@cindex @option{-EB} command-line option, BPF
@item -EB
This option specifies that the assembler should emit big-endian eBPF.
@cindex @option{-EL} command-line option, BPF
@item -EL
This option specifies that the assembler should emit little-endian
eBPF.
@cindex @option{-mdialect} command-line options, BPF
@item -mdialect=@var{dialect}
This option specifies the assembly language dialect to recognize while
assembling. The assembler supports @option{normal} and
@option{pseudoc}.
@cindex @option{-misa-spec} command-line options, BPF
@item -misa-spec=@var{spec}
This option specifies the version of the BPF instruction set to use
when assembling. The BPF ISA versions supported are @option{v1} @option{v2}, @option{v3} and @option{v4}.
The value @option{xbpf} can be specified to recognize extra
instructions that are used by GCC for testing purposes. But beware
this is not valid BPF.
@cindex @option{-mno-relax} command-line options, BPF
@item -mno-relax
This option tells the assembler to not relax instructions.
@end table
Note that if no endianness option is specified in the command line,
the host endianness is used.
@c man end
@node BPF Special Characters
@section BPF Special Characters
@cindex line comment character, BPF
@cindex BPF line comment character
The presence of a @samp{#} or @samp{//} anywhere on a line indicates
the start of a comment that extends to the end of the line.
@cindex block comments, BPF
@cindex BPF block comments
The presence of the @samp{/*} sequence indicates the beginning of a
block (multi-line) comment, whose contents span until the next
@samp{*/} sequence. It is not possible to nest block comments.
@cindex statement separator, BPF
Statements and assembly directives are separated by newlines and
@samp{;} characters.
@node BPF Registers
@section BPF Registers
@cindex BPF register names
@cindex register names, BPF
The eBPF processor provides ten general-purpose 64-bit registers,
which are read-write, and a read-only frame pointer register:
All BPF registers are 64-bit long. However, in the Pseudo-C syntax
registers can be referred using different names, which actually
reflect the kind of instruction they appear on:
@noindent
In pseudoc syntax:
@table @samp
@item r0..r9
General-purpose register in an instruction that operates on its value
as if it was a 64-bit value.
@item w0..w9
General-purpose register in an instruction that operates on its value
as if it was a 32-bit value.
@item r10
Read-only frame pointer register.
@end table
@noindent
Note that in the Pseudo-C syntax register names are not preceded by
@code{%} characters. A consequence of that is that in contexts like
instruction operands, where both register names and expressions
involving symbols are expected, there is no way to disambiguate
between them. In order to keep things simple, this assembler does not
allow to refer to symbols whose names collide with register names in
instruction operands.
@node BPF Directives
@section BPF Directives
@cindex machine directives, BPF
The BPF version of @code{@value{AS}} supports the following additional
machine directives:
@table @code
@cindex @code{half} directive, BPF
@item .word
The @code{.half} directive produces a 16 bit value.
@cindex @code{word} directive, BPF
@item .word
The @code{.word} directive produces a 32 bit value.
@cindex @code{dword} directive, BPF
@item .dword
The @code{.dword} directive produces a 64 bit value.
@end table
@node BPF Instructions
@section BPF Instructions
@cindex BPF opcodes
@cindex opcodes for BPF
In the instruction descriptions below the following field descriptors
are used:
@table @code
@item rd
Destination general-purpose register whose role is to be the
destination of an operation.
@item rs
Source general-purpose register whose role is to be the source of an
operation.
@item disp16
16-bit signed PC-relative offset, measured in number of 64-bit words,
minus one.
@item disp32
32-bit signed PC-relative offset, measured in number of 64-bit words,
minus one.
@item offset16
Signed 16-bit immediate representing an offset in bytes.
@item disp16
Signed 16-bit immediate representing a displacement to a target,
measured in number of 64-bit words @emph{minus one}.
@item disp32
Signed 32-bit immediate representing a displacement to a target,
measured in number of 64-bit words @emph{minus one}.
