#ifndef _SPARC64_CPUDATA_H
#define _SPARC64_CPUDATA_H
+#include <asm/hypervisor.h>
+#include <asm/asi.h>
+
#ifndef __ASSEMBLY__
#include <linux/percpu.h>
unsigned int __softirq_pending; /* must be 1st, see rtrap.S */
unsigned int multiplier;
unsigned int counter;
- unsigned int idle_volume;
+ unsigned int __pad1;
unsigned long clock_tick; /* %tick's per second */
unsigned long udelay_val;
*/
struct thread_info;
struct trap_per_cpu {
-/* D-cache line 1 */
+/* D-cache line 1: Basic thread information, cpu and device mondo queues */
struct thread_info *thread;
unsigned long pgd_paddr;
- unsigned long __pad1[2];
+ unsigned long cpu_mondo_pa;
+ unsigned long dev_mondo_pa;
+
+/* D-cache line 2: Error Mondo Queue and kernel buffer pointers */
+ unsigned long resum_mondo_pa;
+ unsigned long resum_kernel_buf_pa;
+ unsigned long nonresum_mondo_pa;
+ unsigned long nonresum_kernel_buf_pa;
+
+/* Dcache lines 3, 4, 5, and 6: Hypervisor Fault Status */
+ struct hv_fault_status fault_info;
-/* D-cache line 2 */
+/* Dcache line 7: Physical addresses of CPU send mondo block and CPU list. */
+ unsigned long cpu_mondo_block_pa;
+ unsigned long cpu_list_pa;
+ unsigned long tsb_huge;
+ unsigned long tsb_huge_temp;
+
+/* Dcache line 8: Unused, needed to keep trap_block a power-of-2 in size. */
unsigned long __pad2[4];
} __attribute__((aligned(64)));
extern struct trap_per_cpu trap_block[NR_CPUS];
-extern void init_cur_cpu_trap(void);
+extern void init_cur_cpu_trap(struct thread_info *);
extern void setup_tba(void);
-#ifdef CONFIG_SMP
struct cpuid_patch_entry {
unsigned int addr;
unsigned int cheetah_safari[4];
unsigned int cheetah_jbus[4];
unsigned int starfire[4];
+ unsigned int sun4v[4];
};
extern struct cpuid_patch_entry __cpuid_patch, __cpuid_patch_end;
-#endif
+
+struct sun4v_1insn_patch_entry {
+ unsigned int addr;
+ unsigned int insn;
+};
+extern struct sun4v_1insn_patch_entry __sun4v_1insn_patch,
+ __sun4v_1insn_patch_end;
+
+struct sun4v_2insn_patch_entry {
+ unsigned int addr;
+ unsigned int insns[2];
+};
+extern struct sun4v_2insn_patch_entry __sun4v_2insn_patch,
+ __sun4v_2insn_patch_end;
#endif /* !(__ASSEMBLY__) */
-#define TRAP_PER_CPU_THREAD 0x00
-#define TRAP_PER_CPU_PGD_PADDR 0x08
+#define TRAP_PER_CPU_THREAD 0x00
+#define TRAP_PER_CPU_PGD_PADDR 0x08
+#define TRAP_PER_CPU_CPU_MONDO_PA 0x10
+#define TRAP_PER_CPU_DEV_MONDO_PA 0x18
+#define TRAP_PER_CPU_RESUM_MONDO_PA 0x20
+#define TRAP_PER_CPU_RESUM_KBUF_PA 0x28
+#define TRAP_PER_CPU_NONRESUM_MONDO_PA 0x30
+#define TRAP_PER_CPU_NONRESUM_KBUF_PA 0x38
+#define TRAP_PER_CPU_FAULT_INFO 0x40
+#define TRAP_PER_CPU_CPU_MONDO_BLOCK_PA 0xc0
+#define TRAP_PER_CPU_CPU_LIST_PA 0xc8
+#define TRAP_PER_CPU_TSB_HUGE 0xd0
+#define TRAP_PER_CPU_TSB_HUGE_TEMP 0xd8
-#define TRAP_BLOCK_SZ_SHIFT 6
+#define TRAP_BLOCK_SZ_SHIFT 8
-#ifdef CONFIG_SMP
+#include <asm/scratchpad.h>
#define __GET_CPUID(REG) \
/* Spitfire implementation (default). */ \
sllx REG, 9, REG; \
or REG, 0xd0, REG; \
lduwa [REG] ASI_PHYS_BYPASS_EC_E, REG;\
+ /* sun4v implementation. */ \
+ mov SCRATCHPAD_CPUID, REG; \
+ ldxa [REG] ASI_SCRATCHPAD, REG; \
+ nop; \
+ nop; \
.previous;
-/* Clobbers %g1, current address space PGD phys address into %g7. */
-#define TRAP_LOAD_PGD_PHYS \
- __GET_CPUID(%g1) \
- sethi %hi(trap_block), %g7; \
- sllx %g1, TRAP_BLOCK_SZ_SHIFT, %g1; \
- or %g7, %lo(trap_block), %g7; \
- add %g7, %g1, %g7; \
- ldx [%g7 + TRAP_PER_CPU_PGD_PADDR], %g7;
-
-/* Clobbers %g1, loads local processor's IRQ work area into %g6. */
