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_possible(const char *name);
32 'name' simply identifies the instance.
34 There are functions for enabling and disabling the instance::
36 int padata_start(struct padata_instance *pinst);
37 void padata_stop(struct padata_instance *pinst);
39 These functions are setting or clearing the "PADATA_INIT" flag; if that flag is
40 not set, other functions will refuse to work. padata_start() returns zero on
41 success (flag set) or -EINVAL if the padata cpumask contains no active CPU
42 (flag not set). padata_stop() clears the flag and blocks until the padata
45 Finally, complete padata initialization by allocating a padata_shell::
47 struct padata_shell *padata_alloc_shell(struct padata_instance *pinst);
49 A padata_shell is used to submit a job to padata and allows a series of such
50 jobs to be serialized independently. A padata_instance may have one or more
51 padata_shells associated with it, each allowing a separate series of jobs.
56 The CPUs used to run jobs can be changed in two ways, programatically with
57 padata_set_cpumask() or via sysfs. The former is defined::
59 int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type,
60 cpumask_var_t cpumask);
62 Here cpumask_type is one of PADATA_CPU_PARALLEL or PADATA_CPU_SERIAL, where a
63 parallel cpumask describes which processors will be used to execute jobs
64 submitted to this instance in parallel and a serial cpumask defines which
65 processors are allowed to be used as the serialization callback processor.
66 cpumask specifies the new cpumask to use.
68 There may be sysfs files for an instance's cpumasks. For example, pcrypt's
69 live in /sys/kernel/pcrypt/<instance-name>. Within an instance's directory
70 there are two files, parallel_cpumask and serial_cpumask, and either cpumask
71 may be changed by echoing a bitmask into the file, for example::
73 echo f > /sys/kernel/pcrypt/pencrypt/parallel_cpumask
75 Reading one of these files shows the user-supplied cpumask, which may be
76 different from the 'usable' cpumask.
78 Padata maintains two pairs of cpumasks internally, the user-supplied cpumasks
79 and the 'usable' cpumasks. (Each pair consists of a parallel and a serial
80 cpumask.) The user-supplied cpumasks default to all possible CPUs on instance
81 allocation and may be changed as above. The usable cpumasks are always a
82 subset of the user-supplied cpumasks and contain only the online CPUs in the
83 user-supplied masks; these are the cpumasks padata actually uses. So it is
84 legal to supply a cpumask to padata that contains offline CPUs. Once an
85 offline CPU in the user-supplied cpumask comes online, padata is going to use
88 Changing the CPU masks are expensive operations, so it should not be done with
94 Actually submitting work to the padata instance requires the creation of a
95 padata_priv structure, which represents one job::
98 /* Other stuff here... */
99 void (*parallel)(struct padata_priv *padata);
100 void (*serial)(struct padata_priv *padata);
103 This structure will almost certainly be embedded within some larger
104 structure specific to the work to be done. Most of its fields are private to
105 padata, but the structure should be zeroed at initialisation time, and the
106 parallel() and serial() functions should be provided. Those functions will
107 be called in the process of getting the work done as we will see
110 The submission of the job is done with::
112 int padata_do_parallel(struct padata_shell *ps,
113 struct padata_priv *padata, int *cb_cpu);
115 The ps and padata structures must be set up as described above; cb_cpu
116 points to the preferred CPU to be used for the final callback when the job is
117 done; it must be in the current instance's CPU mask (if not the cb_cpu pointer
118 is updated to point to the CPU actually chosen). The return value from
119 padata_do_parallel() is zero on success, indicating that the job is in
120 progress. -EBUSY means that somebody, somewhere else is messing with the
121 instance's CPU mask, while -EINVAL is a complaint about cb_cpu not being in the
122 serial cpumask, no online CPUs in the parallel or serial cpumasks, or a stopped
125 Each job submitted to padata_do_parallel() will, in turn, be passed to
126 exactly one call to the above-mentioned parallel() function, on one CPU, so
127 true parallelism is achieved by submitting multiple jobs. parallel() runs with
128 software interrupts disabled and thus cannot sleep. The parallel()
129 function gets the padata_priv structure pointer as its lone parameter;
130 information about the actual work to be done is probably obtained by using
131 container_of() to find the enclosing structure.
133 Note that parallel() has no return value; the padata subsystem assumes that
134 parallel() will take responsibility for the job from this point. The job
135 need not be completed during this call, but, if parallel() leaves work
136 outstanding, it should be prepared to be called again with a new job before
137 the previous one completes.
142 When a job does complete, parallel() (or whatever function actually finishes
143 the work) should inform padata of the fact with a call to::
145 void padata_do_serial(struct padata_priv *padata);
147 At some point in the future, padata_do_serial() will trigger a call to the
148 serial() function in the padata_priv structure. That call will happen on
149 the CPU requested in the initial call to padata_do_parallel(); it, too, is
150 run with local software interrupts disabled.
151 Note that this call may be deferred for a while since the padata code takes
152 pains to ensure that jobs are completed in the order in which they were
158 Cleaning up a padata instance predictably involves calling the three free
159 functions that correspond to the allocation in reverse::
161 void padata_free_shell(struct padata_shell *ps);
162 void padata_stop(struct padata_instance *pinst);
163 void padata_free(struct padata_instance *pinst);
165 It is the user's responsibility to ensure all outstanding jobs are complete
166 before any of the above are called.
168 Running Multithreaded Jobs
169 ==========================
171 A multithreaded job has a main thread and zero or more helper threads, with the
172 main thread participating in the job and then waiting until all helpers have
173 finished. padata splits the job into units called chunks, where a chunk is a
174 piece of the job that one thread completes in one call to the thread function.
176 A user has to do three things to run a multithreaded job. First, describe the
177 job by defining a padata_mt_job structure, which is explained in the Interface
178 section. This includes a pointer to the thread function, which padata will
179 call each time it assigns a job chunk to a thread. Then, define the thread
180 function, which accepts three arguments, ``start``, ``end``, and ``arg``, where
181 the first two delimit the range that the thread operates on and the last is a
182 pointer to the job's shared state, if any. Prepare the shared state, which is
183 typically allocated on the main thread's stack. Last, call
184 padata_do_multithreaded(), which will return once the job is finished.
189 .. kernel-doc:: include/linux/padata.h
190 .. kernel-doc:: kernel/padata.c