1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _SPARC_TRAP_BLOCK_H
3#define _SPARC_TRAP_BLOCK_H
4
5#include <linux/threads.h>
6
7#include <asm/hypervisor.h>
8#include <asm/asi.h>
9
10#ifndef __ASSEMBLY__
11
12/* Trap handling code needs to get at a few critical values upon
13 * trap entry and to process TSB misses.  These cannot be in the
14 * per_cpu() area as we really need to lock them into the TLB and
15 * thus make them part of the main kernel image.  As a result we
16 * try to make this as small as possible.
17 *
18 * This is padded out and aligned to 64-bytes to avoid false sharing
19 * on SMP.
20 */
21
22/* If you modify the size of this structure, please update
23 * TRAP_BLOCK_SZ_SHIFT below.
24 */
25struct thread_info;
26struct trap_per_cpu {
27/* D-cache line 1: Basic thread information, cpu and device mondo queues */
28	struct thread_info	*thread;
29	unsigned long		pgd_paddr;
30	unsigned long		cpu_mondo_pa;
31	unsigned long		dev_mondo_pa;
32
33/* D-cache line 2: Error Mondo Queue and kernel buffer pointers */
34	unsigned long		resum_mondo_pa;
35	unsigned long		resum_kernel_buf_pa;
36	unsigned long		nonresum_mondo_pa;
37	unsigned long		nonresum_kernel_buf_pa;
38
39/* Dcache lines 3, 4, 5, and 6: Hypervisor Fault Status */
40	struct hv_fault_status	fault_info;
41
42/* Dcache line 7: Physical addresses of CPU send mondo block and CPU list.  */
43	unsigned long		cpu_mondo_block_pa;
44	unsigned long		cpu_list_pa;
45	unsigned long		tsb_huge;
46	unsigned long		tsb_huge_temp;
47
48/* Dcache line 8: IRQ work list, and keep trap_block a power-of-2 in size.  */
49	unsigned long		irq_worklist_pa;
50	unsigned int		cpu_mondo_qmask;
51	unsigned int		dev_mondo_qmask;
52	unsigned int		resum_qmask;
53	unsigned int		nonresum_qmask;
54	unsigned long		__per_cpu_base;
55} __attribute__((aligned(64)));
56extern struct trap_per_cpu trap_block[NR_CPUS];
57void init_cur_cpu_trap(struct thread_info *);
58void setup_tba(void);
59extern int ncpus_probed;
60extern u64 cpu_mondo_counter[NR_CPUS];
61
62unsigned long real_hard_smp_processor_id(void);
63
64struct cpuid_patch_entry {
65	unsigned int	addr;
66	unsigned int	cheetah_safari[4];
67	unsigned int	cheetah_jbus[4];
68	unsigned int	starfire[4];
69	unsigned int	sun4v[4];
70};
71extern struct cpuid_patch_entry __cpuid_patch, __cpuid_patch_end;
72
73struct sun4v_1insn_patch_entry {
74	unsigned int	addr;
75	unsigned int	insn;
76};
77extern struct sun4v_1insn_patch_entry __sun4v_1insn_patch,
78	__sun4v_1insn_patch_end;
79extern struct sun4v_1insn_patch_entry __fast_win_ctrl_1insn_patch,
80	__fast_win_ctrl_1insn_patch_end;
81extern struct sun4v_1insn_patch_entry __sun_m7_1insn_patch,
82	__sun_m7_1insn_patch_end;
83
84struct sun4v_2insn_patch_entry {
85	unsigned int	addr;
86	unsigned int	insns[2];
87};
88extern struct sun4v_2insn_patch_entry __sun4v_2insn_patch,
89	__sun4v_2insn_patch_end;
90extern struct sun4v_2insn_patch_entry __sun_m7_2insn_patch,
91	__sun_m7_2insn_patch_end;
92
93
94#endif /* !(__ASSEMBLY__) */
95
96#define TRAP_PER_CPU_THREAD		0x00
97#define TRAP_PER_CPU_PGD_PADDR		0x08
98#define TRAP_PER_CPU_CPU_MONDO_PA	0x10
99#define TRAP_PER_CPU_DEV_MONDO_PA	0x18
100#define TRAP_PER_CPU_RESUM_MONDO_PA	0x20
101#define TRAP_PER_CPU_RESUM_KBUF_PA	0x28
102#define TRAP_PER_CPU_NONRESUM_MONDO_PA	0x30
103#define TRAP_PER_CPU_NONRESUM_KBUF_PA	0x38
104#define TRAP_PER_CPU_FAULT_INFO		0x40
105#define TRAP_PER_CPU_CPU_MONDO_BLOCK_PA	0xc0
106#define TRAP_PER_CPU_CPU_LIST_PA	0xc8
107#define TRAP_PER_CPU_TSB_HUGE		0xd0
108#define TRAP_PER_CPU_TSB_HUGE_TEMP	0xd8
109#define TRAP_PER_CPU_IRQ_WORKLIST_PA	0xe0
110#define TRAP_PER_CPU_CPU_MONDO_QMASK	0xe8
111#define TRAP_PER_CPU_DEV_MONDO_QMASK	0xec
112#define TRAP_PER_CPU_RESUM_QMASK	0xf0
