[U-Boot] [PATCH RESEND] armv8: LSCH2 early and final mmu needs matching NS attribute

[Huan Wang]

Hi, York,

> On 03/28/2016 02:16 PM, Ed Swarthout wrote:
> > When switching between the early and final mmu tables, the stack will
> > get corrupted if the Non-Secure attribute is different.  For ls1043a,
> > this issue is currently masked because flush_dcache_all is called
> > before the switch when CONFIG_SYS_DPAA_FMAN is defined.
> >
> > Signed-off-by: Ed Swarthout <Ed.Swarthout at nxp.com>
> > ---
> >
> > Resend to fix diff format.
> >
> > denx/master with CONFIG_SYS_DPAA_FMAN undefined:
> > Fixes:
> >
> > U-Boot 2016.03-00530-g1fee6de (Mar 28 2016 - 13:46:36 -0500)
> > SoC:  LS1043E (0x87920010)
> > ...
> > Detected UDIMM 18ASF1G72AZ-2G1A1
> > 4 GiB (DDR4, 32-bit, CL=11, ECC on)
> >        DDR Chip-Select Interleaving Mode: CS0+CS1 "Synchronous Abort"
> > handler, esr 0x8a000000
> > ELR:     deadbeefdeadbeef
> > LR:      deadbeefdeadbeef
> > x0 : 000000ff440c0400 x1 : 0000000000022518
> > x2 : 0000000000000040 x3 : 000000000000003f
> > x4 : 0000000000000004 x5 : 0000000000000001
> > x6 : 0000000900000000 x7 : 0000000000200000
> > x8 : 0000000000000015 x9 : 000000000000000c
> > x10: 0000000000000401 x11: 00000008ffe06000
> > x12: 00000000000001ff x13: 0000000040000000
> > x14: 0000000000200000 x15: 0000000000000001
> > x16: 0000000000000000 x17: 0000000000000002
> > x18: 00000000ffdd8d78 x19: deadbeefdeadbeef
> > x20: deadbeefdeadbeef x21: deadbeefdeadbeef
> > x22: deadbeefdeadbeef x23: deadbeefdeadbeef
> > x24: deadbeefdeadbeef x25: deadbeefdeadbeef
> > x26: deadbeefdeadbeef x27: 0000000000000000
> > x28: 00000000fff6ca90 x29: deadbeefdeadbeef
> >
> >
> >  arch/arm/include/asm/arch-fsl-layerscape/cpu.h | 12 ++++++++----
> >  1 file changed, 8 insertions(+), 4 deletions(-)
> >
> > diff --git a/arch/arm/include/asm/arch-fsl-layerscape/cpu.h
> > b/arch/arm/include/asm/arch-fsl-layerscape/cpu.h
> > index 42ca7df..1f5842a 100644
> > --- a/arch/arm/include/asm/arch-fsl-layerscape/cpu.h
> > +++ b/arch/arm/include/asm/arch-fsl-layerscape/cpu.h
> > @@ -159,9 +159,11 @@ static const struct sys_mmu_table
> early_mmu_table[] = {
> >  	{ CONFIG_SYS_FSL_IFC_BASE, CONFIG_SYS_FSL_IFC_BASE,
> >  	  CONFIG_SYS_FSL_IFC_SIZE, MT_DEVICE_NGNRNE, PTE_BLOCK_NON_SHARE },
> >  	{ CONFIG_SYS_FSL_DRAM_BASE1, CONFIG_SYS_FSL_DRAM_BASE1,
> > -	  CONFIG_SYS_FSL_DRAM_SIZE1, MT_NORMAL, PTE_BLOCK_OUTER_SHARE },
> > +	  CONFIG_SYS_FSL_DRAM_SIZE1, MT_NORMAL,
> > +	  PTE_BLOCK_OUTER_SHARE | PTE_BLOCK_NS },
> >  	{ CONFIG_SYS_FSL_DRAM_BASE2, CONFIG_SYS_FSL_DRAM_BASE2,
> > -	  CONFIG_SYS_FSL_DRAM_SIZE2, MT_NORMAL, PTE_BLOCK_OUTER_SHARE },
> > +	  CONFIG_SYS_FSL_DRAM_SIZE2, MT_NORMAL,
> > +	  PTE_BLOCK_OUTER_SHARE | PTE_BLOCK_NS },
> >  #endif
> >  };
> >
> > @@ -247,7 +249,8 @@ static const struct sys_mmu_table final_mmu_table[]
> = {
> >  	  CONFIG_SYS_FSL_QBMAN_SIZE, MT_DEVICE_NGNRNE,
> >  	  PTE_BLOCK_NON_SHARE | PTE_BLOCK_PXN | PTE_BLOCK_UXN },
> >  	{ CONFIG_SYS_FSL_DRAM_BASE2, CONFIG_SYS_FSL_DRAM_BASE2,
> > -	  CONFIG_SYS_FSL_DRAM_SIZE2, MT_NORMAL, PTE_BLOCK_OUTER_SHARE },
> > +	  CONFIG_SYS_FSL_DRAM_SIZE2, MT_NORMAL,
> > +	  PTE_BLOCK_OUTER_SHARE | PTE_BLOCK_NS },
> >  	{ CONFIG_SYS_PCIE1_PHYS_ADDR, CONFIG_SYS_PCIE1_PHYS_ADDR,
> >  	  CONFIG_SYS_PCIE1_PHYS_SIZE, MT_DEVICE_NGNRNE,
> >  	  PTE_BLOCK_NON_SHARE | PTE_BLOCK_PXN | PTE_BLOCK_UXN }, @@ -258,7
> > +261,8 @@ static const struct sys_mmu_table final_mmu_table[] = {
> >  	  CONFIG_SYS_PCIE3_PHYS_SIZE, MT_DEVICE_NGNRNE,
> >  	  PTE_BLOCK_NON_SHARE | PTE_BLOCK_PXN | PTE_BLOCK_UXN },
> >  	{ CONFIG_SYS_FSL_DRAM_BASE3, CONFIG_SYS_FSL_DRAM_BASE3,
> > -	  CONFIG_SYS_FSL_DRAM_SIZE3, MT_NORMAL, PTE_BLOCK_OUTER_SHARE },
> > +	  CONFIG_SYS_FSL_DRAM_SIZE3, MT_NORMAL,
> > +	  PTE_BLOCK_OUTER_SHARE | PTE_BLOCK_NS },
> >  #endif
> >  };
> >  #endif
> >
> 
> Alison/Qianyu,
> 
> This change looks right to me. Please see if if flushing can be removed.
> 
[Alison Wang] I verified flushing can be removed.


Best Regards,
Alison Wang