| 34 |
- |
1 |
# This is a simple script that checks the contents of /proc/mtrr to see if
|
|
|
2 |
# the BIOS maker for the computer took the easy way out in terms of
|
|
|
3 |
# specifying memory regions when there is a hole below 4GB for PCI access
|
|
|
4 |
# and the machine has 4GB or more of RAM. When the contents of /proc/mtrr
|
|
|
5 |
# show a 4GB mapping of write-back cached RAM, minus punch out hole(s) of
|
|
|
6 |
# uncacheable regions (the area reserved for PCI access), then it becomes
|
|
|
7 |
# impossible for the ib_ipath driver to set write_combining on its PIO
|
|
|
8 |
# buffers. To correct the problem, remap the lower memory region in various
|
|
|
9 |
# chunks up to the start of the punch out hole(s), then delete the punch out
|
|
|
10 |
# hole(s) entirely as they aren't needed any more. That way, ib_ipath will
|
|
|
11 |
# be able to set write_combining on its PIO memory access region.
|
|
|
12 |
|
|
|
13 |
BEGIN {
|
|
|
14 |
regs = 0
|
|
|
15 |
}
|
|
|
16 |
|
|
|
17 |
function check_base(mem)
|
|
|
18 |
{
|
|
|
19 |
printf "Base memory data: base=0x%08x, size=0x%x\n", base[mem], size[mem] > "/dev/stderr"
|
|
|
20 |
if (size[mem] < (512 * 1024 * 1024))
|
|
|
21 |
return 0
|
|
|
22 |
if (type[mem] != "write-back")
|
|
|
23 |
return 0
|
|
|
24 |
if (base[mem] >= (4 * 1024 * 1024 * 1024))
|
|
|
25 |
return 0
|
|
|
26 |
return 1
|
|
|
27 |
}
|
|
|
28 |
|
|
|
29 |
function check_hole(hole)
|
|
|
30 |
{
|
|
|
31 |
printf "Hole data: base=0x%08x, size=0x%x\n", base[hole], size[hole] > "/dev/stderr"
|
|
|
32 |
if (size[hole] > (1 * 1024 * 1024 * 1024))
|
|
|
33 |
return 0
|
|
|
34 |
if (type[hole] != "uncachable")
|
|
|
35 |
return 0
|
|
|
36 |
if ((base[hole] + size[hole]) > (4 * 1024 * 1024 * 1024))
|
|
|
37 |
return 0
|
|
|
38 |
return 1
|
|
|
39 |
}
|
|
|
40 |
|
|
|
41 |
function build_entries(start, end, new_base, new_size, tmp_base)
|
|
|
42 |
{
|
|
|
43 |
# mtrr registers require alignment of blocks, so a 256MB chunk must
|
|
|
44 |
# be 256MB aligned. Additionally, all blocks must be a power of 2
|
|
|
45 |
# in size. So, do the largest power of two size that we can and
|
|
|
46 |
# still have start + block <= end, rinse and repeat.
|
|
|
47 |
tmp_base = start
|
|
|
48 |
do {
|
|
|
49 |
new_base = tmp_base
|
|
|
50 |
new_size = 4096
|
|
|
51 |
while (((new_base + new_size) < end) &&
|
|
|
52 |
((new_base % new_size) == 0))
|
|
|
53 |
new_size = lshift(new_size, 1)
|
|
|
54 |
if (((new_base + new_size) > end) ||
|
|
|
55 |
((new_base % new_size) != 0))
|
|
|
56 |
new_size = rshift(new_size, 1)
|
|
|
57 |
printf "base=0x%x size=0x%x type=%s\n",
|
|
|
58 |
new_base, new_size, type[mem] > "/dev/stderr"
|
|
|
59 |
printf "base=0x%x size=0x%x type=%s\n",
|
|
|
60 |
new_base, new_size, type[mem] > "/proc/mtrr"
|
|
|
61 |
fflush("")
|
|
|
62 |
tmp_base = new_base + new_size
|
|
|
63 |
} while (tmp_base < end)
|
|
|
64 |
}
|
|
|
65 |
|
|
|
66 |
{
|
|
|
67 |
gsub("^reg", "")
|
|
|
68 |
gsub(": base=", " ")
|
|
|
69 |
gsub(" [(].*), size=", " ")
|
|
|
70 |
gsub(": ", " ")
|
|
|
71 |
gsub(", count=.*$", "")
|
|
|
72 |
register[regs] = strtonum($1)
|
|
|
73 |
base[regs] = strtonum($2)
|
|
|
74 |
size[regs] = strtonum($3)
|
|
|
75 |
human_size[regs] = size[regs]
|
|
|
76 |
if (match($3, "MB")) { size[regs] *= 1024*1024; mult[regs] = "MB" }
|
|
|
77 |
else { size[regs] *= 1024; mult[regs] = "KB" }
|
|
|
78 |
type[regs] = $4
|
|
|
79 |
enabled[regs] = 1
|
|
|
80 |
end[regs] = base[regs] + size[regs]
