> > --- a/arch/mips/lib/traps.c
> > +++ b/arch/mips/lib/traps.c
> > @@ -19,6 +19,8 @@
> >  
> >  DECLARE_GLOBAL_DATA_PTR;
> >  
> > +static unsigned long saved_ebase;
> > +
> >  static void show_regs(const struct pt_regs *regs)
> >  {
> >  	const int field = 2 * sizeof(unsigned long);
> > @@ -101,7 +103,24 @@ void trap_init(ulong reloc_addr)
> >  	set_handler(0x180, &except_vec3_generic, 0x80);
> >  	set_handler(0x280, &except_vec_ejtag_debug, 0x80);
> >  
> > +	saved_ebase = read_c0_ebase() & 0xfffff000;
> > +
> >  	write_c0_ebase(ebase);
> >  	clear_c0_status(ST0_BEV);
> >  	execution_hazard_barrier();
> >  }
> > +
> > +void trap_restore(void)
> > +{
> > +	set_c0_status(ST0_BEV);
> > +	execution_hazard_barrier();
> > +
> > +#ifdef CONFIG_OVERRIDE_EXCEPTION_VECTOR_BASE
> > +	write_c0_ebase(CONFIG_NEW_EXCEPTION_VECTOR_BASE & 0xfffff000);
> > +#else
> > +	write_c0_ebase(saved_ebase);
> > +#endif
> > +
> > +	clear_c0_status(ST0_BEV);
> > +	execution_hazard_barrier();
> > +}
> > 
> 
> looks actually good now, thanks. But now I'm thinking that it should be
> enough to simply set ST0_BEV to use the CPU's default EBase. Restoring
> the original EBase and clearing ST0_BEV again seems redundant. How about
> this?
> 
> void trap_restore(void)
> {
> 	set_c0_status(ST0_BEV);
> 	execution_hazard_barrier();
> 
> #ifdef CONFIG_OVERRIDE_EXCEPTION_VECTOR_BASE
> 	write_c0_ebase(CONFIG_NEW_EXCEPTION_VECTOR_BASE & 0xfffff000);
> 	clear_c0_status(ST0_BEV);
> 	execution_hazard_barrier();
> #endif
> }
> 

Actually it's a little bit redundant. saved_ebase is reserved for the
case that u-boot acts as a secondary bootloader, and the first stage
bootloader changes the ebase.

Currently it seems there is no such case. I'll simplify this patch.



More information about the U-Boot mailing list