1		In-kernel memory-mapped I/O tracing
2
3
4Home page and links to optional user space tools:
5
6	http://nouveau.freedesktop.org/wiki/MmioTrace
7
8MMIO tracing was originally developed by Intel around 2003 for their Fault
9Injection Test Harness. In Dec 2006 - Jan 2007, using the code from Intel,
10Jeff Muizelaar created a tool for tracing MMIO accesses with the Nouveau
11project in mind. Since then many people have contributed.
12
13Mmiotrace was built for reverse engineering any memory-mapped IO device with
14the Nouveau project as the first real user. Only x86 and x86_64 architectures
15are supported.
16
17Out-of-tree mmiotrace was originally modified for mainline inclusion and
18ftrace framework by Pekka Paalanen <pq@iki.fi>.
19
20
21Preparation
22-----------
23
24Mmiotrace feature is compiled in by the CONFIG_MMIOTRACE option. Tracing is
25disabled by default, so it is safe to have this set to yes. SMP systems are
26supported, but tracing is unreliable and may miss events if more than one CPU
27is on-line, therefore mmiotrace takes all but one CPU off-line during run-time
28activation. You can re-enable CPUs by hand, but you have been warned, there
29is no way to automatically detect if you are losing events due to CPUs racing.
30
31
32Usage Quick Reference
33---------------------
34
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 &
38Start X or whatever.
39$ echo "X is up" > /sys/kernel/debug/tracing/trace_marker
40$ echo nop > /sys/kernel/debug/tracing/current_tracer
41Check for lost events.
42
43
44Usage
45-----
46
47Make sure debugfs is mounted to /sys/kernel/debug.
48If not (requires root privileges):
49$ mount -t debugfs debugfs /sys/kernel/debug
50
51Check that the driver you are about to trace is not loaded.
52
53Activate mmiotrace (requires root privileges):
54$ echo mmiotrace > /sys/kernel/debug/tracing/current_tracer
55
56Start storing the trace:
57$ cat /sys/kernel/debug/tracing/trace_pipe > mydump.txt &
58The 'cat' process should stay running (sleeping) in the background.
59
60Load the driver you want to trace and use it. Mmiotrace will only catch MMIO
61accesses to areas that are ioremapped while mmiotrace is active.
62
63During tracing you can place comments (markers) into the trace by
64$ echo "X is up" > /sys/kernel/debug/tracing/trace_marker
65This makes it easier to see which part of the (huge) trace corresponds to
66which action. It is recommended to place descriptive markers about what you
67do.
68
69Shut down mmiotrace (requires root privileges):
70$ echo nop > /sys/kernel/debug/tracing/current_tracer
71The 'cat' process exits. If it does not, kill it by issuing 'fg' command and
72pressing ctrl+c.
73
74Check that mmiotrace did not lose events due to a buffer filling up. Either
75$ grep -i lost mydump.txt
76which tells you exactly how many events were lost, or use
77$ dmesg
78to view your kernel log and look for "mmiotrace has lost events" warning. If
79events were lost, the trace is incomplete. You should enlarge the buffers and
80try again. Buffers are enlarged by first seeing how large the current buffers
81are:
82$ cat /sys/kernel/debug/tracing/buffer_size_kb
83gives you a number. Approximately double this number and write it back, for
84instance:
85$ echo 128000 > /sys/kernel/debug/tracing/buffer_size_kb
86Then start again from the top.
87
88If you are doing a trace for a driver project, e.g. Nouveau, you should also
89do the following before sending your results:
90$ lspci -vvv > lspci.txt
91$ dmesg > dmesg.txt
92$ tar zcf pciid-nick-mmiotrace.tar.gz mydump.txt lspci.txt dmesg.txt
93and 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
95under investigation and your nickname.
96
97
98How Mmiotrace Works
99-------------------
100
101Access to hardware IO-memory is gained by mapping addresses from PCI bus by
102calling one of the ioremap_*() functions. Mmiotrace is hooked into the
103__ioremap() function and gets called whenever a mapping is created. Mapping is
104an event that is recorded into the trace log. Note that ISA range mappings
105are not caught, since the mapping always exists and is returned directly.
106
107MMIO accesses are recorded via page faults. Just before __ioremap() returns,
108the mapped pages are marked as not present. Any access to the pages causes a
109fault. The page fault handler calls mmiotrace to handle the fault. Mmiotrace
110marks the page present, sets TF flag to achieve single stepping and exits the
111fault handler. The instruction that faulted is executed and debug trap is
112entered. Here mmiotrace again marks the page as not present. The instruction
113is decoded to get the type of operation (read/write), data width and the value
114read or written. These are stored to the trace log.
115
116Setting the page present in the page fault handler has a race condition on SMP
117machines. During the single stepping other CPUs may run freely on that page
118and events can be missed without a notice. Re-enabling other CPUs during
119tracing is discouraged.
120
121
122Trace Log Format
123----------------
124
125The raw log is text and easily filtered with e.g. grep and awk. One record is
126one line in the log. A record starts with a keyword, followed by keyword-
127dependent arguments. Arguments are separated by a space, or continue until the
128end of line. The format for version 20070824 is as follows:
129
130Explanation	Keyword	Space-separated arguments
131---------------------------------------------------------------------------
132
133read event	R	width, timestamp, map id, physical, value, PC, PID
134write event	W	width, timestamp, map id, physical, value, PC, PID
135ioremap event	MAP	timestamp, map id, physical, virtual, length, PC, PID
136iounmap event	UNMAP	timestamp, map id, PC, PID
137marker		MARK	timestamp, text
138version		VERSION	the string "20070824"
139info for reader	LSPCI	one line from lspci -v
140PCI address map	PCIDEV	space-separated /proc/bus/pci/devices data
141unk. opcode	UNKNOWN	timestamp, map id, physical, data, PC, PID
142
143Timestamp is in seconds with decimals. Physical is a PCI bus address, virtual
144is a kernel virtual address. Width is the data width in bytes and value is the
145data value. Map id is an arbitrary id number identifying the mapping that was
146used in an operation. PC is the program counter and PID is process id. PC is
147zero if it is not recorded. PID is always zero as tracing MMIO accesses
148originating in user space memory is not yet supported.
149
150For instance, the following awk filter will pass all 32-bit writes that target
151physical addresses in the range [0xfb73ce40, 0xfb800000[
152
153$ awk '/W 4 / { adr=strtonum($5); if (adr >= 0xfb73ce40 &&
154adr < 0xfb800000) print; }'
155
156
157Tools for Developers
158--------------------
159
160The 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
163
164
165