Re: rseq(2) man page

From: Alejandro Colomar
Date: Fri Jan 06 2023 - 12:51:24 EST

Hi Mathieu,

See some comments below.

Cheers,

Alex

.\" Copyright 2015-2020 Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx>
.\"
.\" %%%LICENSE_START(VERBATIM)
.\" Permission is granted to make and distribute verbatim copies of this
.\" manual provided the copyright notice and this permission notice are
.\" preserved on all copies.
.\"
.\" Permission is granted to copy and distribute modified versions of this
.\" manual under the conditions for verbatim copying, provided that the
.\" entire resulting derived work is distributed under the terms of a
.\" permission notice identical to this one.
.\"
.\" Since the Linux kernel and libraries are constantly changing, this
.\" manual page may be incorrect or out-of-date.  The author(s) assume no
.\" responsibility for errors or omissions, or for damages resulting from
.\" the use of the information contained herein.  The author(s) may not
.\" have taken the same level of care in the production of this manual,
.\" which is licensed free of charge, as they might when working
.\" professionally.
.\"
.\" Formatted or processed versions of this manual, if unaccompanied by
.\" the source, must acknowledge the copyright and authors of this work.
.\" %%%LICENSE_END

We now use SPDX-License-Identifier.

.\"
.TH RSEQ 2 2020-06-05 "Linux" "Linux Programmer's Manual"

We use lowercase for the function name (or to be more precise, the same case that the function name uses.

The date is specified with a placeholder that is replaced at the time of creation of the tarball.

The 5th argument is unspecified.

The 4th argument is now the project name and a placeholder for the version.

See an example:

$ cat man2/membarrier.2 | grep '^.TH'
.TH membarrier 2 (date) "Linux man-pages (unreleased)"


.SH NAME
rseq \- Restartable sequences system call

We use lowercase here, so s/Restartable/restartable/

.SH SYNOPSIS
.nf
.B #include <linux/rseq.h>

Is there a glibc wrapper for this syscall? Do you expect that it will be added relatively soon? Or is it expected that this syscall will be called through syscall(2) for many years?

If so, it may be better to document it directly as such, like for example membarrier:

SYNOPSIS
#include <linux/membarrier.h> /* Definition of MEMBARRIER_* constants */
#include <sys/syscall.h> /* Definition of SYS_* constants */
#include <unistd.h>

int syscall(SYS_membarrier, int cmd, unsigned int flags, int cpu_id);

Note: glibc provides no wrapper for membarrier(), necessitating the use
of syscall(2).

.sp

s/sp/PP/

.BI "int rseq(struct rseq * " rseq ", uint32_t " rseq_len ", int " flags ",

Is it valid to pass NULL in 'rseq'? If it is, we now document that using _Nullable. See for example recv(2):

SYNOPSIS
#include <sys/socket.h>

ssize_t recv(int sockfd, void buf[.len], size_t len,
int flags);
ssize_t recvfrom(int sockfd, void buf[restrict .len], size_t len,
int flags,
struct sockaddr *_Nullable restrict src_addr,
socklen_t *_Nullable restrict addrlen);
ssize_t recvmsg(int sockfd, struct msghdr *msg, int flags);




uint32_t " sig ");
.sp

.sp is unnecessary here.

.fi is missing (it's the "closing" pair to .nf).

.SH DESCRIPTION


Use .PP instead of blank lines.

The
.BR rseq ()
ABI accelerates specific user-space operations by registering a
per-thread data structure shared between kernel and user-space. This

Use semantic newlines. See man-pages(7):

Use semantic newlines
In the source of a manual page, new sentences should be started on new
lines, long sentences should be split into lines at clause breaks (com‐
mas, semicolons, colons, and so on), and long clauses should be split
at phrase boundaries. This convention, sometimes known as "semantic
newlines", makes it easier to see the effect of patches, which often
operate at the level of individual sentences, clauses, or phrases.


data structure can be read from or written to by user-space to skip
otherwise expensive system calls.

A restartable sequence is a sequence of instructions guaranteed to be executed
atomically with respect to other threads and signal handlers on the current
CPU. If its execution does not complete atomically, the kernel changes the
execution flow by jumping to an abort handler defined by user-space for that
restartable sequence.