@item imm32
Signed 32-bit immediate.
@item imm64
Signed 64-bit immediate.
@end table
@noindent
Note that the assembler allows to express the value for an immediate
using any numerical literal whose two's complement encoding fits in
the immediate field. For example, @code{-2}, @code{0xfffffffe} and
@code{4294967294} all denote the same encoded 32-bit immediate, whose
value may be then interpreted by different instructions as either as a
negative or a positive number.
@subsection Arithmetic instructions
The destination register in these instructions act like an
accumulator.
Note that in pseudoc syntax these instructions should use @code{r}
registers.
@item mov rd, rs
@itemx mov rd, imm32
@itemx rd = rs
@itemx rd = imm32
Move the 64-bit value of @code{rs} in @code{rd}, or load @code{imm32}
in @code{rd}.
@item movs rd, rs, 8
@itemx rd = (s8) rs
Move the sign-extended 8-bit value in @code{rs} to @code{rd}.
@item movs rd, rs, 16
@itemx rd = (s16) rs
Move the sign-extended 16-bit value in @code{rs} to @code{rd}.
@item movs rd, rs, 32
@itemx rd = (s32) rs
Move the sign-extended 32-bit value in @code{rs} to @code{rd}.
@end table
@subsection 32-bit arithmetic instructions
The destination register in these instructions act as an accumulator.
Note that in pseudoc syntax these instructions should use @code{w}
registers. It is not allowed to mix @code{w} and @code{r} registers
in the same instruction.
@item mov32 rd, rs
@itemx mov32 rd, imm32
@itemx rd = rs
@itemx rd = imm32
Move the 32-bit value of @code{rs} in @code{rd}, or load @code{imm32}
in @code{rd}.
@item mov32s rd, rs, 8
@itemx rd = (s8) rs
Move the sign-extended 8-bit value in @code{rs} to @code{rd}.
@item mov32s rd, rs, 16
@itemx rd = (s16) rs
Move the sign-extended 16-bit value in @code{rs} to @code{rd}.
@item mov32s rd, rs, 32
@itemx rd = (s32) rs
Move the sign-extended 32-bit value in @code{rs} to @code{rd}.
@end table
@subsection Endianness conversion instructions
@table @code
@item endle rd, 16
@itemx endle rd, 32
@itemx endle rd, 64
@itemx rd = le16 rd
@itemx rd = le32 rd
@itemx rd = le64 rd
Convert the 16-bit, 32-bit or 64-bit value in @code{rd} to
little-endian and store it back in @code{rd}.
@item endbe %d, 16
@itemx endbe %d, 32
@itemx endbe %d, 64
@itemx rd = be16 rd
@itemx rd = be32 rd
@itemx rd = be64 rd
Convert the 16-bit, 32-bit or 64-bit value in @code{rd} to big-endian
and store it back in @code{rd}.
@end table
@subsection Byte swap instructions
@table @code
@item bswap rd, 16
@itemx rd = bswap16 rd
Swap the least-significant 16-bit word in @code{rd} with the
most-significant 16-bit word.
@item bswap rd, 32
@itemx rd = bswap32 rd
Swap the least-significant 32-bit word in @code{rd} with the
most-significant 32-bit word.
@item bswap rd, 64
@itemx rd = bswap64 rd
Swap the least-significant 64-bit word in @code{rd} with the
most-significant 64-bit word.
@end table
@subsection 64-bit load and pseudo maps
@table @code
@item lddw rd, imm64
@itemx rd = imm64 ll
Load the given signed 64-bit immediate to the destination register
@code{rd}.
@end table
@subsection Load instructions for socket filters
The following instructions are intended to be used in socket filters,
and are therefore not general-purpose: they make assumptions on the
contents of several registers. See the file
@file{Documentation/networking/filter.txt} in the Linux kernel source
tree for more information.
@table @code
@item stdw [rd + offset16], imm32
@itemx *(u64 *) (rd + offset16) = imm32
Store immediate as 64-bit.
