1===================================
2In-kernel memory-mapped I/O tracing
3===================================
4
5
6Home page and links to optional user space tools:
7
8	http://nouveau.freedesktop.org/wiki/MmioTrace
9
10MMIO tracing was originally developed by Intel around 2003 for their Fault
11Injection Test Harness. In Dec 2006 - Jan 2007, using the code from Intel,
12Jeff Muizelaar created a tool for tracing MMIO accesses with the Nouveau
13project in mind. Since then many people have contributed.
14
15Mmiotrace was built for reverse engineering any memory-mapped IO device with
16the Nouveau project as the first real user. Only x86 and x86_64 architectures
17are supported.
18
19Out-of-tree mmiotrace was originally modified for mainline inclusion and
20ftrace framework by Pekka Paalanen <pq@iki.fi>.
21
22
23Preparation
24-----------
25
26Mmiotrace feature is compiled in by the CONFIG_MMIOTRACE option. Tracing is
27disabled by default, so it is safe to have this set to yes. SMP systems are
28supported, but tracing is unreliable and may miss events if more than one CPU
29is on-line, therefore mmiotrace takes all but one CPU off-line during run-time
30activation. You can re-enable CPUs by hand, but you have been warned, there
31is no way to automatically detect if you are losing events due to CPUs racing.
32
33
34Usage Quick Reference
35---------------------
36::
37
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 &
41	Start X or whatever.
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.
45
46
47Usage
48-----
49
50Make sure debugfs is mounted to /sys/kernel/debug.
51If not (requires root privileges)::
52
53	$ mount -t debugfs debugfs /sys/kernel/debug
54
55Check that the driver you are about to trace is not loaded.
56
57Activate mmiotrace (requires root privileges)::
58
59	$ echo mmiotrace > /sys/kernel/debug/tracing/current_tracer
60
61Start storing the trace::
62
63	$ cat /sys/kernel/debug/tracing/trace_pipe > mydump.txt &
64
65The 'cat' process should stay running (sleeping) in the background.
66
67Load the driver you want to trace and use it. Mmiotrace will only catch MMIO
68accesses to areas that are ioremapped while mmiotrace is active.
69
70During tracing you can place comments (markers) into the trace by
71$ echo "X is up" > /sys/kernel/debug/tracing/trace_marker
72This makes it easier to see which part of the (huge) trace corresponds to
73which action. It is recommended to place descriptive markers about what you
74do.
75
76Shut down mmiotrace (requires root privileges)::
77
78	$ echo nop > /sys/kernel/debug/tracing/current_tracer
79
80The 'cat' process exits. If it does not, kill it by issuing 'fg' command and
81pressing ctrl+c.
82
83Check that mmiotrace did not lose events due to a buffer filling up. Either::
84
85	$ grep -i lost mydump.txt
86
87which tells you exactly how many events were lost, or use::
88
89	$ dmesg
90
91to view your kernel log and look for "mmiotrace has lost events" warning. If
92events were lost, the trace is incomplete. You should enlarge the buffers and
93try again. Buffers are enlarged by first seeing how large the current buffers
94are::
95
96	$ cat /sys/kernel/debug/tracing/buffer_size_kb
97
98gives you a number. Approximately double this number and write it back, for
99instance::
100
101	$ echo 128000 > /sys/kernel/debug/tracing/buffer_size_kb
102
103Then start again from the top.
104
105If you are doing a trace for a driver project, e.g. Nouveau, you should also
106do the following before sending your results::
107
108	$ lspci -vvv > lspci.txt
109	$ dmesg > dmesg.txt
110	$ tar zcf pciid-nick-mmiotrace.tar.gz mydump.txt lspci.txt dmesg.txt
111
112and 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
114under investigation and your nickname.
115
116
117How Mmiotrace Works
118-------------------
119
120Access to hardware IO-memory is gained by mapping addresses from PCI bus by
121calling one of the ioremap_*() functions. Mmiotrace is hooked into the
122__ioremap() function and gets called whenever a mapping is created. Mapping is
123an event that is recorded into the trace log. Note that ISA range mappings
124are not caught, since the mapping always exists and is returned directly.
125
126MMIO accesses are recorded via page faults. Just before __ioremap() returns,
127the mapped pages are marked as not present. Any access to the pages causes a
128fault. The page fault handler calls mmiotrace to handle the fault. Mmiotrace
129marks the page present, sets TF flag to achieve single stepping and exits the
130fault handler. The instruction that faulted is executed and debug trap is
131entered. Here mmiotrace again marks the page as not present. The instruction
132is decoded to get the type of operation (read/write), data width and the value
133read or written. These are stored to the trace log.
134
135Setting the page present in the page fault handler has a race condition on SMP
136machines. During the single stepping other CPUs may run freely on that page
137and events can be missed without a notice. Re-enabling other CPUs during
138tracing is discouraged.
139
140
141Trace Log Format
142----------------
143
144The raw log is text and easily filtered with e.g. grep and awk. One record is
145one line in the log. A record starts with a keyword, followed by keyword-
146dependent arguments. Arguments are separated by a space, or continue until the
147end of line. The format for version 20070824 is as follows:
148
149Explanation	Keyword	Space-separated arguments
150---------------------------------------------------------------------------
151
152read event	R	width, timestamp, map id, physical, value, PC, PID
153write event	W	width, timestamp, map id, physical, value, PC, PID
154ioremap event	MAP	timestamp, map id, physical, virtual, length, PC, PID
155iounmap event	UNMAP	timestamp, map id, PC, PID
156marker		MARK	timestamp, text
157version		VERSION	the string "20070824"
158info for reader	LSPCI	one line from lspci -v
159PCI address map	PCIDEV	space-separated /proc/bus/pci/devices data
160unk. opcode	UNKNOWN	timestamp, map id, physical, data, PC, PID
161
162Timestamp is in seconds with decimals. Physical is a PCI bus address, virtual
163is a kernel virtual address. Width is the data width in bytes and value is the
164data value. Map id is an arbitrary id number identifying the mapping that was
165used in an operation. PC is the program counter and PID is process id. PC is
166zero if it is not recorded. PID is always zero as tracing MMIO accesses
167originating in user space memory is not yet supported.
168
169For instance, the following awk filter will pass all 32-bit writes that target
170physical addresses in the range [0xfb73ce40, 0xfb800000]
171::
172
173	$ awk '/W 4 / { adr=strtonum($5); if (adr >= 0xfb73ce40 &&
174	adr < 0xfb800000) print; }'
175
176
177Tools for Developers
178--------------------
179
180The 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
183
184
185