1 .. SPDX-License-Identifier: GPL-2.0
2
3 =============
4 Kernel Stacks
5 =============
6
7 Kernel stacks on x86-64 bit
8 ===========================
9
10 Most of the text from Keith Owens, hacked by A
11
12 x86_64 page size (PAGE_SIZE) is 4K.
13
14 Like all other architectures, x86_64 has a ker
15 active thread. These thread stacks are THREAD
16 These stacks contain useful data as long as a
17 zombie. While the thread is in user space the
18 except for the thread_info structure at the bo
19
20 In addition to the per thread stacks, there ar
21 associated with each CPU. These stacks are on
22 is in control on that CPU; when a CPU returns
23 specialized stacks contain no useful data. Th
24
25 * Interrupt stack. IRQ_STACK_SIZE
26
27 Used for external hardware interrupts. If t
28 hardware interrupt (i.e. not a nested hardwa
29 kernel switches from the current task to the
30 the split thread and interrupt stacks on i38
31 for kernel interrupt processing without havi
32 of every per thread stack.
33
34 The interrupt stack is also used when proces
35
36 Switching to the kernel interrupt stack is don
37 per CPU interrupt nest counter. This is needed
38 hardware stacks cannot nest without races.
39
40 x86_64 also has a feature which is not availab
41 to automatically switch to a new stack for des
42 double fault or NMI, which makes it easier to
43 events on x86_64. This feature is called the
44 (IST). There can be up to 7 IST entries per C
45 index into the Task State Segment (TSS). The I
46 point to dedicated stacks; each stack can be a
47
48 An IST is selected by a non-zero value in the
49 interrupt-gate descriptor. When an interrupt
50 loads such a descriptor, the hardware automati
51 pointer based on the IST value, then invokes t
52 the interrupt came from user mode, then the in
53 will switch back to the per-thread stack. If
54 nested IST interrupts then the handler must ad
55 entry to and exit from the interrupt handler.
56 done, e.g. for debug exceptions.)
57
58 Events with different IST codes (i.e. with dif
59 nested. For example, a debug interrupt can sa
60 NMI. arch/x86_64/kernel/entry.S::paranoidentr
61 pointers on entry to and exit from all IST eve
62 IST events with the same code to be nested. H
63 stack size allocated to an IST assumes no nest
64 If that assumption is ever broken then the sta
65
66 The currently assigned IST stacks are:
67
68 * ESTACK_DF. EXCEPTION_STKSZ (PAGE_SIZE).
69
70 Used for interrupt 8 - Double Fault Exceptio
71
72 Invoked when handling one exception causes a
73 when the kernel is very confused (e.g. kerne
74 Using a separate stack allows the kernel to
75 in many cases to still output an oops.
76
77 * ESTACK_NMI. EXCEPTION_STKSZ (PAGE_SIZE).
78
79 Used for non-maskable interrupts (NMI).
80
81 NMI can be delivered at any time, including
82 middle of switching stacks. Using IST for N
83 assumptions about the previous state of the
84
85 * ESTACK_DB. EXCEPTION_STKSZ (PAGE_SIZE).
86
87 Used for hardware debug interrupts (interrup
88 debug interrupts (INT3).
89
90 When debugging a kernel, debug interrupts (b
91 software) can occur at any time. Using IST
92 avoids making assumptions about the previous
93 stack.
94
95 To handle nested #DB correctly there exist t
96 #DB entry the IST stackpointer for #DB is sw
97 so a nested #DB starts from a clean stack. T
98 the IST stackpointer to a guard hole to catc
99
100 * ESTACK_MCE. EXCEPTION_STKSZ (PAGE_SIZE).
101
102 Used for interrupt 18 - Machine Check Except
103
104 MCE can be delivered at any time, including
105 middle of switching stacks. Using IST for M
106 assumptions about the previous state of the
107
108 For more details see the Intel IA32 or AMD AMD
109
110
111 Printing backtraces on x86
112 ==========================
113
114 The question about the '?' preceding function
115 keeps popping up, here's an indepth explanatio
116 stares at print_context_stack() and the whole
117 arch/x86/kernel/dumpstack.c.
118
119 Adapted from Ingo's mail, Message-ID: <20150521
120
121 We always scan the full kernel stack for retur
122 the kernel stack(s) [1]_, from stack top to st
123 anything that 'looks like' a kernel text addre
124
125 If it fits into the frame pointer chain, we pr
126 mark, knowing that it's part of the real backt
127
128 If the address does not fit into our expected
129 still print it, but we print a '?'. It can mea
130
131 - either the address is not part of the call
132 values on the kernel stack, from earlier fu
133 the common case.
134
135 - or it is part of the call chain, but the fr
136 up properly within the function, so we don'
137
138 This way we will always print out the real cal
139 entries), regardless of whether the frame poin
140 or not - but in most cases we'll get the call
141 entries printed are strictly in stack order, s
142 information from that as well.
143
144 The most important property of this method is
145 information: we always strive to print _all_ a
146 that look like kernel text addresses, so if de
147 we still print out the real call chain as well
148 marks than ideal.
149
150 .. [1] For things like IRQ and IST stacks, we
151 the right order, and try to cross from
152 reconstructing the call chain. This wor
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.