113#define TRAP_PER_CPU_NONRESUM_QMASK	0xf4
114#define TRAP_PER_CPU_PER_CPU_BASE	0xf8
115
116#define TRAP_BLOCK_SZ_SHIFT		8
117
118#include <asm/scratchpad.h>
119
120#define __GET_CPUID(REG)				\
121	/* Spitfire implementation (default). */	\
122661:	ldxa		[%g0] ASI_UPA_CONFIG, REG;	\
123	srlx		REG, 17, REG;			\
124	 and		REG, 0x1f, REG;			\
125	nop;						\
126	.section	.cpuid_patch, "ax";		\
127	/* Instruction location. */			\
128	.word		661b;				\
129	/* Cheetah Safari implementation. */		\
130	ldxa		[%g0] ASI_SAFARI_CONFIG, REG;	\
131	srlx		REG, 17, REG;			\
132	and		REG, 0x3ff, REG;		\
133	nop;						\
134	/* Cheetah JBUS implementation. */		\
135	ldxa		[%g0] ASI_JBUS_CONFIG, REG;	\
136	srlx		REG, 17, REG;			\
137	and		REG, 0x1f, REG;			\
138	nop;						\
139	/* Starfire implementation. */			\
140	sethi		%hi(0x1fff40000d0 >> 9), REG;	\
141	sllx		REG, 9, REG;			\
142	or		REG, 0xd0, REG;			\
143	lduwa		[REG] ASI_PHYS_BYPASS_EC_E, REG;\
144	/* sun4v implementation. */			\
145	mov		SCRATCHPAD_CPUID, REG;		\
146	ldxa		[REG] ASI_SCRATCHPAD, REG;	\
147	nop;						\
148	nop;						\
149	.previous;
150
151#ifdef CONFIG_SMP
152
153#define TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
154	__GET_CPUID(TMP)			\
155	sethi	%hi(trap_block), DEST;		\
156	sllx	TMP, TRAP_BLOCK_SZ_SHIFT, TMP;	\
157	or	DEST, %lo(trap_block), DEST;	\
158	add	DEST, TMP, DEST;		\
159
160/* Clobbers TMP, current address space PGD phys address into DEST.  */
161#define TRAP_LOAD_PGD_PHYS(DEST, TMP)		\
162	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
163	ldx	[DEST + TRAP_PER_CPU_PGD_PADDR], DEST;
164
165/* Clobbers TMP, loads local processor's IRQ work area into DEST.  */
166#define TRAP_LOAD_IRQ_WORK_PA(DEST, TMP)	\
167	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
168	add	DEST, TRAP_PER_CPU_IRQ_WORKLIST_PA, DEST;
169
170/* Clobbers TMP, loads DEST with current thread info pointer.  */
171#define TRAP_LOAD_THREAD_REG(DEST, TMP)		\
172	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
173	ldx	[DEST + TRAP_PER_CPU_THREAD], DEST;
174
175/* Given the current thread info pointer in THR, load the per-cpu
176 * area base of the current processor into DEST.  REG1, REG2, and REG3 are
177 * clobbered.
178 *
179 * You absolutely cannot use DEST as a temporary in this code.  The
180 * reason is that traps can happen during execution, and return from
181 * trap will load the fully resolved DEST per-cpu base.  This can corrupt
182 * the calculations done by the macro mid-stream.
183 */
184#define LOAD_PER_CPU_BASE(DEST, THR, REG1, REG2, REG3)	\
185	lduh	[THR + TI_CPU], REG1;			\
186	sethi	%hi(trap_block), REG2;			\
187	sllx	REG1, TRAP_BLOCK_SZ_SHIFT, REG1;	\
188	or	REG2, %lo(trap_block), REG2;		\
189	add	REG2, REG1, REG2;			\
190	ldx	[REG2 + TRAP_PER_CPU_PER_CPU_BASE], DEST;
191
192#else
193
194#define TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
195	sethi	%hi(trap_block), DEST;		\
196	or	DEST, %lo(trap_block), DEST;	\
197
198/* Uniprocessor versions, we know the cpuid is zero.  */
199#define TRAP_LOAD_PGD_PHYS(DEST, TMP)		\
200	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
201	ldx	[DEST + TRAP_PER_CPU_PGD_PADDR], DEST;
202
203/* Clobbers TMP, loads local processor's IRQ work area into DEST.  */
204#define TRAP_LOAD_IRQ_WORK_PA(DEST, TMP)	\
205	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
206	add	DEST, TRAP_PER_CPU_IRQ_WORKLIST_PA, DEST;
207
208#define TRAP_LOAD_THREAD_REG(DEST, TMP)		\
209	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
210	ldx	[DEST + TRAP_PER_CPU_THREAD], DEST;
211
212/* No per-cpu areas on uniprocessor, so no need to load DEST.  */
213#define LOAD_PER_CPU_BASE(DEST, THR, REG1, REG2, REG3)
214
215#endif /* !(CONFIG_SMP) */
216
217#endif /* _SPARC_TRAP_BLOCK_H */
218