Using restartable sequences requires to register a
rseq ABI per-thread data structure (struct rseq) through the
.BR rseq ()
system call. Only one rseq ABI can be registered per thread, so
user-space libraries and applications must follow a user-space ABI
defining how to share this resource. The ABI defining how to share this
resource between applications and libraries is defined by the C library.
Allocation of the per-thread rseq ABI and its registration to the kernel
is handled by glibc since version 2.35.

The rseq ABI per-thread data structure contains a
.I rseq_cs
field which points to the currently executing critical section. For each
thread, a single rseq critical section can run at any given point. Each
critical section need to be implemented in assembly.

The
.BR rseq ()
ABI accelerates user-space operations on per-cpu data by defining a
shared data structure ABI between each user-space thread and the kernel.

It allows user-space to perform update operations on per-cpu data
without requiring heavy-weight atomic operations.

The term CPU used in this documentation refers to a hardware execution
context. For instance, each CPU number returned by
.BR sched_getcpu ()
is a CPU. The current CPU means to the CPU on which the registered thread is
running.

Restartable sequences are atomic with respect to preemption (making it
atomic with respect to other threads running on the same CPU), as well
as signal delivery (user-space execution contexts nested over the same
thread). They either complete atomically with respect to preemption on
the current CPU and signal delivery, or they are aborted.

Restartable sequences are suited for update operations on per-cpu data.

Restartable sequences can be used on data structures shared between threads
within a process, and on data structures shared between threads across
different processes.

.PP
Some examples of operations that can be accelerated or improved
by this ABI:
.IP \[bu] 2

Use 3 instead of 2. See man-pages:

Lists
There are different kinds of lists:

[...]

Bullet lists
Elements are preceded by bullet symbols (\(bu). Anything that
doesn’t fit elsewhere is usually covered by this type of list.

[...]

There should always be exactly 2 spaces between the list symbol and the
elements. This doesn’t apply to "tagged paragraphs", which use the de‐
fault indentation rules.


The rationale for that was that if you use 1 space, then the list introducer can be confused with the list contents. Two spaces makes the difference more clear.


Also, we use \(bu instead of \[bu]. I'm not particularly worried by using it, but I prefer being consistent at which one we use.


Memory allocator per-cpu free-lists,
.IP \[bu] 2
Querying the current CPU number,
.IP \[bu] 2
Incrementing per-CPU counters,
.IP \[bu] 2
Modifying data protected by per-CPU spinlocks,
.IP \[bu] 2
Inserting/removing elements in per-CPU linked-lists,
.IP \[bu] 2
Writing/reading per-CPU ring buffers content.
.IP \[bu] 2
Accurately reading performance monitoring unit counters
with respect to thread migration.

.PP
Restartable sequences must not perform system calls. Doing so may result
in termination of the process by a segmentation fault.

.PP
The
.I rseq
argument is a pointer to the thread-local rseq structure to be shared
between kernel and user-space.

.PP
The structure
.B struct rseq
is an extensible structure. Additional feature fields can be added in
future kernel versions. Its layout is as follows:
.TP
.B Structure alignment
This structure is aligned on either 32-byte boundary, or on the
alignment value returned by
.I getauxval(AT_RSEQ_ALIGN)
if the structure size differs from 32 bytes.
.TP
.B Structure size
This structure size needs to be at least 32 bytes. It can be either
32 bytes, or it needs to be large enough to hold the result of
.I getauxval(AT_RSEQ_FEATURE_SIZE) .
Its size is passed as parameter to the rseq system call.
.PP
.in +8n
.EX
struct rseq {
    __u32 cpu_id_start;
    __u32 cpu_id;
    union {
        /* Edited out for conciseness. [...] */
    } rseq_cs;
    __u32 flags;
    __u32 node_id;
    __u32 mm_cid;
} __attribute__((aligned(32)));
.EE
.TP
.B Fields

.TP
.in +4n

I guess you're looking for .RS/.RE. It would wrap all the indented stuff, replacing .in.

