Hybrid FPRs is a scheme where scalar FP registers are 64b wide, but
accesses to odd indexed single registers use bits 63:32 of the
preceeding even indexed 64b register. In this mode all FP code
except that built for the plain FP64 ABI can execute correctly. Most
notably a combination of FP64A & FP32 code can execute correctly,
allowing for existing FP32 binaries to be linked with new FP64A binaries
that can make use of 64 bit FP & MSA.
Hybrid FPRs are implemented by setting both the FR & FRE bits, trapping
& emulating single precision FP instructions (via Reserved Instruction
exceptions) whilst allowing others to execute natively. It therefore has
a penalty in terms of execution speed, and should only be used when no
fully native mode can be. As more binaries are recompiled to use either
the FPXX or FP64(A) ABIs, the need for hybrid FPRs should diminish.
However in the short to mid term it allows for a gradual transition
towards that world, rather than a complete ABI break which is not
feasible for some users & not desirable for many.
A task will be executed using the hybrid FPR scheme when its
TIF_HYBRID_FPREGS flag is set & TIF_32BIT_FPREGS is clear. A further
patch will set the flags as necessary, this patch simply adds the
infrastructure necessary for the hybrid FPR mode to work.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/7683/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Based on the spatch
@@
expression e;
@@
- return (e);
+ return e;
with heavy hand editing because some of the changes are either whitespace
or identation only or result in excessivly long lines.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Starting with version 2.24.51.20140728 MIPS binutils complain loudly
about mixing soft-float and hard-float object files, leading to this
build failure since GCC is invoked with "-msoft-float" on MIPS:
{standard input}: Warning: .gnu_attribute 4,3 requires `softfloat'
LD arch/mips/alchemy/common/built-in.o
mipsel-softfloat-linux-gnu-ld: Warning: arch/mips/alchemy/common/built-in.o
uses -msoft-float (set by arch/mips/alchemy/common/prom.o),
arch/mips/alchemy/common/sleeper.o uses -mhard-float
To fix this, we detect if GAS is new enough to support "-msoft-float" command
option, and if it does, we can let GCC pass it to GAS; but then we also need
to sprinkle the files which make use of floating point registers with the
necessary ".set hardfloat" directives.
Signed-off-by: Manuel Lauss <manuel.lauss@gmail.com>
Cc: Linux-MIPS <linux-mips@linux-mips.org>
Cc: Matthew Fortune <Matthew.Fortune@imgtec.com>
Cc: Markos Chandras <Markos.Chandras@imgtec.com>
Cc: Maciej W. Rozycki <macro@linux-mips.org>
Patchwork: https://patchwork.linux-mips.org/patch/8355/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Commit 08a07904e1 ("MIPS: math-emu: Remove most ifdefery") removed
the #ifdef ISA conditions and switched to runtime detection. However,
according to the instruction set manual, the cop1x_op instructions are
available in >=MIPS32r2 as well. This fixes a problem on MIPS32r2
with the ntpd package which failed to execute with a SIGILL exit code due
to the fact that a madd.d instruction was not being emulated.
Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
Fixes: 08a07904e1 ("MIPS: math-emu: Remove most ifdefery")
Cc: <stable@vger.kernel.org> # v3.16+
Cc: linux-mips@linux-mips.org
Reviewed-by: Paul Burton <paul.burton@imgtec.com>
Reviewed-by: James Hogan <james.hogan@imgtec.com>
Cc: Markos Chandras <markos.chandras@imgtec.com>
Patchwork: https://patchwork.linux-mips.org/patch/8173/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Commit bbd426f542 "MIPS: Simplify FP context access" modified the
SIFROMREG & SIFROMHREG macros such that they return unsigned rather
than signed 32b integers. I had believed that to be fine, but
inadvertently missed the MFC1 & MFHC1 cases which write to a struct
pt_regs regs element. On MIPS32 this is fine, but on 64 bit those
saved regs' fields are 64 bit wide. Using unsigned values caused the
32 bit value from the FP register to be zero rather than sign extended
as the architecture specifies, causing incorrect emulation of the
MFC1 & MFHc1 instructions. Fix by reintroducing the casts to signed
integers, and therefore the sign extension.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: stable@vger.kernel.org # v3.15+
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/7848/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The fix in the preceeding commit did do exactly the same thing in two
places showing some code cleanup was due.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Older GCC doesn't get named initializations of anonymous structs right,
that is members are not initializable in the containing structure through
name however old style initializations are working fine.
