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