1*4882a593Smuzhiyun /**
2*4882a593Smuzhiyun * @file arch/alpha/oprofile/op_model_ev5.c
3*4882a593Smuzhiyun *
4*4882a593Smuzhiyun * @remark Copyright 2002 OProfile authors
5*4882a593Smuzhiyun * @remark Read the file COPYING
6*4882a593Smuzhiyun *
7*4882a593Smuzhiyun * @author Richard Henderson <rth@twiddle.net>
8*4882a593Smuzhiyun */
9*4882a593Smuzhiyun
10*4882a593Smuzhiyun #include <linux/oprofile.h>
11*4882a593Smuzhiyun #include <linux/smp.h>
12*4882a593Smuzhiyun #include <asm/ptrace.h>
13*4882a593Smuzhiyun
14*4882a593Smuzhiyun #include "op_impl.h"
15*4882a593Smuzhiyun
16*4882a593Smuzhiyun
17*4882a593Smuzhiyun /* Compute all of the registers in preparation for enabling profiling.
18*4882a593Smuzhiyun
19*4882a593Smuzhiyun The 21164 (EV5) and 21164PC (PCA65) vary in the bit placement and
20*4882a593Smuzhiyun meaning of the "CBOX" events. Given that we don't care about meaning
21*4882a593Smuzhiyun at this point, arrange for the difference in bit placement to be
22*4882a593Smuzhiyun handled by common code. */
23*4882a593Smuzhiyun
24*4882a593Smuzhiyun static void
common_reg_setup(struct op_register_config * reg,struct op_counter_config * ctr,struct op_system_config * sys,int cbox1_ofs,int cbox2_ofs)25*4882a593Smuzhiyun common_reg_setup(struct op_register_config *reg,
26*4882a593Smuzhiyun struct op_counter_config *ctr,
27*4882a593Smuzhiyun struct op_system_config *sys,
28*4882a593Smuzhiyun int cbox1_ofs, int cbox2_ofs)
29*4882a593Smuzhiyun {
30*4882a593Smuzhiyun int i, ctl, reset, need_reset;
31*4882a593Smuzhiyun
32*4882a593Smuzhiyun /* Select desired events. The event numbers are selected such
33*4882a593Smuzhiyun that they map directly into the event selection fields:
34*4882a593Smuzhiyun
35*4882a593Smuzhiyun PCSEL0: 0, 1
36*4882a593Smuzhiyun PCSEL1: 24-39
37*4882a593Smuzhiyun CBOX1: 40-47
38*4882a593Smuzhiyun PCSEL2: 48-63
39*4882a593Smuzhiyun CBOX2: 64-71
40*4882a593Smuzhiyun
41*4882a593Smuzhiyun There are two special cases, in that CYCLES can be measured
42*4882a593Smuzhiyun on PCSEL[02], and SCACHE_WRITE can be measured on CBOX[12].
43*4882a593Smuzhiyun These event numbers are canonicalizes to their first appearance. */
44*4882a593Smuzhiyun
45*4882a593Smuzhiyun ctl = 0;
46*4882a593Smuzhiyun for (i = 0; i < 3; ++i) {
47*4882a593Smuzhiyun unsigned long event = ctr[i].event;
48*4882a593Smuzhiyun if (!ctr[i].enabled)
49*4882a593Smuzhiyun continue;
50*4882a593Smuzhiyun
51*4882a593Smuzhiyun /* Remap the duplicate events, as described above. */
52*4882a593Smuzhiyun if (i == 2) {
53*4882a593Smuzhiyun if (event == 0)
54*4882a593Smuzhiyun event = 12+48;
55*4882a593Smuzhiyun else if (event == 2+41)
56*4882a593Smuzhiyun event = 4+65;
57*4882a593Smuzhiyun }
58*4882a593Smuzhiyun
59*4882a593Smuzhiyun /* Convert the event numbers onto mux_select bit mask. */
60*4882a593Smuzhiyun if (event < 2)
61*4882a593Smuzhiyun ctl |= event << 31;
62*4882a593Smuzhiyun else if (event < 24)
63*4882a593Smuzhiyun /* error */;
64*4882a593Smuzhiyun else if (event < 40)
65*4882a593Smuzhiyun ctl |= (event - 24) << 4;
66*4882a593Smuzhiyun else if (event < 48)
67*4882a593Smuzhiyun ctl |= (event - 40) << cbox1_ofs | 15 << 4;
68*4882a593Smuzhiyun else if (event < 64)
69*4882a593Smuzhiyun ctl |= event - 48;
70*4882a593Smuzhiyun else if (event < 72)
71*4882a593Smuzhiyun ctl |= (event - 64) << cbox2_ofs | 15;
72*4882a593Smuzhiyun }
73*4882a593Smuzhiyun reg->mux_select = ctl;
74*4882a593Smuzhiyun
75*4882a593Smuzhiyun /* Select processor mode. */
76*4882a593Smuzhiyun /* ??? Need to come up with some mechanism to trace only selected
77*4882a593Smuzhiyun processes. For now select from pal, kernel and user mode. */
78*4882a593Smuzhiyun ctl = 0;
79*4882a593Smuzhiyun ctl |= !sys->enable_pal << 9;
80*4882a593Smuzhiyun ctl |= !sys->enable_kernel << 8;
81*4882a593Smuzhiyun ctl |= !sys->enable_user << 30;
82*4882a593Smuzhiyun reg->proc_mode = ctl;
83*4882a593Smuzhiyun
84*4882a593Smuzhiyun /* Select interrupt frequencies. Take the interrupt count selected
85*4882a593Smuzhiyun by the user, and map it onto one of the possible counter widths.
