brcm47xx: add kernel 4.4 support
[openwrt/openwrt.git] / target / linux / brcm47xx / patches-4.4 / 159-cpu_fixes.patch
diff --git a/target/linux/brcm47xx/patches-4.4/159-cpu_fixes.patch b/target/linux/brcm47xx/patches-4.4/159-cpu_fixes.patch
new file mode 100644 (file)
index 0000000..50df04f
--- /dev/null
@@ -0,0 +1,391 @@
+--- a/arch/mips/include/asm/r4kcache.h
++++ b/arch/mips/include/asm/r4kcache.h
+@@ -25,6 +25,20 @@
+ extern void (*r4k_blast_dcache)(void);
+ extern void (*r4k_blast_icache)(void);
++#ifdef CONFIG_BCM47XX
++#include <asm/paccess.h>
++#include <linux/ssb/ssb.h>
++#define BCM4710_DUMMY_RREG() ((void) *((u8 *) KSEG1ADDR(SSB_ENUM_BASE)))
++
++#define BCM4710_FILL_TLB(addr) (*(volatile unsigned long *)(addr))
++#define BCM4710_PROTECTED_FILL_TLB(addr) ({ unsigned long x; get_dbe(x, (volatile unsigned long *)(addr)); })
++#else
++#define BCM4710_DUMMY_RREG()
++
++#define BCM4710_FILL_TLB(addr)
++#define BCM4710_PROTECTED_FILL_TLB(addr)
++#endif
++
+ /*
+  * This macro return a properly sign-extended address suitable as base address
+  * for indexed cache operations.  Two issues here:
+@@ -98,6 +112,7 @@ static inline void flush_icache_line_ind
+ static inline void flush_dcache_line_indexed(unsigned long addr)
+ {
+       __dflush_prologue
++      BCM4710_DUMMY_RREG();
+       cache_op(Index_Writeback_Inv_D, addr);
+       __dflush_epilogue
+ }
+@@ -125,6 +140,7 @@ static inline void flush_icache_line(uns
+ static inline void flush_dcache_line(unsigned long addr)
+ {
+       __dflush_prologue
++      BCM4710_DUMMY_RREG();
+       cache_op(Hit_Writeback_Inv_D, addr);
+       __dflush_epilogue
+ }
+@@ -132,6 +148,7 @@ static inline void flush_dcache_line(uns
+ static inline void invalidate_dcache_line(unsigned long addr)
+ {
+       __dflush_prologue
++      BCM4710_DUMMY_RREG();
+       cache_op(Hit_Invalidate_D, addr);
+       __dflush_epilogue
+ }
+@@ -187,6 +204,7 @@ static inline void protected_flush_icach
+ #ifdef CONFIG_EVA
+               protected_cachee_op(Hit_Invalidate_I, addr);
+ #else
++              BCM4710_DUMMY_RREG();
+               protected_cache_op(Hit_Invalidate_I, addr);
+ #endif
+               break;
+@@ -201,6 +219,7 @@ static inline void protected_flush_icach
+  */
+ static inline void protected_writeback_dcache_line(unsigned long addr)
+ {
++      BCM4710_DUMMY_RREG();
+ #ifdef CONFIG_EVA
+       protected_cachee_op(Hit_Writeback_Inv_D, addr);
+ #else
+@@ -554,8 +573,51 @@ static inline void invalidate_tcache_pag
+               : "r" (base),                                           \
+                 "i" (op));
++static inline void blast_dcache(void)
++{
++      unsigned long start = KSEG0;
++      unsigned long dcache_size = current_cpu_data.dcache.waysize * current_cpu_data.dcache.ways;
++      unsigned long end = (start + dcache_size);
++
++      do {
++              BCM4710_DUMMY_RREG();
++              cache_op(Index_Writeback_Inv_D, start);
++              start += current_cpu_data.dcache.linesz;
++      } while(start < end);
++}
++
++static inline void blast_dcache_page(unsigned long page)
++{
++      unsigned long start = page;
++      unsigned long end = start + PAGE_SIZE;
++
++      BCM4710_FILL_TLB(start);
++      do {
++              BCM4710_DUMMY_RREG();
++              cache_op(Hit_Writeback_Inv_D, start);
++              start += current_cpu_data.dcache.linesz;
++      } while(start < end);
++}
++
++static inline void blast_dcache_page_indexed(unsigned long page)
++{
++      unsigned long start = page;
++      unsigned long end = start + PAGE_SIZE;
++      unsigned long ws_inc = 1UL << current_cpu_data.dcache.waybit;
++      unsigned long ws_end = current_cpu_data.dcache.ways <<
++                             current_cpu_data.dcache.waybit;
++      unsigned long ws, addr;
++      for (ws = 0; ws < ws_end; ws += ws_inc) {
++              start = page + ws;
++              for (addr = start; addr < end; addr += current_cpu_data.dcache.linesz) {
++                      BCM4710_DUMMY_RREG();
++                      cache_op(Index_Writeback_Inv_D, addr);
++              }
++      }
++}
++
+ /* build blast_xxx, blast_xxx_page, blast_xxx_page_indexed */
+-#define __BUILD_BLAST_CACHE(pfx, desc, indexop, hitop, lsize, extra)  \
