1 #ifndef _SPARC_TRAP_BLOCK_H
2 #define _SPARC_TRAP_BLOCK_H
3 
4 #include <asm/hypervisor.h>
5 #include <asm/asi.h>
6 
7 #ifndef __ASSEMBLY__
8 
9 /* Trap handling code needs to get at a few critical values upon
10  * trap entry and to process TSB misses.  These cannot be in the
11  * per_cpu() area as we really need to lock them into the TLB and
12  * thus make them part of the main kernel image.  As a result we
13  * try to make this as small as possible.
14  *
15  * This is padded out and aligned to 64-bytes to avoid false sharing
16  * on SMP.
17  */
18 
19 /* If you modify the size of this structure, please update
20  * TRAP_BLOCK_SZ_SHIFT below.
21  */
22 struct thread_info;
23 struct trap_per_cpu {
24 /* D-cache line 1: Basic thread information, cpu and device mondo queues */
25 	struct thread_info	*thread;
26 	unsigned long		pgd_paddr;
27 	unsigned long		cpu_mondo_pa;
28 	unsigned long		dev_mondo_pa;
29 
30 /* D-cache line 2: Error Mondo Queue and kernel buffer pointers */
31 	unsigned long		resum_mondo_pa;
32 	unsigned long		resum_kernel_buf_pa;
33 	unsigned long		nonresum_mondo_pa;
34 	unsigned long		nonresum_kernel_buf_pa;
35 
36 /* Dcache lines 3, 4, 5, and 6: Hypervisor Fault Status */
37 	struct hv_fault_status	fault_info;
38 
39 /* Dcache line 7: Physical addresses of CPU send mondo block and CPU list.  */
40 	unsigned long		cpu_mondo_block_pa;
41 	unsigned long		cpu_list_pa;
42 	unsigned long		tsb_huge;
43 	unsigned long		tsb_huge_temp;
44 
45 /* Dcache line 8: IRQ work list, and keep trap_block a power-of-2 in size.  */
46 	unsigned long		irq_worklist_pa;
47 	unsigned int		cpu_mondo_qmask;
48 	unsigned int		dev_mondo_qmask;
49 	unsigned int		resum_qmask;
50 	unsigned int		nonresum_qmask;
51 	unsigned long		__per_cpu_base;
52 } __attribute__((aligned(64)));
53 extern struct trap_per_cpu trap_block[NR_CPUS];
54 void init_cur_cpu_trap(struct thread_info *);
55 void setup_tba(void);
56 extern int ncpus_probed;
57 extern u64 cpu_mondo_counter[NR_CPUS];
58 
59 unsigned long real_hard_smp_processor_id(void);
60 
61 struct cpuid_patch_entry {
62 	unsigned int	addr;
63 	unsigned int	cheetah_safari[4];
64 	unsigned int	cheetah_jbus[4];
65 	unsigned int	starfire[4];
66 	unsigned int	sun4v[4];
67 };
68 extern struct cpuid_patch_entry __cpuid_patch, __cpuid_patch_end;
69 
70 struct sun4v_1insn_patch_entry {
71 	unsigned int	addr;
72 	unsigned int	insn;
73 };
74 extern struct sun4v_1insn_patch_entry __sun4v_1insn_patch,
75 	__sun4v_1insn_patch_end;
76 extern struct sun4v_1insn_patch_entry __fast_win_ctrl_1insn_patch,
77 	__fast_win_ctrl_1insn_patch_end;
78 
79 struct sun4v_2insn_patch_entry {
80 	unsigned int	addr;
81 	unsigned int	insns[2];
82 };
83 extern struct sun4v_2insn_patch_entry __sun4v_2insn_patch,
84 	__sun4v_2insn_patch_end;
85 extern struct sun4v_2insn_patch_entry __sun_m7_2insn_patch,
86 	__sun_m7_2insn_patch_end;
87 
88 
89 #endif /* !(__ASSEMBLY__) */
90 
91 #define TRAP_PER_CPU_THREAD		0x00
92 #define TRAP_PER_CPU_PGD_PADDR		0x08
93 #define TRAP_PER_CPU_CPU_MONDO_PA	0x10
94 #define TRAP_PER_CPU_DEV_MONDO_PA	0x18
95 #define TRAP_PER_CPU_RESUM_MONDO_PA	0x20
96 #define TRAP_PER_CPU_RESUM_KBUF_PA	0x28
97 #define TRAP_PER_CPU_NONRESUM_MONDO_PA	0x30
98 #define TRAP_PER_CPU_NONRESUM_KBUF_PA	0x38
99 #define TRAP_PER_CPU_FAULT_INFO		0x40
100 #define TRAP_PER_CPU_CPU_MONDO_BLOCK_PA	0xc0
101 #define TRAP_PER_CPU_CPU_LIST_PA	0xc8
102 #define TRAP_PER_CPU_TSB_HUGE		0xd0
103 #define TRAP_PER_CPU_TSB_HUGE_TEMP	0xd8
104 #define TRAP_PER_CPU_IRQ_WORKLIST_PA	0xe0
105 #define TRAP_PER_CPU_CPU_MONDO_QMASK	0xe8
106 #define TRAP_PER_CPU_DEV_MONDO_QMASK	0xec
107 #define TRAP_PER_CPU_RESUM_QMASK	0xf0
108 #define TRAP_PER_CPU_NONRESUM_QMASK	0xf4
109 #define TRAP_PER_CPU_PER_CPU_BASE	0xf8
110 