.I cpu_id_start
Always-updated value of the CPU number on which the registered thread is
running. Its value is guaranteed to always be a possible CPU number,
even when rseq is not registered. Its value should always be confirmed by

rseq (and maybe other cases around too) should be formatted in italics, since it's a variable name (.I).

reading the cpu_id field before user-space performs any side-effect (e.g.
storing to memory).

This field is always guaranteed to hold a valid CPU number in the range
[ 0 ..  nr_possible_cpus - 1 ]. It can therefore be loaded by user-space
and used as an offset in per-cpu data structures without having to check
whether its value is within the valid bounds compared to the number of
possible CPUs in the system.

Initialized by user-space to a possible CPU number (e.g., 0), updated
by the kernel for threads registered with rseq.

For user-space applications executed on a kernel without rseq support,
the cpu_id_start field stays initialized at 0, which is indeed a valid
CPU number. It is therefore valid to use it as an offset in per-cpu data
structures, and only validate whether it's actually the current CPU
number by comparing it with the cpu_id field within the rseq critical
section. If the kernel does not provide rseq support, that cpu_id field
stays initialized at -1, so the comparison always fails, as intended.

This field should only be read by the thread which registered this data
structure. Aligned on 32-bit.

It is up to user-space to implement a fall-back mechanism for scenarios where
rseq is not available.
.in
.TP
.in +4n
.I cpu_id
Always-updated value of the CPU number on which the registered thread is
running. Initialized by user-space to -1, updated by the kernel for
threads registered with rseq.

This field should only be read by the thread which registered this data
structure. Aligned on 32-bit.
.in
.TP
.in +4n
.I rseq_cs
The rseq_cs field is a pointer to a struct rseq_cs. Is is NULL when no
rseq assembly block critical section is active for the registered thread.
Setting it to point to a critical section descriptor (struct rseq_cs)
marks the beginning of the critical section.

Initialized by user-space to NULL.

Updated by user-space, which sets the address of the currently
active rseq_cs at the beginning of assembly instruction sequence
block, and set to NULL by the kernel when it restarts an assembly
instruction sequence block, as well as when the kernel detects that
it is preempting or delivering a signal outside of the range
targeted by the rseq_cs. Also needs to be set to NULL by user-space
before reclaiming memory that contains the targeted struct rseq_cs.

Read and set by the kernel.

This field should only be updated by the thread which registered this
data structure. Aligned on 64-bit.
.in
.TP
.in +4n
.I flags
Flags indicating the restart behavior for the registered thread. This is
mainly used for debugging purposes. Can be a combination of:
.IP \[bu]
RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT: Inhibit instruction sequence block restart
on preemption for this thread. This flag is deprecated since kernel 6.1.
.IP \[bu]
RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL: Inhibit instruction sequence block restart
on signal delivery for this thread. This flag is deprecated since kernel 6.1.
.IP \[bu]
RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE: Inhibit instruction sequence block restart
on migration for this thread. This flag is deprecated since kernel 6.1.

Initialized by user-space, used by the kernel.
.in
.TP
.in +4n
.I node_id
Always-updated value of the current NUMA node ID.

Initialized by user-space to 0.

Updated by the kernel. Read by user-space with single-copy atomicity
semantics. This field should only be read by the thread which registered
this data structure. Aligned on 32-bit.
.in
.TP
.in +4n
.I mm_cid
Contains the current thread's concurrency ID (allocated uniquely within
a memory map).

Updated by the kernel. Read by user-space with single-copy atomicity
semantics. This field should only be read by the thread which registered
this data structure. Aligned on 32-bit.

This concurrency ID is within the possible cpus range, and is
temporarily (and uniquely) assigned while threads are actively running
within a memory map. If a memory map has fewer threads than cores, or is
limited to run on few cores concurrently through sched affinity or
cgroup cpusets, the concurrency IDs will be values close to 0, thus
allowing efficient use of user-space memory for per-cpu data structures.