++#define __BUILD_BLAST_CACHE(pfx, desc, indexop, hitop, lsize, extra, war) \
+ static inline void extra##blast_##pfx##cache##lsize(void)             \
+ {                                                                     \
+       unsigned long start = INDEX_BASE;                               \
+@@ -567,6 +629,7 @@ static inline void extra##blast_##pfx##c
+                                                                       \
+       __##pfx##flush_prologue                                         \
+                                                                       \
++      war                                                             \
+       for (ws = 0; ws < ws_end; ws += ws_inc)                         \
+               for (addr = start; addr < end; addr += lsize * 32)      \
+                       cache##lsize##_unroll32(addr|ws, indexop);      \
+@@ -581,6 +644,7 @@ static inline void extra##blast_##pfx##c
+                                                                       \
+       __##pfx##flush_prologue                                         \
+                                                                       \
++      war                                                             \
+       do {                                                            \
+               cache##lsize##_unroll32(start, hitop);                  \
+               start += lsize * 32;                                    \
+@@ -599,6 +663,8 @@ static inline void extra##blast_##pfx##c
+                              current_cpu_data.desc.waybit;            \
+       unsigned long ws, addr;                                         \
+                                                                       \
++      war                                                             \
++                                                                      \
+       __##pfx##flush_prologue                                         \
+                                                                       \
+       for (ws = 0; ws < ws_end; ws += ws_inc)                         \
+@@ -608,26 +674,26 @@ static inline void extra##blast_##pfx##c
+       __##pfx##flush_epilogue                                         \
+ }
+-__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 16, )
+-__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 16, )
+-__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 16, )
+-__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 32, )
+-__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 32, )
+-__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I_Loongson2, 32, loongson2_)
+-__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 32, )
+-__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 64, )
+-__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 64, )
+-__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 64, )
+-__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 128, )
+-__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 128, )
+-__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 128, )
+-
+-__BUILD_BLAST_CACHE(inv_d, dcache, Index_Writeback_Inv_D, Hit_Invalidate_D, 16, )
+-__BUILD_BLAST_CACHE(inv_d, dcache, Index_Writeback_Inv_D, Hit_Invalidate_D, 32, )
+-__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 16, )
+-__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 32, )
+-__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 64, )
+-__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 128, )
++__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 16, , )
++__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 16, , BCM4710_FILL_TLB(start);)
++__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 16, , )
++__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 32, , )
++__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 32, , BCM4710_FILL_TLB(start);)
++__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I_Loongson2, 32, loongson2_, BCM4710_FILL_TLB(start);)
++__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 32, , )
++__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 64, , )
++__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 64, , BCM4710_FILL_TLB(start);)
++__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 64, , )
