1 ===================================
2 In-kernel memory-mapped I/O tracing
3 ===================================
6 Home page and links to optional user space tools:
8 http://nouveau.freedesktop.org/wiki/MmioTrace
10 MMIO tracing was originally developed by Intel around 2003 for their Fault
11 Injection Test Harness. In Dec 2006 - Jan 2007, using the code from Intel,
12 Jeff Muizelaar created a tool for tracing MMIO accesses with the Nouveau
13 project in mind. Since then many people have contributed.
15 Mmiotrace was built for reverse engineering any memory-mapped IO device with
16 the Nouveau project as the first real user. Only x86 and x86_64 architectures
19 Out-of-tree mmiotrace was originally modified for mainline inclusion and
20 ftrace framework by Pekka Paalanen <pq@iki.fi>.
26 Mmiotrace feature is compiled in by the CONFIG_MMIOTRACE option. Tracing is
27 disabled by default, so it is safe to have this set to yes. SMP systems are
28 supported, but tracing is unreliable and may miss events if more than one CPU
29 is on-line, therefore mmiotrace takes all but one CPU off-line during run-time
30 activation. You can re-enable CPUs by hand, but you have been warned, there
31 is no way to automatically detect if you are losing events due to CPUs racing.
38 $ mount -t debugfs debugfs /sys/kernel/debug
39 $ echo mmiotrace > /sys/kernel/debug/tracing/current_tracer
40 $ cat /sys/kernel/debug/tracing/trace_pipe > mydump.txt &
42 $ echo "X is up" > /sys/kernel/debug/tracing/trace_marker
43 $ echo nop > /sys/kernel/debug/tracing/current_tracer
44 Check for lost events.
50 Make sure debugfs is mounted to /sys/kernel/debug.
51 If not (requires root privileges)::
53 $ mount -t debugfs debugfs /sys/kernel/debug
55 Check that the driver you are about to trace is not loaded.
57 Activate mmiotrace (requires root privileges)::
59 $ echo mmiotrace > /sys/kernel/debug/tracing/current_tracer
61 Start storing the trace::
63 $ cat /sys/kernel/debug/tracing/trace_pipe > mydump.txt &
65 The 'cat' process should stay running (sleeping) in the background.
67 Load the driver you want to trace and use it. Mmiotrace will only catch MMIO
68 accesses to areas that are ioremapped while mmiotrace is active.
70 During tracing you can place comments (markers) into the trace by
71 $ echo "X is up" > /sys/kernel/debug/tracing/trace_marker
72 This makes it easier to see which part of the (huge) trace corresponds to
73 which action. It is recommended to place descriptive markers about what you
76 Shut down mmiotrace (requires root privileges)::
78 $ echo nop > /sys/kernel/debug/tracing/current_tracer
80 The 'cat' process exits. If it does not, kill it by issuing 'fg' command and
83 Check that mmiotrace did not lose events due to a buffer filling up. Either::
85 $ grep -i lost mydump.txt
87 which tells you exactly how many events were lost, or use::
91 to view your kernel log and look for "mmiotrace has lost events" warning. If
92 events were lost, the trace is incomplete. You should enlarge the buffers and
93 try again. Buffers are enlarged by first seeing how large the current buffers
96 $ cat /sys/kernel/debug/tracing/buffer_size_kb
98 gives you a number. Approximately double this number and write it back, for
101 $ echo 128000 > /sys/kernel/debug/tracing/buffer_size_kb
103 Then start again from the top.
105 If you are doing a trace for a driver project, e.g. Nouveau, you should also
106 do the following before sending your results::
108 $ lspci -vvv > lspci.txt
110 $ tar zcf pciid-nick-mmiotrace.tar.gz mydump.txt lspci.txt dmesg.txt
112 and then send the .tar.gz file. The trace compresses considerably. Replace
113 "pciid" and "nick" with the PCI ID or model name of your piece of hardware
114 under investigation and your nickname.
120 Access to hardware IO-memory is gained by mapping addresses from PCI bus by
121 calling one of the ioremap_*() functions. Mmiotrace is hooked into the
122 __ioremap() function and gets called whenever a mapping is created. Mapping is
123 an event that is recorded into the trace log. Note that ISA range mappings
124 are not caught, since the mapping always exists and is returned directly.
126 MMIO accesses are recorded via page faults. Just before __ioremap() returns,
127 the mapped pages are marked as not present. Any access to the pages causes a
128 fault. The page fault handler calls mmiotrace to handle the fault. Mmiotrace
129 marks the page present, sets TF flag to achieve single stepping and exits the
130 fault handler. The instruction that faulted is executed and debug trap is
131 entered. Here mmiotrace again marks the page as not present. The instruction
132 is decoded to get the type of operation (read/write), data width and the value
133 read or written. These are stored to the trace log.
135 Setting the page present in the page fault handler has a race condition on SMP
136 machines. During the single stepping other CPUs may run freely on that page
137 and events can be missed without a notice. Re-enabling other CPUs during
138 tracing is discouraged.
144 The raw log is text and easily filtered with e.g. grep and awk. One record is
145 one line in the log. A record starts with a keyword, followed by keyword-
146 dependent arguments. Arguments are separated by a space, or continue until the
147 end of line. The format for version 20070824 is as follows:
149 Explanation Keyword Space-separated arguments
150 ---------------------------------------------------------------------------
152 read event R width, timestamp, map id, physical, value, PC, PID
153 write event W width, timestamp, map id, physical, value, PC, PID
154 ioremap event MAP timestamp, map id, physical, virtual, length, PC, PID
155 iounmap event UNMAP timestamp, map id, PC, PID
156 marker MARK timestamp, text
157 version VERSION the string "20070824"
158 info for reader LSPCI one line from lspci -v
159 PCI address map PCIDEV space-separated /proc/bus/pci/devices data
160 unk. opcode UNKNOWN timestamp, map id, physical, data, PC, PID
162 Timestamp is in seconds with decimals. Physical is a PCI bus address, virtual
163 is a kernel virtual address. Width is the data width in bytes and value is the
164 data value. Map id is an arbitrary id number identifying the mapping that was
165 used in an operation. PC is the program counter and PID is process id. PC is
166 zero if it is not recorded. PID is always zero as tracing MMIO accesses
167 originating in user space memory is not yet supported.
169 For instance, the following awk filter will pass all 32-bit writes that target
170 physical addresses in the range [0xfb73ce40, 0xfb800000]
173 $ awk '/W 4 / { adr=strtonum($5); if (adr >= 0xfb73ce40 &&
174 adr < 0xfb800000) print; }'
180 The user space tools include utilities for:
181 - replacing numeric addresses and values with hardware register names
182 - replaying MMIO logs, i.e., re-executing the recorded writes