1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3 * lppaca.h
4 * Copyright (C) 2001 Mike Corrigan IBM Corporation
5 */
6 #ifndef _ASM_POWERPC_LPPACA_H
7 #define _ASM_POWERPC_LPPACA_H
8
9 /*
10 * The below VPHN macros are outside the __KERNEL__ check since these are
11 * used for compiling the vphn selftest in userspace
12 */
13
14 /* The H_HOME_NODE_ASSOCIATIVITY h_call returns 6 64-bit registers. */
15 #define VPHN_REGISTER_COUNT 6
16
17 /*
18 * 6 64-bit registers unpacked into up to 24 be32 associativity values. To
19 * form the complete property we have to add the length in the first cell.
20 */
21 #define VPHN_ASSOC_BUFSIZE (VPHN_REGISTER_COUNT*sizeof(u64)/sizeof(u16) + 1)
22
23 /*
24 * The H_HOME_NODE_ASSOCIATIVITY hcall takes two values for flags:
25 * 1 for retrieving associativity information for a guest cpu
26 * 2 for retrieving associativity information for a host/hypervisor cpu
27 */
28 #define VPHN_FLAG_VCPU 1
29 #define VPHN_FLAG_PCPU 2
30
31 #ifdef __KERNEL__
32
33 /*
34 * These definitions relate to hypervisors that only exist when using
35 * a server type processor
36 */
37 #ifdef CONFIG_PPC_BOOK3S
38
39 /*
40 * This control block contains the data that is shared between the
41 * hypervisor and the OS.
42 */
43 #include <linux/cache.h>
44 #include <linux/threads.h>
45 #include <linux/spinlock_types.h>
46 #include <asm/types.h>
47 #include <asm/mmu.h>
48 #include <asm/firmware.h>
49
50 /*
51 * The lppaca is the "virtual processor area" registered with the hypervisor,
52 * H_REGISTER_VPA etc.
53 *
54 * According to PAPR, the structure is 640 bytes long, must be L1 cache line
55 * aligned, and must not cross a 4kB boundary. Its size field must be at
56 * least 640 bytes (but may be more).
57 *
58 * Pre-v4.14 KVM hypervisors reject the VPA if its size field is smaller than
59 * 1kB, so we dynamically allocate 1kB and advertise size as 1kB, but keep
60 * this structure as the canonical 640 byte size.
61 */
62 struct lppaca {
63 /* cacheline 1 contains read-only data */
64
65 __be32 desc; /* Eye catcher 0xD397D781 */
66 __be16 size; /* Size of this struct */
67 u8 reserved1[3];
68 u8 __old_status; /* Old status, including shared proc */
69 u8 reserved3[14];
70 volatile __be32 dyn_hw_node_id; /* Dynamic hardware node id */
71 volatile __be32 dyn_hw_proc_id; /* Dynamic hardware proc id */
72 u8 reserved4[56];
73 volatile u8 vphn_assoc_counts[8]; /* Virtual processor home node */
74 /* associativity change counters */
75 u8 reserved5[32];
76
77 /* cacheline 2 contains local read-write data */
78
79 u8 reserved6[48];
80 u8 cede_latency_hint;
81 u8 ebb_regs_in_use;
82 u8 reserved7[6];
83 u8 dtl_enable_mask; /* Dispatch Trace Log mask */
84 u8 donate_dedicated_cpu; /* Donate dedicated CPU cycles */
85 u8 fpregs_in_use;
86 u8 pmcregs_in_use;
87 u8 reserved8[28];
88 __be64 wait_state_cycles; /* Wait cycles for this proc */
89 u8 reserved9[28];
90 __be16 slb_count; /* # of SLBs to maintain */
91 u8 idle; /* Indicate OS is idle */
92 u8 vmxregs_in_use;
93
94 /* cacheline 3 is shared with other processors */
95
96 /*
97 * This is the yield_count. An "odd" value (low bit on) means that
98 * the processor is yielded (either because of an OS yield or a
99 * hypervisor preempt). An even value implies that the processor is
100 * currently executing.