@item stw [rd + offset16], imm32
@itemx *(u32 *) (rd + offset16) = imm32
Store immediate as 32-bit.
@item sth [rd + offset16], imm32
@itemx *(u16 *) (rd + offset16) = imm32
Store immediate as 16-bit.
@item stb [rd + offset16], imm32
@itemx *(u8 *) (rd + offset16) = imm32
Store immediate as 8-bit.
@end table
@subsection Jump instructions
eBPF provides the following compare-and-jump instructions, which
compare the values of the two given registers, or the values of a
register and an immediate, and perform a branch in case the comparison
holds true.
@table @code
@item ja disp16
@itemx goto disp16
Jump-always.
@item jal disp32
@itemx gotol disp32
Jump-always, long range.
@item jeq rd, rs, disp16
@itemx jeq rd, imm32, disp16
@itemx if rd == rs goto disp16
@itemx if rd == imm32 goto disp16
Jump if equal, unsigned.
@item jgt rd, rs, disp16
@itemx jgt rd, imm32, disp16
@itemx if rd > rs goto disp16
@itemx if rd > imm32 goto disp16
Jump if greater, unsigned.
@item jge rd, rs, disp16
@itemx jge rd, imm32, disp16
@itemx if rd >= rs goto disp16
@itemx if rd >= imm32 goto disp16
Jump if greater or equal.
@item jlt rd, rs, disp16
@itemx jlt rd, imm32, disp16
@itemx if rd < rs goto disp16
@itemx if rd < imm32 goto disp16
Jump if lesser.
@item jle rd , rs, disp16
@itemx jle rd, imm32, disp16
@itemx if rd <= rs goto disp16
@itemx if rd <= imm32 goto disp16
Jump if lesser or equal.
@item jset rd, rs, disp16
@itemx jset rd, imm32, disp16
@itemx if rd & rs goto disp16
@itemx if rd & imm32 goto disp16
Jump if signed equal.
@item jne rd, rs, disp16
@itemx jne rd, imm32, disp16
@itemx if rd != rs goto disp16
@itemx if rd != imm32 goto disp16
Jump if not equal.
@item jsgt rd, rs, disp16
@itemx jsgt rd, imm32, disp16
@itemx if rd s> rs goto disp16
@itemx if rd s> imm32 goto disp16
Jump if signed greater.
@item jsge rd, rs, disp16
@itemx jsge rd, imm32, disp16
@itemx if rd s>= rd goto disp16
@itemx if rd s>= imm32 goto disp16
Jump if signed greater or equal.
@item jslt rd, rs, disp16
@itemx jslt rd, imm32, disp16
@itemx if rd s< rs goto disp16
@itemx if rd s< imm32 goto disp16
Jump if signed lesser.
@item jsle rd, rs, disp16
@itemx jsle rd, imm32, disp16
@itemx if rd s<= rs goto disp16
@itemx if rd s<= imm32 goto disp16
Jump if signed lesser or equal.
@end table
A call instruction is provided in order to perform calls to other eBPF
functions, or to external kernel helpers:
@table @code
@item call disp32
@item call imm32
Jump and link to the offset @emph{disp32}, or to the kernel helper
function identified by @emph{imm32}.
@end table
Finally:
@table @code
@item exit
Terminate the eBPF program.
@end table
@subsection 32-bit jump instructions
eBPF provides the following compare-and-jump instructions, which
compare the 32-bit values of the two given registers, or the values of
a register and an immediate, and perform a branch in case the
comparison holds true.
These instructions are only available in BPF v3 or later.
@table @code
@item jeq32 rd, rs, disp16
@itemx jeq32 rd, imm32, disp16
@itemx if rd == rs goto disp16
@itemx if rd == imm32 goto disp16
Jump if equal, unsigned.
@item jgt32 rd, rs, disp16
@itemx jgt32 rd, imm32, disp16
@itemx if rd > rs goto disp16
@itemx if rd > imm32 goto disp16
Jump if greater, unsigned.