++__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 128, , )
++__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 128, , )
++__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 128, , )
++
++__BUILD_BLAST_CACHE(inv_d, dcache, Index_Writeback_Inv_D, Hit_Invalidate_D, 16, , )
++__BUILD_BLAST_CACHE(inv_d, dcache, Index_Writeback_Inv_D, Hit_Invalidate_D, 32, , )
++__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 16, , )
++__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 32, , )
++__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 64, , )
++__BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 128, , )
+ #define __BUILD_BLAST_USER_CACHE(pfx, desc, indexop, hitop, lsize) \
+ static inline void blast_##pfx##cache##lsize##_user_page(unsigned long page) \
+@@ -656,17 +722,19 @@ __BUILD_BLAST_USER_CACHE(d, dcache, Inde
+ __BUILD_BLAST_USER_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 64)
+ /* build blast_xxx_range, protected_blast_xxx_range */
+-#define __BUILD_BLAST_CACHE_RANGE(pfx, desc, hitop, prot, extra)      \
++#define __BUILD_BLAST_CACHE_RANGE(pfx, desc, hitop, prot, extra, war, war2)   \
+ static inline void prot##extra##blast_##pfx##cache##_range(unsigned long start, \
+                                                   unsigned long end)  \
+ {                                                                     \
+       unsigned long lsize = cpu_##desc##_line_size();                 \
+       unsigned long addr = start & ~(lsize - 1);                      \
+       unsigned long aend = (end - 1) & ~(lsize - 1);                  \
++      war                                                             \
+                                                                       \
+       __##pfx##flush_prologue                                         \
+                                                                       \
+       while (1) {                                                     \
++              war2                                                    \
+               prot##cache_op(hitop, addr);                            \
+               if (addr == aend)                                       \
+                       break;                                          \
+@@ -678,8 +746,8 @@ static inline void prot##extra##blast_##
+ #ifndef CONFIG_EVA
+-__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, protected_, )
+-__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I, protected_, )
++__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, protected_, , BCM4710_PROTECTED_FILL_TLB(addr); BCM4710_PROTECTED_FILL_TLB(aend);, BCM4710_DUMMY_RREG();)
++__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I, protected_, , , )
+ #else
+@@ -716,14 +784,14 @@ __BUILD_PROT_BLAST_CACHE_RANGE(d, dcache
+ __BUILD_PROT_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I)
+ #endif
+-__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, protected_, )
++__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, protected_, , , )
+ __BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I_Loongson2, \
+-      protected_, loongson2_)
+-__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, , )
+-__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I, , )
+-__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, , )
++      protected_, loongson2_, , )
++__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, , , BCM4710_FILL_TLB(addr); BCM4710_FILL_TLB(aend);, BCM4710_DUMMY_RREG();)
++__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I, , , , )
++__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, , , , )
+ /* blast_inv_dcache_range */
+-__BUILD_BLAST_CACHE_RANGE(inv_d, dcache, Hit_Invalidate_D, , )
+-__BUILD_BLAST_CACHE_RANGE(inv_s, scache, Hit_Invalidate_SD, , )
++__BUILD_BLAST_CACHE_RANGE(inv_d, dcache, Hit_Invalidate_D, , , , BCM4710_DUMMY_RREG();)
++__BUILD_BLAST_CACHE_RANGE(inv_s, scache, Hit_Invalidate_SD, , , , )
+ #endif /* _ASM_R4KCACHE_H */
+--- a/arch/mips/include/asm/stackframe.h
++++ b/arch/mips/include/asm/stackframe.h
+@@ -358,6 +358,10 @@
+               .macro  RESTORE_SP_AND_RET
+               LONG_L  sp, PT_R29(sp)
+               .set    arch=r4000
++#ifdef CONFIG_BCM47XX
++              nop
++              nop
++#endif