101 * NOTE: Even dedicated processor partitions can yield so this
102 * field cannot be used to determine if we are shared or dedicated.
103 */
104 volatile __be32 yield_count;
105 volatile __be32 dispersion_count; /* dispatch changed physical cpu */
106 volatile __be64 cmo_faults; /* CMO page fault count */
107 volatile __be64 cmo_fault_time; /* CMO page fault time */
108 u8 reserved10[104];
109
110 /* cacheline 4-5 */
111
112 __be32 page_ins; /* CMO Hint - # page ins by OS */
113 u8 reserved11[148];
114 volatile __be64 dtl_idx; /* Dispatch Trace Log head index */
115 u8 reserved12[96];
116 } ____cacheline_aligned;
117
118 #define lppaca_of(cpu) (*paca_ptrs[cpu]->lppaca_ptr)
119
120 /*
121 * We are using a non architected field to determine if a partition is
122 * shared or dedicated. This currently works on both KVM and PHYP, but
123 * we will have to transition to something better.
124 */
125 #define LPPACA_OLD_SHARED_PROC 2
126
lppaca_shared_proc(struct lppaca * l)127 static inline bool lppaca_shared_proc(struct lppaca *l)
128 {
129 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
130 return false;
131 return !!(l->__old_status & LPPACA_OLD_SHARED_PROC);
132 }
133
134 /*
135 * SLB shadow buffer structure as defined in the PAPR. The save_area
136 * contains adjacent ESID and VSID pairs for each shadowed SLB. The
137 * ESID is stored in the lower 64bits, then the VSID.
138 */
139 struct slb_shadow {
140 __be32 persistent; /* Number of persistent SLBs */
141 __be32 buffer_length; /* Total shadow buffer length */
142 __be64 reserved;
143 struct {
144 __be64 esid;
145 __be64 vsid;
146 } save_area[SLB_NUM_BOLTED];
147 } ____cacheline_aligned;
148
149 /*
150 * Layout of entries in the hypervisor's dispatch trace log buffer.
151 */
152 struct dtl_entry {
153 u8 dispatch_reason;
154 u8 preempt_reason;
155 __be16 processor_id;
156 __be32 enqueue_to_dispatch_time;
157 __be32 ready_to_enqueue_time;
158 __be32 waiting_to_ready_time;
159 __be64 timebase;
160 __be64 fault_addr;
161 __be64 srr0;
162 __be64 srr1;
163 };
164
165 #define DISPATCH_LOG_BYTES 4096 /* bytes per cpu */
166 #define N_DISPATCH_LOG (DISPATCH_LOG_BYTES / sizeof(struct dtl_entry))
167
168 /*
169 * Dispatch trace log event enable mask:
170 * 0x1: voluntary virtual processor waits
171 * 0x2: time-slice preempts
172 * 0x4: virtual partition memory page faults
173 */
174 #define DTL_LOG_CEDE 0x1
175 #define DTL_LOG_PREEMPT 0x2
176 #define DTL_LOG_FAULT 0x4
177 #define DTL_LOG_ALL (DTL_LOG_CEDE | DTL_LOG_PREEMPT | DTL_LOG_FAULT)
178
179 extern struct kmem_cache *dtl_cache;
180 extern rwlock_t dtl_access_lock;
181
182 /*
183 * When CONFIG_VIRT_CPU_ACCOUNTING_NATIVE = y, the cpu accounting code controls
184 * reading from the dispatch trace log. If other code wants to consume
185 * DTL entries, it can set this pointer to a function that will get
186 * called once for each DTL entry that gets processed.
187 */
188 extern void (*dtl_consumer)(struct dtl_entry *entry, u64 index);
189
190 extern void register_dtl_buffer(int cpu);
191 extern void alloc_dtl_buffers(unsigned long *time_limit);
192 extern long hcall_vphn(unsigned long cpu, u64 flags, __be32 *associativity);
193
194 #endif /* CONFIG_PPC_BOOK3S */
195 #endif /* __KERNEL__ */
196 #endif /* _ASM_POWERPC_LPPACA_H */
197