@item jge32 rd, rs, disp16
@itemx jge32 rd, imm32, disp16
@itemx if rd >= rs goto disp16
@itemx if rd >= imm32 goto disp16
Jump if greater or equal.
@item jlt32 rd, rs, disp16
@itemx jlt32 rd, imm32, disp16
@itemx if rd < rs goto disp16
@itemx if rd < imm32 goto disp16
Jump if lesser.
@item jle32 rd , rs, disp16
@itemx jle32 rd, imm32, disp16
@itemx if rd <= rs goto disp16
@itemx if rd <= imm32 goto disp16
Jump if lesser or equal.
@item jset32 rd, rs, disp16
@itemx jset32 rd, imm32, disp16
@itemx if rd & rs goto disp16
@itemx if rd & imm32 goto disp16
Jump if signed equal.
@item jne32 rd, rs, disp16
@itemx jne32 rd, imm32, disp16
@itemx if rd != rs goto disp16
@itemx if rd != imm32 goto disp16
Jump if not equal.
@item jsgt32 rd, rs, disp16
@itemx jsgt32 rd, imm32, disp16
@itemx if rd s> rs goto disp16
@itemx if rd s> imm32 goto disp16
Jump if signed greater.
@item jsge32 rd, rs, disp16
@itemx jsge32 rd, imm32, disp16
@itemx if rd s>= rd goto disp16
@itemx if rd s>= imm32 goto disp16
Jump if signed greater or equal.
@item jslt32 rd, rs, disp16
@itemx jslt32 rd, imm32, disp16
@itemx if rd s< rs goto disp16
@itemx if rd s< imm32 goto disp16
Jump if signed lesser.
@item jsle32 rd, rs, disp16
@itemx jsle32 rd, imm32, disp16
@itemx if rd s<= rs goto disp16
@itemx if rd s<= imm32 goto disp16
Jump if signed lesser or equal.
@end table
@subsection Atomic instructions
Atomic exchange instructions are provided in two flavors: one for
compare-and-swap, one for unconditional exchange.
@table @code
@item acmp [rd + offset16], rs
@itemx r0 = cmpxchg_64 (rd + offset16, r0, rs)
Atomic compare-and-swap. Compares value in @code{r0} to value
addressed by @code{rd + offset16}. On match, the value addressed by
@code{rd + offset16} is replaced with the value in @code{rs}.
Regardless, the value that was at @code{rd + offset16} is
zero-extended and loaded into @code{r0}.
@item axchg [rd + offset16], rs
@itemx rs = xchg_64 (rd + offset16, rs)
Atomic exchange. Atomically exchanges the value in @code{rs} with
the value addressed by @code{rd + offset16}.
@end table
@noindent
The following instructions provide atomic arithmetic operations.
The above instructions were introduced in the V3 of the BPF
instruction set. The following instruction is supported for backwards
compatibility:
@table @code
@item xadddw [rd + offset16], rs
Alias to @code{aadd}.
@end table
@subsection 32-bit atomic instructions
32-bit atomic exchange instructions are provided in two flavors: one
for compare-and-swap, one for unconditional exchange.
@table @code
@item acmp32 [rd + offset16], rs
@itemx w0 = cmpxchg32_32 (rd + offset16, w0, ws)
Atomic compare-and-swap. Compares value in @code{w0} to value
addressed by @code{rd + offset16}. On match, the value addressed by
@code{rd + offset16} is replaced with the value in @code{ws}.
Regardless, the value that was at @code{rd + offset16} is
zero-extended and loaded into @code{w0}.
@item axchg [rd + offset16], rs
@itemx ws = xchg32_32 (rd + offset16, ws)
Atomic exchange. Atomically exchanges the value in @code{ws} with
the value addressed by @code{rd + offset16}.
@end table
@noindent
The following instructions provide 32-bit atomic arithmetic operations.