+               eret
+               .set    mips0
+               .endm
+--- a/arch/mips/kernel/genex.S
++++ b/arch/mips/kernel/genex.S
+@@ -32,6 +32,10 @@
+ NESTED(except_vec3_generic, 0, sp)
+       .set    push
+       .set    noat
++#ifdef CONFIG_BCM47XX
++      nop
++      nop
++#endif
+ #if R5432_CP0_INTERRUPT_WAR
+       mfc0    k0, CP0_INDEX
+ #endif
+--- a/arch/mips/mm/c-r4k.c
++++ b/arch/mips/mm/c-r4k.c
+@@ -39,6 +39,9 @@
+ #include <asm/dma-coherence.h>
+ #include <asm/mips-cm.h>
++/* For enabling BCM4710 cache workarounds */
++int bcm4710 = 0;
++
+ /*
+  * Special Variant of smp_call_function for use by cache functions:
+  *
+@@ -157,6 +160,9 @@ static void r4k_blast_dcache_user_page_s
+ {
+       unsigned long  dc_lsize = cpu_dcache_line_size();
++      if (bcm4710)
++              r4k_blast_dcache_page = blast_dcache_page;
++      else
+       if (dc_lsize == 0)
+               r4k_blast_dcache_user_page = (void *)cache_noop;
+       else if (dc_lsize == 16)
+@@ -175,6 +181,9 @@ static void r4k_blast_dcache_page_indexe
+ {
+       unsigned long dc_lsize = cpu_dcache_line_size();
++      if (bcm4710)
++              r4k_blast_dcache_page_indexed = blast_dcache_page_indexed;
++      else
+       if (dc_lsize == 0)
+               r4k_blast_dcache_page_indexed = (void *)cache_noop;
+       else if (dc_lsize == 16)
+@@ -194,6 +203,9 @@ static void r4k_blast_dcache_setup(void)
+ {
+       unsigned long dc_lsize = cpu_dcache_line_size();
++      if (bcm4710)
++              r4k_blast_dcache = blast_dcache;
++      else
+       if (dc_lsize == 0)
+               r4k_blast_dcache = (void *)cache_noop;
+       else if (dc_lsize == 16)
+@@ -793,6 +805,8 @@ static void local_r4k_flush_cache_sigtra
+       unsigned long addr = (unsigned long) arg;
+       R4600_HIT_CACHEOP_WAR_IMPL;
++      BCM4710_PROTECTED_FILL_TLB(addr);
++      BCM4710_PROTECTED_FILL_TLB(addr + 4);
+       if (dc_lsize)
+               protected_writeback_dcache_line(addr & ~(dc_lsize - 1));
+       if (!cpu_icache_snoops_remote_store && scache_size)
+@@ -1602,6 +1616,17 @@ static void coherency_setup(void)
+        * silly idea of putting something else there ...
+        */
+       switch (current_cpu_type()) {
++      case CPU_BMIPS3300:
++              {
++                      u32 cm;
++                      cm = read_c0_diag();
++                      /* Enable icache */
++                      cm |= (1 << 31);
++                      /* Enable dcache */
++                      cm |= (1 << 30);
++                      write_c0_diag(cm);
++              }
++              break;
+       case CPU_R4000PC:
+       case CPU_R4000SC:
+       case CPU_R4000MC:
+@@ -1648,6 +1673,15 @@ void r4k_cache_init(void)
+       extern void build_copy_page(void);
+       struct cpuinfo_mips *c = &current_cpu_data;
++      /* Check if special workarounds are required */
++#ifdef CONFIG_BCM47XX
++      if (current_cpu_data.cputype == CPU_BMIPS32 && (current_cpu_data.processor_id & 0xff) == 0) {
++              printk("Enabling BCM4710A0 cache workarounds.\n");
++              bcm4710 = 1;
++      } else
++#endif
++              bcm4710 = 0;
++
+       probe_pcache();
+       setup_scache();
+@@ -1717,7 +1751,15 @@ void r4k_cache_init(void)
+        */
+       local_r4k___flush_cache_all(NULL);
++#ifdef CONFIG_BCM47XX
++      {
++              static void (*_coherency_setup)(void);
++              _coherency_setup = (void (*)(void)) KSEG1ADDR(coherency_setup);
++              _coherency_setup();
++      }
++#else
+       coherency_setup();
++#endif
+       board_cache_error_setup = r4k_cache_error_setup;
+       /*
+--- a/arch/mips/mm/tlbex.c
++++ b/arch/mips/mm/tlbex.c
+@@ -1281,6 +1281,9 @@ static void build_r4000_tlb_refill_handl
+                       /* No need for uasm_i_nop */
+               }
++#ifdef CONFIG_BCM47XX
++              uasm_i_nop(&p);
++#endif
+ #ifdef CONFIG_64BIT
+               build_get_pmde64(&p, &l, &r, K0, K1); /* get pmd in K1 */
+ #else
+@@ -1868,6 +1871,9 @@ build_r4000_tlbchange_handler_head(u32 *
+ {
+       struct work_registers wr = build_get_work_registers(p);
++#ifdef CONFIG_BCM47XX
++      uasm_i_nop(p);
++#endif
+ #ifdef CONFIG_64BIT
+       build_get_pmde64(p, l, r, wr.r1, wr.r2); /* get pmd in ptr */
+ #else