1.. SPDX-License-Identifier: GPL-2.0
2
3=======================================
4The padata parallel execution mechanism
5=======================================
6
7:Date: May 2020
8
9Padata is a mechanism by which the kernel can farm jobs out to be done in
10parallel on multiple CPUs while optionally retaining their ordering.
11
12It was originally developed for IPsec, which needs to perform encryption and
13decryption on large numbers of packets without reordering those packets.  This
14is currently the sole consumer of padata's serialized job support.
15
16Padata also supports multithreaded jobs, splitting up the job evenly while load
17balancing and coordinating between threads.
18
19Running Serialized Jobs
20=======================
21
22Initializing
23------------
24
25The first step in using padata to run serialized jobs is to set up a
26padata_instance structure for overall control of how jobs are to be run::
27
28    #include <linux/padata.h>
29
30    struct padata_instance *padata_alloc(const char *name);
31
32'name' simply identifies the instance.
33
34Then, complete padata initialization by allocating a padata_shell::
35
36   struct padata_shell *padata_alloc_shell(struct padata_instance *pinst);
37
38A padata_shell is used to submit a job to padata and allows a series of such
39jobs to be serialized independently.  A padata_instance may have one or more
40padata_shells associated with it, each allowing a separate series of jobs.
41
42Modifying cpumasks
43------------------
44
45The CPUs used to run jobs can be changed in two ways, programatically with
46padata_set_cpumask() or via sysfs.  The former is defined::
47
48    int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type,
49			   cpumask_var_t cpumask);
50
51Here cpumask_type is one of PADATA_CPU_PARALLEL or PADATA_CPU_SERIAL, where a
52parallel cpumask describes which processors will be used to execute jobs
53submitted to this instance in parallel and a serial cpumask defines which
54processors are allowed to be used as the serialization callback processor.
55cpumask specifies the new cpumask to use.
56
57There may be sysfs files for an instance's cpumasks.  For example, pcrypt's
58live in /sys/kernel/pcrypt/<instance-name>.  Within an instance's directory
59there are two files, parallel_cpumask and serial_cpumask, and either cpumask
60may be changed by echoing a bitmask into the file, for example::
61
62    echo f > /sys/kernel/pcrypt/pencrypt/parallel_cpumask
63
64Reading one of these files shows the user-supplied cpumask, which may be
65different from the 'usable' cpumask.
66
67Padata maintains two pairs of cpumasks internally, the user-supplied cpumasks
68and the 'usable' cpumasks.  (Each pair consists of a parallel and a serial
69cpumask.)  The user-supplied cpumasks default to all possible CPUs on instance
70allocation and may be changed as above.  The usable cpumasks are always a
71subset of the user-supplied cpumasks and contain only the online CPUs in the
72user-supplied masks; these are the cpumasks padata actually uses.  So it is
73legal to supply a cpumask to padata that contains offline CPUs.  Once an
74offline CPU in the user-supplied cpumask comes online, padata is going to use
75it.
76
77Changing the CPU masks are expensive operations, so it should not be done with
78great frequency.
79
80Running A Job
81-------------
82
83Actually submitting work to the padata instance requires the creation of a
84padata_priv structure, which represents one job::
85
86    struct padata_priv {
87        /* Other stuff here... */
88	void                    (*parallel)(struct padata_priv *padata);
89	void                    (*serial)(struct padata_priv *padata);
90    };
91
92This structure will almost certainly be embedded within some larger
93structure specific to the work to be done.  Most of its fields are private to
94padata, but the structure should be zeroed at initialisation time, and the
95parallel() and serial() functions should be provided.  Those functions will
96be called in the process of getting the work done as we will see
97momentarily.
98
99The submission of the job is done with::
100
101    int padata_do_parallel(struct padata_shell *ps,
102		           struct padata_priv *padata, int *cb_cpu);
103
104The ps and padata structures must be set up as described above; cb_cpu
105points to the preferred CPU to be used for the final callback when the job is
106done; it must be in the current instance's CPU mask (if not the cb_cpu pointer
107is updated to point to the CPU actually chosen).  The return value from
108padata_do_parallel() is zero on success, indicating that the job is in
109progress. -EBUSY means that somebody, somewhere else is messing with the
110instance's CPU mask, while -EINVAL is a complaint about cb_cpu not being in the
111serial cpumask, no online CPUs in the parallel or serial cpumasks, or a stopped
112instance.
113
114Each job submitted to padata_do_parallel() will, in turn, be passed to
115exactly one call to the above-mentioned parallel() function, on one CPU, so
116true parallelism is achieved by submitting multiple jobs.  parallel() runs with
117software interrupts disabled and thus cannot sleep.  The parallel()
118function gets the padata_priv structure pointer as its lone parameter;
119information about the actual work to be done is probably obtained by using
120container_of() to find the enclosing structure.
121
122Note that parallel() has no return value; the padata subsystem assumes that
123parallel() will take responsibility for the job from this point.  The job
124need not be completed during this call, but, if parallel() leaves work
125outstanding, it should be prepared to be called again with a new job before
126the previous one completes.
127
128Serializing Jobs
129----------------
130
131When a job does complete, parallel() (or whatever function actually finishes
132the work) should inform padata of the fact with a call to::
133
134    void padata_do_serial(struct padata_priv *padata);
135
136At some point in the future, padata_do_serial() will trigger a call to the
137serial() function in the padata_priv structure.  That call will happen on
138the CPU requested in the initial call to padata_do_parallel(); it, too, is
139run with local software interrupts disabled.
140Note that this call may be deferred for a while since the padata code takes
141pains to ensure that jobs are completed in the order in which they were
142submitted.
143
144Destroying
145----------
146
147Cleaning up a padata instance predictably involves calling the two free
148functions that correspond to the allocation in reverse::
149
150    void padata_free_shell(struct padata_shell *ps);
151    void padata_free(struct padata_instance *pinst);
152
153It is the user's responsibility to ensure all outstanding jobs are complete
154before any of the above are called.
155
156Running Multithreaded Jobs
157==========================
158
159A multithreaded job has a main thread and zero or more helper threads, with the
160main thread participating in the job and then waiting until all helpers have
161finished.  padata splits the job into units called chunks, where a chunk is a
162piece of the job that one thread completes in one call to the thread function.
163
164A user has to do three things to run a multithreaded job.  First, describe the
165job by defining a padata_mt_job structure, which is explained in the Interface
166section.  This includes a pointer to the thread function, which padata will
167call each time it assigns a job chunk to a thread.  Then, define the thread
168function, which accepts three arguments, ``start``, ``end``, and ``arg``, where
169the first two delimit the range that the thread operates on and the last is a
170pointer to the job's shared state, if any.  Prepare the shared state, which is
171typically allocated on the main thread's stack.  Last, call
172padata_do_multithreaded(), which will return once the job is finished.
173
174Interface
175=========
176
177.. kernel-doc:: include/linux/padata.h
178.. kernel-doc:: kernel/padata.c
179