-#define TRAP_LOAD_IRQ_WORK \
- __GET_CPUID(%g1) \
- sethi %hi(__irq_work), %g6; \
- sllx %g1, 6, %g1; \
- or %g6, %lo(__irq_work), %g6; \
- add %g6, %g1, %g6;
-
-/* Clobbers %g1, loads %g6 with current thread info pointer. */
-#define TRAP_LOAD_THREAD_REG \
- __GET_CPUID(%g1) \
- sethi %hi(trap_block), %g6; \
- sllx %g1, TRAP_BLOCK_SZ_SHIFT, %g1; \
- or %g6, %lo(trap_block), %g6; \
- ldx [%g6 + %g1], %g6;
-
-/* Given the current thread info pointer in %g6, load the per-cpu
- * area base of the current processor into %g5. REG1, REG2, and REG3 are
+#ifdef CONFIG_SMP
+
+#define TRAP_LOAD_TRAP_BLOCK(DEST, TMP) \
+ __GET_CPUID(TMP) \
+ sethi %hi(trap_block), DEST; \
+ sllx TMP, TRAP_BLOCK_SZ_SHIFT, TMP; \
+ or DEST, %lo(trap_block), DEST; \
+ add DEST, TMP, DEST; \
+
+/* Clobbers TMP, current address space PGD phys address into DEST. */
+#define TRAP_LOAD_PGD_PHYS(DEST, TMP) \
+ TRAP_LOAD_TRAP_BLOCK(DEST, TMP) \
+ ldx [DEST + TRAP_PER_CPU_PGD_PADDR], DEST;
+
+/* Clobbers TMP, loads local processor's IRQ work area into DEST. */
+#define TRAP_LOAD_IRQ_WORK(DEST, TMP) \
+ __GET_CPUID(TMP) \
+ sethi %hi(__irq_work), DEST; \
+ sllx TMP, 6, TMP; \
+ or DEST, %lo(__irq_work), DEST; \
+ add DEST, TMP, DEST;
+
+/* Clobbers TMP, loads DEST with current thread info pointer. */
+#define TRAP_LOAD_THREAD_REG(DEST, TMP) \
+ TRAP_LOAD_TRAP_BLOCK(DEST, TMP) \
+ ldx [DEST + TRAP_PER_CPU_THREAD], DEST;
+
+/* Given the current thread info pointer in THR, load the per-cpu
+ * area base of the current processor into DEST. REG1, REG2, and REG3 are
* clobbered.
*
- * You absolutely cannot use %g5 as a temporary in this code. The
+ * You absolutely cannot use DEST as a temporary in this code. The
* reason is that traps can happen during execution, and return from
- * trap will load the fully resolved %g5 per-cpu base. This can corrupt
+ * trap will load the fully resolved DEST per-cpu base. This can corrupt
* the calculations done by the macro mid-stream.
*/
-#define LOAD_PER_CPU_BASE(REG1, REG2, REG3) \
- ldub [%g6 + TI_CPU], REG1; \
+#define LOAD_PER_CPU_BASE(DEST, THR, REG1, REG2, REG3) \
+ ldub [THR + TI_CPU], REG1; \
sethi %hi(__per_cpu_shift), REG3; \
sethi %hi(__per_cpu_base), REG2; \
ldx [REG3 + %lo(__per_cpu_shift)], REG3; \
ldx [REG2 + %lo(__per_cpu_base)], REG2; \
sllx REG1, REG3, REG3; \
- add REG3, REG2, %g5;
+ add REG3, REG2, DEST;
#else
+#define TRAP_LOAD_TRAP_BLOCK(DEST, TMP) \
+ sethi %hi(trap_block), DEST; \
+ or DEST, %lo(trap_block), DEST; \
+
/* Uniprocessor versions, we know the cpuid is zero. */
-#define TRAP_LOAD_PGD_PHYS \
- sethi %hi(trap_block), %g7; \
- or %g7, %lo(trap_block), %g7; \
- ldx [%g7 + TRAP_PER_CPU_PGD_PADDR], %g7;
+#define TRAP_LOAD_PGD_PHYS(DEST, TMP) \
+ TRAP_LOAD_TRAP_BLOCK(DEST, TMP) \
+ ldx [DEST + TRAP_PER_CPU_PGD_PADDR], DEST;
-#define TRAP_LOAD_IRQ_WORK \
- sethi %hi(__irq_work), %g6; \
- or %g6, %lo(__irq_work), %g6;
+#define TRAP_LOAD_IRQ_WORK(DEST, TMP) \
+ sethi %hi(__irq_work), DEST; \
+ or DEST, %lo(__irq_work), DEST;
-#define TRAP_LOAD_THREAD_REG \
- sethi %hi(trap_block), %g6; \
- ldx [%g6 + %lo(trap_block)], %g6;
+#define TRAP_LOAD_THREAD_REG(DEST, TMP) \
+ TRAP_LOAD_TRAP_BLOCK(DEST, TMP) \
+ ldx [DEST + TRAP_PER_CPU_THREAD], DEST;
-/* No per-cpu areas on uniprocessor, so no need to load %g5. */
-#define LOAD_PER_CPU_BASE(REG1, REG2, REG3)
+/* No per-cpu areas on uniprocessor, so no need to load DEST. */
+#define LOAD_PER_CPU_BASE(DEST, THR, REG1, REG2, REG3)
#endif /* !(CONFIG_SMP) */