From 7d79cee2c6540ea64dd917a14e2fd63d4ac3d3c0 Mon Sep 17 00:00:00 2001 From: Alexey Brodkin Date: Tue, 1 Aug 2017 12:58:47 +0300 Subject: ARCv2: PAE40: Explicitly set MSB counterpart of SLC region ops addresses It is necessary to explicitly set both SLC_AUX_RGN_START1 and SLC_AUX_RGN_END1 which hold MSB bits of the physical address correspondingly of region start and end otherwise SLC region operation is executed in unpredictable manner Without this patch, SLC flushes on HSDK (IOC disabled) were taking seconds. Cc: stable@vger.kernel.org #4.4+ Reported-by: Vladimir Kondratiev Signed-off-by: Alexey Brodkin Signed-off-by: Vineet Gupta [vgupta: PAR40 regs only written if PAE40 exist] --- arch/arc/include/asm/cache.h | 2 ++ 1 file changed, 2 insertions(+) (limited to 'arch/arc/include') diff --git a/arch/arc/include/asm/cache.h b/arch/arc/include/asm/cache.h index 19ebddffb279..02fd1cece6ef 100644 --- a/arch/arc/include/asm/cache.h +++ b/arch/arc/include/asm/cache.h @@ -96,7 +96,9 @@ extern unsigned long perip_base, perip_end; #define ARC_REG_SLC_FLUSH 0x904 #define ARC_REG_SLC_INVALIDATE 0x905 #define ARC_REG_SLC_RGN_START 0x914 +#define ARC_REG_SLC_RGN_START1 0x915 #define ARC_REG_SLC_RGN_END 0x916 +#define ARC_REG_SLC_RGN_END1 0x917 /* Bit val in SLC_CONTROL */ #define SLC_CTRL_DIS 0x001 -- cgit v1.2.3 From b5ddb6d54729d814356937572d6c9b599f10c29f Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Thu, 3 Aug 2017 17:45:44 +0530 Subject: ARCv2: PAE40: set MSB even if !CONFIG_ARC_HAS_PAE40 but PAE exists in SoC PAE40 confiuration in hardware extends some of the address registers for TLB/cache ops to 2 words. So far kernel was NOT setting the higher word if feature was not enabled in software which is wrong. Those need to be set to 0 in such case. Normally this would be done in the cache flush / tlb ops, however since these registers only exist conditionally, this would have to be conditional to a flag being set on boot which is expensive/ugly - specially for the more common case of PAE exists but not in use. Optimize that by zero'ing them once at boot - nobody will write to them afterwards Cc: stable@vger.kernel.org #4.4+ Signed-off-by: Vineet Gupta --- arch/arc/include/asm/mmu.h | 2 ++ arch/arc/mm/cache.c | 34 ++++++++++++++++++++++++++++------ arch/arc/mm/tlb.c | 12 +++++++++++- 3 files changed, 41 insertions(+), 7 deletions(-) (limited to 'arch/arc/include') diff --git a/arch/arc/include/asm/mmu.h b/arch/arc/include/asm/mmu.h index db7319e9b506..efb79fafff1d 100644 --- a/arch/arc/include/asm/mmu.h +++ b/arch/arc/include/asm/mmu.h @@ -94,6 +94,8 @@ static inline int is_pae40_enabled(void) return IS_ENABLED(CONFIG_ARC_HAS_PAE40); } +extern int pae40_exist_but_not_enab(void); + #endif /* !__ASSEMBLY__ */ #endif diff --git a/arch/arc/mm/cache.c b/arch/arc/mm/cache.c index 874913b3e826..7db283b46ebd 100644 --- a/arch/arc/mm/cache.c +++ b/arch/arc/mm/cache.c @@ -1123,6 +1123,13 @@ noinline void __init arc_ioc_setup(void) __dc_enable(); } +/* + * Cache related boot time checks/setups only needed on master CPU: + * - Geometry checks (kernel build and hardware agree: e.g. L1_CACHE_BYTES) + * Assume SMP only, so all cores will have same cache config. A check on + * one core suffices for all + * - IOC setup / dma callbacks only need to be done once + */ void __init arc_cache_init_master(void) { unsigned int __maybe_unused cpu = smp_processor_id(); @@ -1202,12 +1209,27 @@ void __ref arc_cache_init(void) printk(arc_cache_mumbojumbo(0, str, sizeof(str))); - /* - * Only master CPU needs to execute rest of function: - * - Assume SMP so all cores will have same cache config so - * any geomtry checks will be same for all - * - IOC setup / dma callbacks only need to be setup once - */ if (!cpu) arc_cache_init_master(); + + /* + * In PAE regime, TLB and cache maintenance ops take wider addresses + * And even if PAE is not enabled in kernel, the upper 32-bits still need + * to be zeroed to keep the ops sane. + * As an optimization for more common !PAE enabled case, zero them out + * once at init, rather than checking/setting to 0 for every runtime op + */ + if (is_isa_arcv2() && pae40_exist_but_not_enab()) { + + if (IS_ENABLED(CONFIG_ARC_HAS_ICACHE)) + write_aux_reg(ARC_REG_IC_PTAG_HI, 0); + + if (IS_ENABLED(CONFIG_ARC_HAS_DCACHE)) + write_aux_reg(ARC_REG_DC_PTAG_HI, 0); + + if (l2_line_sz) { + write_aux_reg(ARC_REG_SLC_RGN_END1, 0); + write_aux_reg(ARC_REG_SLC_RGN_START1, 0); + } + } } diff --git a/arch/arc/mm/tlb.c b/arch/arc/mm/tlb.c index d0126fdfe2d8..b181f3ee38aa 100644 --- a/arch/arc/mm/tlb.c +++ b/arch/arc/mm/tlb.c @@ -104,6 +104,8 @@ /* A copy of the ASID from the PID reg is kept in asid_cache */ DEFINE_PER_CPU(unsigned int, asid_cache) = MM_CTXT_FIRST_CYCLE; +static int __read_mostly pae_exists; + /* * Utility Routine to erase a J-TLB entry * Caller needs to setup Index Reg (manually or via getIndex) @@ -784,7 +786,7 @@ void read_decode_mmu_bcr(void) mmu->u_dtlb = mmu4->u_dtlb * 4; mmu->u_itlb = mmu4->u_itlb * 4; mmu->sasid = mmu4->sasid; - mmu->pae = mmu4->pae; + pae_exists = mmu->pae = mmu4->pae; } } @@ -809,6 +811,11 @@ char *arc_mmu_mumbojumbo(int cpu_id, char *buf, int len) return buf; } +int pae40_exist_but_not_enab(void) +{ + return pae_exists && !is_pae40_enabled(); +} + void arc_mmu_init(void) { char str[256]; @@ -859,6 +866,9 @@ void arc_mmu_init(void) /* swapper_pg_dir is the pgd for the kernel, used by vmalloc */ write_aux_reg(ARC_REG_SCRATCH_DATA0, swapper_pg_dir); #endif + + if (pae40_exist_but_not_enab()) + write_aux_reg(ARC_REG_TLBPD1HI, 0); } /* -- cgit v1.2.3