1 In-kernel memory-mapped I/O tracing
4 Home page and links to optional user space tools:
6 http://nouveau.freedesktop.org/wiki/MmioTrace
8 MMIO tracing was originally developed by Intel around 2003 for their Fault
9 Injection Test Harness. In Dec 2006 - Jan 2007, using the code from Intel,
10 Jeff Muizelaar created a tool for tracing MMIO accesses with the Nouveau
11 project in mind. Since then many people have contributed.
13 Mmiotrace was built for reverse engineering any memory-mapped IO device with
14 the Nouveau project as the first real user. Only x86 and x86_64 architectures
17 Out-of-tree mmiotrace was originally modified for mainline inclusion and
18 ftrace framework by Pekka Paalanen <pq@iki.fi>.
24 Mmiotrace feature is compiled in by the CONFIG_MMIOTRACE option. Tracing is
25 disabled by default, so it is safe to have this set to yes. SMP systems are
26 supported, but tracing is unreliable and may miss events if more than one CPU
27 is on-line, therefore mmiotrace takes all but one CPU off-line during run-time
28 activation. You can re-enable CPUs by hand, but you have been warned, there
29 is no way to automatically detect if you are losing events due to CPUs racing.
35 $ mount -t debugfs debugfs /sys/kernel/debug
36 $ echo mmiotrace > /sys/kernel/debug/tracing/current_tracer
37 $ cat /sys/kernel/debug/tracing/trace_pipe > mydump.txt &
39 $ echo "X is up" > /sys/kernel/debug/tracing/trace_marker
40 $ echo nop > /sys/kernel/debug/tracing/current_tracer
41 Check for lost events.
47 Make sure debugfs is mounted to /sys/kernel/debug.
48 If not (requires root privileges):
49 $ mount -t debugfs debugfs /sys/kernel/debug
51 Check that the driver you are about to trace is not loaded.
53 Activate mmiotrace (requires root privileges):
54 $ echo mmiotrace > /sys/kernel/debug/tracing/current_tracer
56 Start storing the trace:
57 $ cat /sys/kernel/debug/tracing/trace_pipe > mydump.txt &
58 The 'cat' process should stay running (sleeping) in the background.
60 Load the driver you want to trace and use it. Mmiotrace will only catch MMIO
61 accesses to areas that are ioremapped while mmiotrace is active.
63 During tracing you can place comments (markers) into the trace by
64 $ echo "X is up" > /sys/kernel/debug/tracing/trace_marker
65 This makes it easier to see which part of the (huge) trace corresponds to
66 which action. It is recommended to place descriptive markers about what you
69 Shut down mmiotrace (requires root privileges):
70 $ echo nop > /sys/kernel/debug/tracing/current_tracer
71 The 'cat' process exits. If it does not, kill it by issuing 'fg' command and
74 Check that mmiotrace did not lose events due to a buffer filling up. Either
75 $ grep -i lost mydump.txt
76 which tells you exactly how many events were lost, or use
78 to view your kernel log and look for "mmiotrace has lost events" warning. If
79 events were lost, the trace is incomplete. You should enlarge the buffers and
80 try again. Buffers are enlarged by first seeing how large the current buffers
82 $ cat /sys/kernel/debug/tracing/buffer_size_kb
83 gives you a number. Approximately double this number and write it back, for
85 $ echo 128000 > /sys/kernel/debug/tracing/buffer_size_kb
86 Then start again from the top.
88 If you are doing a trace for a driver project, e.g. Nouveau, you should also
89 do the following before sending your results:
90 $ lspci -vvv > lspci.txt
92 $ tar zcf pciid-nick-mmiotrace.tar.gz mydump.txt lspci.txt dmesg.txt
93 and then send the .tar.gz file. The trace compresses considerably. Replace
94 "pciid" and "nick" with the PCI ID or model name of your piece of hardware
95 under investigation and your nickname.
101 Access to hardware IO-memory is gained by mapping addresses from PCI bus by
102 calling one of the ioremap_*() functions. Mmiotrace is hooked into the
103 __ioremap() function and gets called whenever a mapping is created. Mapping is
104 an event that is recorded into the trace log. Note that ISA range mappings
105 are not caught, since the mapping always exists and is returned directly.
107 MMIO accesses are recorded via page faults. Just before __ioremap() returns,
108 the mapped pages are marked as not present. Any access to the pages causes a
109 fault. The page fault handler calls mmiotrace to handle the fault. Mmiotrace
110 marks the page present, sets TF flag to achieve single stepping and exits the
111 fault handler. The instruction that faulted is executed and debug trap is
112 entered. Here mmiotrace again marks the page as not present. The instruction
113 is decoded to get the type of operation (read/write), data width and the value
114 read or written. These are stored to the trace log.
116 Setting the page present in the page fault handler has a race condition on SMP
117 machines. During the single stepping other CPUs may run freely on that page
118 and events can be missed without a notice. Re-enabling other CPUs during
119 tracing is discouraged.
125 The raw log is text and easily filtered with e.g. grep and awk. One record is
126 one line in the log. A record starts with a keyword, followed by keyword-
127 dependent arguments. Arguments are separated by a space, or continue until the
128 end of line. The format for version 20070824 is as follows:
130 Explanation Keyword Space-separated arguments
131 ---------------------------------------------------------------------------
133 read event R width, timestamp, map id, physical, value, PC, PID
134 write event W width, timestamp, map id, physical, value, PC, PID
135 ioremap event MAP timestamp, map id, physical, virtual, length, PC, PID
136 iounmap event UNMAP timestamp, map id, PC, PID
137 marker MARK timestamp, text
138 version VERSION the string "20070824"
139 info for reader LSPCI one line from lspci -v
140 PCI address map PCIDEV space-separated /proc/bus/pci/devices data
141 unk. opcode UNKNOWN timestamp, map id, physical, data, PC, PID
143 Timestamp is in seconds with decimals. Physical is a PCI bus address, virtual
144 is a kernel virtual address. Width is the data width in bytes and value is the
145 data value. Map id is an arbitrary id number identifying the mapping that was
146 used in an operation. PC is the program counter and PID is process id. PC is
147 zero if it is not recorded. PID is always zero as tracing MMIO accesses
148 originating in user space memory is not yet supported.
150 For instance, the following awk filter will pass all 32-bit writes that target
151 physical addresses in the range [0xfb73ce40, 0xfb800000[
153 $ awk '/W 4 / { adr=strtonum($5); if (adr >= 0xfb73ce40 &&
154 adr < 0xfb800000) print; }'
160 The user space tools include utilities for:
161 - replacing numeric addresses and values with hardware register names
162 - replaying MMIO logs, i.e., re-executing the recorded writes