Commit 716850ab10 ("MIPS: eBPF: Initial eBPF support for MIPS32
architecture.") enabled our eBPF JIT for MIPS32 kernels, whereas it has
previously only been availailable for MIPS64. It was my understanding at
the time that the BPF test suite was passing & JITing a comparable
number of tests to our cBPF JIT [1], but it turns out that was not the
case.
The eBPF JIT has a number of problems on MIPS32:
- Most notably various code paths still result in emission of MIPS64
instructions which will cause reserved instruction exceptions & kernel
panics when run on MIPS32 CPUs.
- The eBPF JIT doesn't account for differences between the O32 ABI used
by MIPS32 kernels versus the N64 ABI used by MIPS64 kernels. Notably
arguments beyond the first 4 are passed on the stack in O32, and this
is entirely unhandled when JITing a BPF_CALL instruction. Stack space
must be reserved for arguments even if they all fit in registers, and
the callee is free to assume that stack space has been reserved for
its use - with the eBPF JIT this is not the case, so calling any
function can result in clobbering values on the stack & unpredictable
behaviour. Function arguments in eBPF are always 64-bit values which
is also entirely unhandled - the JIT still uses a single (32-bit)
register per argument. As a result all function arguments are always
passed incorrectly when JITing a BPF_CALL instruction, leading to
kernel crashes or strange behavior.
- The JIT attempts to bail our on use of ALU64 instructions or 64-bit
memory access instructions. The code doing this at the start of
build_one_insn() incorrectly checks whether BPF_OP() equals BPF_DW,
when it should really be checking BPF_SIZE() & only doing so when
BPF_CLASS() is one of BPF_{LD,LDX,ST,STX}. This results in false
positives that cause more bailouts than intended, and that in turns
hides some of the problems described above.
- The kernel's cBPF->eBPF translation makes heavy use of 64-bit eBPF
instructions that the MIPS32 eBPF JIT bails out on, leading to most
cBPF programs not being JITed at all.
Until these problems are resolved, revert the removal of the cBPF JIT
performed by commit 716850ab10 ("MIPS: eBPF: Initial eBPF support for
MIPS32 architecture."). Together with commit f8fffebdea ("MIPS: BPF:
Disable MIPS32 eBPF JIT") this restores MIPS32 BPF JIT behavior back to
the same state it was prior to the introduction of the broken eBPF JIT
support.
[1] https://lore.kernel.org/linux-mips/MWHPR2201MB13583388481F01A422CE7D66D4410@MWHPR2201MB1358.namprd22.prod.outlook.com/
Signed-off-by: Paul Burton <paulburton@kernel.org>
Fixes: 716850ab10 ("MIPS: eBPF: Initial eBPF support for MIPS32 architecture.")
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Hassan Naveed <hnaveed@wavecomp.com>
Cc: Tony Ambardar <itugrok@yahoo.com>
Cc: bpf@vger.kernel.org
Cc: netdev@vger.kernel.org
Cc: linux-mips@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Currently MIPS32 supports a JIT for classic BPF only, not extended BPF.
This patch adds JIT support for extended BPF on MIPS32, so code is
actually JIT'ed instead of being only interpreted. Instructions with
64-bit operands are not supported at this point.
We can delete classic BPF because the kernel will translate classic BPF
programs into extended BPF and JIT them, eliminating the need for
classic BPF.
Signed-off-by: Hassan Naveed <hnaveed@wavecomp.com>
Signed-off-by: Paul Burton <paul.burton@mips.com>
Cc: kafai@fb.com
Cc: songliubraving@fb.com
Cc: yhs@fb.com
Cc: netdev@vger.kernel.org
Cc: bpf@vger.kernel.org
Cc: linux-mips@vger.kernel.org
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: James Hogan <jhogan@kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: open list:MIPS <linux-mips@linux-mips.org>
Cc: open list <linux-kernel@vger.kernel.org>
Remove the need to check that __mips_isa_rev is defined by using the
newly added MIPS_ISA_REV.
Signed-off-by: Matt Redfearn <matt.redfearn@mips.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paul Burton <paul.burton@mips.com>
Cc: David Daney <david.daney@cavium.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/18677/
Signed-off-by: James Hogan <jhogan@kernel.org>
o Socket data is unsigned, so use unsigned accessors instructions.
o Fix path result pointer generation arithmetic.
o Fix half-word byte swapping code for unsigned semantics.