86*4882a593Smuzhiyun If the user value is in between, compute a value to which the
87*4882a593Smuzhiyun counter is reset at each interrupt. */
88*4882a593Smuzhiyun
89*4882a593Smuzhiyun ctl = reset = need_reset = 0;
90*4882a593Smuzhiyun for (i = 0; i < 3; ++i) {
91*4882a593Smuzhiyun unsigned long max, hilo, count = ctr[i].count;
92*4882a593Smuzhiyun if (!ctr[i].enabled)
93*4882a593Smuzhiyun continue;
94*4882a593Smuzhiyun
95*4882a593Smuzhiyun if (count <= 256)
96*4882a593Smuzhiyun count = 256, hilo = 3, max = 256;
97*4882a593Smuzhiyun else {
98*4882a593Smuzhiyun max = (i == 2 ? 16384 : 65536);
99*4882a593Smuzhiyun hilo = 2;
100*4882a593Smuzhiyun if (count > max)
101*4882a593Smuzhiyun count = max;
102*4882a593Smuzhiyun }
103*4882a593Smuzhiyun ctr[i].count = count;
104*4882a593Smuzhiyun
105*4882a593Smuzhiyun ctl |= hilo << (8 - i*2);
106*4882a593Smuzhiyun reset |= (max - count) << (48 - 16*i);
107*4882a593Smuzhiyun if (count != max)
108*4882a593Smuzhiyun need_reset |= 1 << i;
109*4882a593Smuzhiyun }
110*4882a593Smuzhiyun reg->freq = ctl;
111*4882a593Smuzhiyun reg->reset_values = reset;
112*4882a593Smuzhiyun reg->need_reset = need_reset;
113*4882a593Smuzhiyun }
114*4882a593Smuzhiyun
115*4882a593Smuzhiyun static void
ev5_reg_setup(struct op_register_config * reg,struct op_counter_config * ctr,struct op_system_config * sys)116*4882a593Smuzhiyun ev5_reg_setup(struct op_register_config *reg,
117*4882a593Smuzhiyun struct op_counter_config *ctr,
118*4882a593Smuzhiyun struct op_system_config *sys)
119*4882a593Smuzhiyun {
120*4882a593Smuzhiyun common_reg_setup(reg, ctr, sys, 19, 22);
121*4882a593Smuzhiyun }
122*4882a593Smuzhiyun
123*4882a593Smuzhiyun static void
pca56_reg_setup(struct op_register_config * reg,struct op_counter_config * ctr,struct op_system_config * sys)124*4882a593Smuzhiyun pca56_reg_setup(struct op_register_config *reg,
125*4882a593Smuzhiyun struct op_counter_config *ctr,
126*4882a593Smuzhiyun struct op_system_config *sys)
127*4882a593Smuzhiyun {
128*4882a593Smuzhiyun common_reg_setup(reg, ctr, sys, 8, 11);
129*4882a593Smuzhiyun }
130*4882a593Smuzhiyun
131*4882a593Smuzhiyun /* Program all of the registers in preparation for enabling profiling. */
132*4882a593Smuzhiyun
133*4882a593Smuzhiyun static void
ev5_cpu_setup(void * x)134*4882a593Smuzhiyun ev5_cpu_setup (void *x)
135*4882a593Smuzhiyun {
136*4882a593Smuzhiyun struct op_register_config *reg = x;
137*4882a593Smuzhiyun
138*4882a593Smuzhiyun wrperfmon(2, reg->mux_select);
139*4882a593Smuzhiyun wrperfmon(3, reg->proc_mode);
140*4882a593Smuzhiyun wrperfmon(4, reg->freq);
141*4882a593Smuzhiyun wrperfmon(6, reg->reset_values);
142*4882a593Smuzhiyun }
143*4882a593Smuzhiyun
144*4882a593Smuzhiyun /* CTR is a counter for which the user has requested an interrupt count
145*4882a593Smuzhiyun in between one of the widths selectable in hardware. Reset the count
146*4882a593Smuzhiyun for CTR to the value stored in REG->RESET_VALUES.