111 #define TRAP_BLOCK_SZ_SHIFT		8
112 
113 #include <asm/scratchpad.h>
114 
115 #define __GET_CPUID(REG)				\
116 	/* Spitfire implementation (default). */	\
117 661:	ldxa		[%g0] ASI_UPA_CONFIG, REG;	\
118 	srlx		REG, 17, REG;			\
119 	 and		REG, 0x1f, REG;			\
120 	nop;						\
121 	.section	.cpuid_patch, "ax";		\
122 	/* Instruction location. */			\
123 	.word		661b;				\
124 	/* Cheetah Safari implementation. */		\
125 	ldxa		[%g0] ASI_SAFARI_CONFIG, REG;	\
126 	srlx		REG, 17, REG;			\
127 	and		REG, 0x3ff, REG;		\
128 	nop;						\
129 	/* Cheetah JBUS implementation. */		\
130 	ldxa		[%g0] ASI_JBUS_CONFIG, REG;	\
131 	srlx		REG, 17, REG;			\
132 	and		REG, 0x1f, REG;			\
133 	nop;						\
134 	/* Starfire implementation. */			\
135 	sethi		%hi(0x1fff40000d0 >> 9), REG;	\
136 	sllx		REG, 9, REG;			\
137 	or		REG, 0xd0, REG;			\
138 	lduwa		[REG] ASI_PHYS_BYPASS_EC_E, REG;\
139 	/* sun4v implementation. */			\
140 	mov		SCRATCHPAD_CPUID, REG;		\
141 	ldxa		[REG] ASI_SCRATCHPAD, REG;	\
142 	nop;						\
143 	nop;						\
144 	.previous;
145 
146 #ifdef CONFIG_SMP
147 
148 #define TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
149 	__GET_CPUID(TMP)			\
150 	sethi	%hi(trap_block), DEST;		\
151 	sllx	TMP, TRAP_BLOCK_SZ_SHIFT, TMP;	\
152 	or	DEST, %lo(trap_block), DEST;	\
153 	add	DEST, TMP, DEST;		\
154 
155 /* Clobbers TMP, current address space PGD phys address into DEST.  */
156 #define TRAP_LOAD_PGD_PHYS(DEST, TMP)		\
157 	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
158 	ldx	[DEST + TRAP_PER_CPU_PGD_PADDR], DEST;
159 
160 /* Clobbers TMP, loads local processor's IRQ work area into DEST.  */
161 #define TRAP_LOAD_IRQ_WORK_PA(DEST, TMP)	\
162 	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
163 	add	DEST, TRAP_PER_CPU_IRQ_WORKLIST_PA, DEST;
164 
165 /* Clobbers TMP, loads DEST with current thread info pointer.  */
166 #define TRAP_LOAD_THREAD_REG(DEST, TMP)		\
167 	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
168 	ldx	[DEST + TRAP_PER_CPU_THREAD], DEST;
169 
170 /* Given the current thread info pointer in THR, load the per-cpu
171  * area base of the current processor into DEST.  REG1, REG2, and REG3 are
172  * clobbered.
173  *
174  * You absolutely cannot use DEST as a temporary in this code.  The
175  * reason is that traps can happen during execution, and return from
176  * trap will load the fully resolved DEST per-cpu base.  This can corrupt
177  * the calculations done by the macro mid-stream.
178  */
179 #define LOAD_PER_CPU_BASE(DEST, THR, REG1, REG2, REG3)	\
180 	lduh	[THR + TI_CPU], REG1;			\
181 	sethi	%hi(trap_block), REG2;			\
182 	sllx	REG1, TRAP_BLOCK_SZ_SHIFT, REG1;	\
183 	or	REG2, %lo(trap_block), REG2;		\
184 	add	REG2, REG1, REG2;			\
185 	ldx	[REG2 + TRAP_PER_CPU_PER_CPU_BASE], DEST;
186 
187 #else
188 
189 #define TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
190 	sethi	%hi(trap_block), DEST;		\
191 	or	DEST, %lo(trap_block), DEST;	\
192 
193 /* Uniprocessor versions, we know the cpuid is zero.  */
194 #define TRAP_LOAD_PGD_PHYS(DEST, TMP)		\
195 	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
196 	ldx	[DEST + TRAP_PER_CPU_PGD_PADDR], DEST;
197 
198 /* Clobbers TMP, loads local processor's IRQ work area into DEST.  */
199 #define TRAP_LOAD_IRQ_WORK_PA(DEST, TMP)	\
200 	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
201 	add	DEST, TRAP_PER_CPU_IRQ_WORKLIST_PA, DEST;
202 
203 #define TRAP_LOAD_THREAD_REG(DEST, TMP)		\
204 	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
205 	ldx	[DEST + TRAP_PER_CPU_THREAD], DEST;
206 
207 /* No per-cpu areas on uniprocessor, so no need to load DEST.  */
208 #define LOAD_PER_CPU_BASE(DEST, THR, REG1, REG2, REG3)
209 
210 #endif /* !(CONFIG_SMP) */
211 
212 #endif /* _SPARC_TRAP_BLOCK_H */
213