Signed-off-by: David Daney <david.daney@cavium.com>
Cc: James Hogan <james.hogan@imgtec.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Steven J. Hill <steven.hill@cavium.com>
Cc: linux-mips@linux-mips.org
Cc: netdev@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/15747/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The entire bpf_jit_asm.S is written in noreorder mode because "we know
better" according to a comment. This also prevented the assembler from
throwing in the required NOPs for MIPS I processors which have no
load-use interlock, thus the load's consumer might end up using the
old value of the register from prior to the load.
Fixed by putting the assembler in reorder mode for just the affected
load instructions. This is not enough for gas to actually try to be
clever by looking at the next instruction and inserting a nop only
when needed but as the comment said "we know better", so getting gas
to unconditionally emit a NOP is just right in this case and prevents
adding further ifdefery.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
FEXPORT also marks the symbol as code using .type symbol, @function.
Without objdump -d will output only a hexdump for code following the
affected symbols.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The rotr, seh and wsbh instructions have been introduced with the R2
ISA. Thus the current BPF code fails to build on pre-R2 little endian
CPUs:
CC arch/mips/net/bpf_jit.o
AS arch/mips/net/bpf_jit_asm.o
/home/aurel32/linux-4.2/arch/mips/net/bpf_jit_asm.S: Assembler messages:
/home/aurel32/linux-4.2/arch/mips/net/bpf_jit_asm.S:67: Error: opcode not supported on this processor: mips32 (mips32) `wsbh $8,$19'
/home/aurel32/linux-4.2/arch/mips/net/bpf_jit_asm.S:68: Error: opcode not supported on this processor: mips32 (mips32) `rotr $19,$8,16'
/home/aurel32/linux-4.2/arch/mips/net/bpf_jit_asm.S:83: Error: opcode not supported on this processor: mips32 (mips32) `wsbh $8,$19'
/home/aurel32/linux-4.2/arch/mips/net/bpf_jit_asm.S:84: Error: opcode not supported on this processor: mips32 (mips32) `seh $19,$8'
/home/aurel32/linux-4.2/arch/mips/net/bpf_jit_asm.S:151: Error: opcode not supported on this processor: mips32 (mips32) `wsbh $8,$12'
/home/aurel32/linux-4.2/arch/mips/net/bpf_jit_asm.S:153: Error: opcode not supported on this processor: mips32 (mips32) `rotr $19,$8,16'
/home/aurel32/linux-4.2/arch/mips/net/bpf_jit_asm.S:164: Error: opcode not supported on this processor: mips32 (mips32) `wsbh $19,$12'
/home/aurel32/linux-4.2/scripts/Makefile.build:294: recipe for target 'arch/mips/net/bpf_jit_asm.o' failed
Fix that by providing equivalent code for these CPUs.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Reviewed-by: Markos Chandras <markos.chandras@imgtec.com>
Cc: stable@vger.kernel.org # v4.2+
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/11098/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
On little endian, avoid generating the big endian version of the code
by using #else in addition to #ifdef #endif. Also fix one alignment
issue wrt delay slot.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Reviewed-by: Markos Chandras <markos.chandras@imgtec.com>
Cc: stable@vger.kernel.org # v4.2+
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/11097/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
This commit introduces BPF ASM helpers for MIPS and MIPS64 kernels.
The purpose of this patch is to twofold:
1) We are now able to handle negative offsets instead of either
falling back to the interpreter or to simply not do anything and
bail out.
2) Optimize reads from the packet header instead of calling the C
helpers
Because of this patch, we are now able to get rid of quite a bit of
code in the JIT generation process by using MIPS optimized assembly
code. The new assembly code makes the test_bpf testsuite happy with
all 60 test passing successfully compared to the previous
implementation where 2 tests were failing.
Doing some basic analysis in the results between the old
implementation and the new one we can obtain the following
summary running current mainline on an ER8 board (+/- 30us delta is
ignored to prevent noise from kernel scheduling or IRQ latencies):
Summary: 22 tests are faster, 7 are slower and 47 saw no improvement
with the most notable improvement being the tcpdump tests. The 7 tests
that seem to be a bit slower is because they all follow the slow path
(bpf_internal_load_pointer_neg_helper) which is meant to be slow so
that's not a problem.
Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
Cc: netdev@vger.kernel.org
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Daniel Borkmann <dborkman@redhat.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: linux-kernel@vger.kernel.org
Cc: linux-mips@linux-mips.org
Cc: netdev@vger.kernel.org
Patchwork: http://patchwork.linux-mips.org/patch/10530/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