147*4882a593Smuzhiyun
148*4882a593Smuzhiyun For EV5, this means disabling profiling, reading the current values,
149*4882a593Smuzhiyun masking in the value for the desired register, writing, then turning
150*4882a593Smuzhiyun profiling back on.
151*4882a593Smuzhiyun
152*4882a593Smuzhiyun This can be streamlined if profiling is only enabled for user mode.
153*4882a593Smuzhiyun In that case we know that the counters are not currently incrementing
154*4882a593Smuzhiyun (due to being in kernel mode). */
155*4882a593Smuzhiyun
156*4882a593Smuzhiyun static void
ev5_reset_ctr(struct op_register_config * reg,unsigned long ctr)157*4882a593Smuzhiyun ev5_reset_ctr(struct op_register_config *reg, unsigned long ctr)
158*4882a593Smuzhiyun {
159*4882a593Smuzhiyun unsigned long values, mask, not_pk, reset_values;
160*4882a593Smuzhiyun
161*4882a593Smuzhiyun mask = (ctr == 0 ? 0xfffful << 48
162*4882a593Smuzhiyun : ctr == 1 ? 0xfffful << 32
163*4882a593Smuzhiyun : 0x3fff << 16);
164*4882a593Smuzhiyun
165*4882a593Smuzhiyun not_pk = 1 << 9 | 1 << 8;
166*4882a593Smuzhiyun
167*4882a593Smuzhiyun reset_values = reg->reset_values;
168*4882a593Smuzhiyun
169*4882a593Smuzhiyun if ((reg->proc_mode & not_pk) == not_pk) {
170*4882a593Smuzhiyun values = wrperfmon(5, 0);
171*4882a593Smuzhiyun values = (reset_values & mask) | (values & ~mask & -2);
172*4882a593Smuzhiyun wrperfmon(6, values);
173*4882a593Smuzhiyun } else {
174*4882a593Smuzhiyun wrperfmon(0, -1);
175*4882a593Smuzhiyun values = wrperfmon(5, 0);
176*4882a593Smuzhiyun values = (reset_values & mask) | (values & ~mask & -2);
177*4882a593Smuzhiyun wrperfmon(6, values);
178*4882a593Smuzhiyun wrperfmon(1, reg->enable);
179*4882a593Smuzhiyun }
180*4882a593Smuzhiyun }
181*4882a593Smuzhiyun
182*4882a593Smuzhiyun static void
ev5_handle_interrupt(unsigned long which,struct pt_regs * regs,struct op_counter_config * ctr)183*4882a593Smuzhiyun ev5_handle_interrupt(unsigned long which, struct pt_regs *regs,
184*4882a593Smuzhiyun struct op_counter_config *ctr)
185*4882a593Smuzhiyun {
186*4882a593Smuzhiyun /* Record the sample. */
187*4882a593Smuzhiyun oprofile_add_sample(regs, which);
188*4882a593Smuzhiyun }
189*4882a593Smuzhiyun
190*4882a593Smuzhiyun
191*4882a593Smuzhiyun struct op_axp_model op_model_ev5 = {
192*4882a593Smuzhiyun .reg_setup = ev5_reg_setup,
193*4882a593Smuzhiyun .cpu_setup = ev5_cpu_setup,
194*4882a593Smuzhiyun .reset_ctr = ev5_reset_ctr,
195*4882a593Smuzhiyun .handle_interrupt = ev5_handle_interrupt,
196*4882a593Smuzhiyun .cpu_type = "alpha/ev5",
197*4882a593Smuzhiyun .num_counters = 3,
198*4882a593Smuzhiyun .can_set_proc_mode = 1,
199*4882a593Smuzhiyun };
200*4882a593Smuzhiyun
201*4882a593Smuzhiyun struct op_axp_model op_model_pca56 = {
202*4882a593Smuzhiyun .reg_setup = pca56_reg_setup,
203*4882a593Smuzhiyun .cpu_setup = ev5_cpu_setup,
204*4882a593Smuzhiyun .reset_ctr = ev5_reset_ctr,
205*4882a593Smuzhiyun .handle_interrupt = ev5_handle_interrupt,
206*4882a593Smuzhiyun .cpu_type = "alpha/pca56",
207*4882a593Smuzhiyun .num_counters = 3,
208*4882a593Smuzhiyun .can_set_proc_mode = 1,
209*4882a593Smuzhiyun };
210