|
|
|
81 |
regs++
|
|
|
82 |
}
|
|
|
83 |
|
|
|
84 |
END {
|
|
|
85 |
# First we need to find our base memory region. We only care about
|
|
|
86 |
# the memory register that starts at base 0. This is the only one
|
|
|
87 |
# that we can reliably know is our global memory region, and the
|
|
|
88 |
# only one that we can reliably check against overlaps. It's entirely
|
|
|
89 |
# possible that any memory region not starting at 0 and having an
|
|
|
90 |
# overlap with another memory region is in fact intentional and we
|
|
|
91 |
# shouldn't touch it.
|
|
|
92 |
for(i=0; i<regs; i++)
|
|
|
93 |
if (base[i] == 0)
|
|
|
94 |
break
|
|
|
95 |
# Did we get a valid base register?
|
|
|
96 |
if (i == regs)
|
|
|
97 |
exit 1
|
|
|
98 |
mem = i
|
|
|
99 |
if (!check_base(mem))
|
|
|
100 |
exit 1
|
|
|
101 |
|
|
|
102 |
cur_hole = 0
|
|
|
103 |
for(i=0; i<regs; i++) {
|
|
|
104 |
if (i == mem)
|
|
|
105 |
continue
|
|
|
106 |
if (base[i] < end[mem] && check_hole(i))
|
|
|
107 |
holes[cur_hole++] = i
|
|
|
108 |
}
|
|
|
109 |
if (cur_hole == 0) {
|
|
|
110 |
print "Nothing to do" > "/dev/stderr"
|
|
|
111 |
exit 1
|
|
|
112 |
}
|
|
|
113 |
printf "Found %d punch-out holes\n", cur_hole > "/dev/stderr"
|
|
|
114 |
|
|
|
115 |
# We need to sort the holes according to base address
|
|
|
116 |
for(j = 0; j < cur_hole - 1; j++) {
|
|
|
117 |
for(i = cur_hole - 1; i > j; i--) {
|
|
|
118 |
if(base[holes[i]] < base[holes[i-1]]) {
|
|
|
119 |
tmp = holes[i]
|
|
|
120 |
holes[i] = holes[i-1]
|
|
|
121 |
holes[i-1] = tmp
|
|
|
122 |
}
|
|
|
123 |
}
|
|
|
124 |
}
|
|
|
125 |
# OK, the common case would be that the BIOS is mapping holes out
|
|
|
126 |
# of the 4GB memory range, and that our hole(s) are consecutive and
|
|
|
127 |
# that our holes and our memory region end at the same place. However,
|
|
|
128 |
# things like machines with 8GB of RAM or more can foul up these
|
|
|
129 |
# common traits.
|
|
|
130 |
#
|
|
|
131 |
# So, our modus operandi is to disable all of the memory/hole regions
|
|
|
132 |
# to start, then build new base memory zones that in the end add
|
|
|
133 |
# up to the same as our original zone minus the holes. We know that
|
|
|
134 |
# we will never have a hole listed here that belongs to a valid
|
|
|
135 |
# hole punched in a write-combining memory region because you can't
|
|
|
136 |
# overlay write-combining on top of write-back and we know our base
|
|
|
137 |
# memory region is write-back, so in order for this hole to overlap
|
|
|
138 |
# our base memory region it can't be also overlapping a write-combining
|
|
|
139 |
# region.
|
|
|
140 |
printf "disable=%d\n", register[mem] > "/dev/stderr"
|
|
|
141 |
printf "disable=%d\n", register[mem] > "/proc/mtrr"
|
|
|
142 |
fflush("")
|
|
|
143 |
enabled[mem] = 0
|
|
|
144 |
for(i=0; i < cur_hole; i++) {
|
|
|
145 |
printf "disable=%d\n", register[holes[i]] > "/dev/stderr"
|
|
|
146 |
printf "disable=%d\n", register[holes[i]] > "/proc/mtrr"
|
|
|
147 |
fflush("")
|
|
|
148 |
enabled[holes[i]] = 0
|
|
|
149 |
}
|
|
|
150 |
build_entries(base[mem], base[holes[0]])
|
|
|
151 |
for(i=0; i < cur_hole - 1; i++)
|
|
|
152 |
if (base[holes[i+1]] > end[holes[i]])
|
|
|
153 |
build_entries(end[holes[i]], base[holes[i+1]])
|
|
|
154 |
if (end[mem] > end[holes[i]])
|
|
|
155 |
build_entries(end[holes[i]], end[mem])
|
|
|
156 |
# We changed up the mtrr regs, so signal to the rdma script to
|
|
|
157 |
# reload modules that need the mtrr regs to be right.
|
|
|
158 |
exit 0
|
|
|
159 |
}
|
|
|
160 |
|