.PP
The layout of
.B struct rseq_cs
version 0 is as follows:
.TP
.B Structure alignment
This structure is aligned on 32-byte boundary.
.TP
.B Structure size
This structure has a fixed size of 32 bytes.
.PP
.in +8n
.EX
struct rseq_cs {
    __u32   version;
    __u32   flags;
    __u64   start_ip;
    __u64   post_commit_offset;
    __u64   abort_ip;
} __attribute__((aligned(32)));
.EE
.TP
.B Fields

.TP
.in +4n
.I version
Version of this structure. Should be initialized to 0.
.in
.TP
.in +4n
.I flags
Flags indicating the restart behavior of this structure. Can be a combination
of:
.IP \[bu]

This list should be a tagged paragraph instead. See man-pages(7):

Lists
There are different kinds of lists:

Tagged paragraphs
These are used for a list of tags and their descriptions. When
the tags are constants (either macros or numbers) they are in
bold. Use the .TP macro.

An example is this "Tagged paragraphs" subsection is itself.

[...]

Bullet lists
Elements are preceded by bullet symbols (\(bu). Anything that
doesn’t fit elsewhere is usually covered by this type of list.

[...]


RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT: Inhibit instruction sequence block restart
on preemption for this critical section. This flag is deprecated since kernel
6.1.
.IP \[bu]
RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL: Inhibit instruction sequence block restart
on signal delivery for this critical section. This flag is deprecated since
kernel 6.1.
.IP \[bu]
RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE: Inhibit instruction sequence block restart
on migration for this critical section. This flag is deprecated since kernel
6.1.
.TP
.in +4n
.I start_ip
Instruction pointer address of the first instruction of the sequence of
consecutive assembly instructions.
.in
.TP
.in +4n
.I post_commit_offset
Offset (from start_ip address) of the address after the last instruction
of the sequence of consecutive assembly instructions.
.in
.TP
.in +4n
.I abort_ip
Instruction pointer address where to move the execution flow in case of
abort of the sequence of consecutive assembly instructions.
.in

.PP
The
.I rseq_len
argument is the size of the
.I struct rseq
to register.

.PP
The
.I flags
argument is 0 for registration, and
.IR RSEQ_FLAG_UNREGISTER
for unregistration.

.PP
The
.I sig
argument is the 32-bit signature to be expected before the abort
handler code.

.PP
A single library per process should keep the rseq structure in a
per-thread data structure.
The
.I cpu_id
field should be initialized to -1, and the
.I cpu_id_start
field should be initialized to a possible CPU value (typically 0).

.PP
Each thread is responsible for registering and unregistering its rseq
structure. No more than one rseq structure address can be registered
per thread at a given time.

.PP
Reclaim of rseq object's memory must only be done after either an
explicit rseq unregistration is performed or after the thread exits.

.PP
In a typical usage scenario, the thread registering the rseq
structure will be performing loads and stores from/to that structure. It
is however also allowed to read that structure from other threads.
The rseq field updates performed by the kernel provide relaxed atomicity
semantics (atomic store, without memory ordering), which guarantee that other
threads performing relaxed atomic reads (atomic load, without memory ordering)
of the cpu number fields will always observe a consistent value.

.SH RETURN VALUE
A return value of 0 indicates success. On error, \-1 is returned, and
.I errno
is set appropriately.

.SH ERRORS
.TP
.B EINVAL
Either
.I flags
contains an invalid value, or
.I rseq
contains an address which is not appropriately aligned, or
.I rseq_len
contains an incorrect size.
.TP
.B ENOSYS
The
.BR rseq ()
system call is not implemented by this kernel.
.TP
.B EFAULT
.I rseq
is an invalid address.

Doesn't this result in a SEGV? It's trying to access invalid memory. We had some discussion about this in other syscalls, and concluded that that's undefined behavior, and a crash is valid behavior (and probably a good thing to do), right? I'm just curious about the view from the kernel point of view.

.TP
.B EBUSY
Restartable sequence is already registered for this thread.
.TP
.B EPERM
The
.I sig
argument on unregistration does not match the signature received
on registration.

.SH VERSIONS
The
.BR rseq ()
system call was added in Linux 4.18.

.SH CONFORMING TO

We call that section STANDARDS now.

.BR rseq ()
is Linux-specific.

.in
.SH SEE ALSO
.BR sched_getcpu (3) ,
.BR membarrier (2) ,
.BR getauxval (3)


Cheers,

Alex



--
<http://www.alejandro-colomar.es/>

Attachment: OpenPGP_signature
Description: OpenPGP digital signature