The issue exists with gcc up to 4.5.x.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Some versions of the assembler will not assemble CFC1 for OCTEON, so
override the ISA for these.
Add r4k_fpu.o to handle low level FPU initialization.
Modify octeon_switch.S to save the FPU registers. And include
r4k_switch.S to pick up more FPU support.
Get rid of "#define cpu_has_fpu 0"
Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: Andreas Herrmann <andreas.herrmann@caviumnetworks.com>
Cc: linux-mips@linux-mips.org
Cc: James Hogan <james.hogan@imgtec.com>
Cc: kvm@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/7006/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Sometimes it's useful to let the user, while doing performance research,
know what in the IEEE754 exceptions has caused many times of FP emulation
when running a specific application. This patch adds 5 more files to
/sys/kernel/debug/mips/fpuemustats/, whose filenames begin with "ieee754".
These stats are in addition to the existing cp1ops, cp1xops, errors, loads
and stores, which may not be useful in understanding the reasons of ieee754
exceptions.
[ralf@linux-mips.org: Fixed reject due to other changes to the kernel
FP assist software.]
Signed-off-by: Deng-Cheng Zhu <dengcheng.zhu@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: Steven.Hill@imgtec.com
Cc: james.hogan@imgtec.com
Patchwork: http://patchwork.linux-mips.org/patch/7044/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Move microMIPS32_to_MIPS32() to a separate file which only gets built
for mipsMIPS configurations; for other configurations the optimizer
eleminates calls to microMIPS32_to_MIPS32().
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Previously math-emu was using the IEEE-754 constants internally. These
were differing by having the constants for rounding to +/- infinity
switched, so a conversion was necessary. This would be entirely
avoidable if the MIPS constants were used throughout, so get rid of
the bloat.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
mm_isBranchInstr() did reside in the math emu code even though it logically
is separate and also is used outside the math emu code. In addition GCC 4.9.0
leaves the following unnnecessarily bloated function body for a non-microMIPS
configuration:
<mm_isBranchInstr>:
105c: afa50004 sw a1,4(sp)
1060: afa60008 sw a2,8(sp)
1064: afa7000c sw a3,12(sp)
1068: 03e00008 jr ra
106c: 00001021 move v0,zero
which stores arguments that are never going to be used on the stack frame.
Move mm_isBranchInstr() from cp1emu.c to branch.c, then split mm_isBranchInstr()
into a __mm_isBranchInstr() core and a mm_isBranchInstr() wrapper inline function
which only invokes __mm_isBranchInstr() on microMIPS configurations.
This shaves off 112 bytes off the kernel and improves code flow a bit.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
o Only define variables in the outermost block
o One empty line at most
o Format comments as per CodingStyle
o Update FSF address in licensing term comment
o Spell FPU and MIPS in all capitals.
o Remove ####-type of lines in comments.
o Try to make things a bit most consistent between sp_*.c / dp_*.c files.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
There are two version of get_rounding(), one for single precision, one
for double precision. Add a ieee754sp_ rsp. ieee754dp_ prefix for
clarity.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
ieee754d.o contains only debug code and dp_sqrt.o and sp_sqrt.o contain
code which for MIPS I/II/III systems we don't want to link. Again the
savings can be considerable for some systems:
$ mips-linux-size --totals ieee754d.o dp_sqrt.o sp_sqrt.o
text data bss dec hex filename
1624 0 0 1624 658 ieee754d.o
2016 0 0 2016 7e0 dp_sqrt.o
736 0 0 736 2e0 sp_sqrt.o
4376 0 0 4376 1118 (TOTALS)
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Both are unused since lmo commit fdffbafbb38723618626c70ffdc6ff9175cdffa2
[Lots of FPU bug fixes from Kjeld Borch Egevang.]
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
LONG_LONG_MAX is a symbol defined in <limits.h> which may not be available
so better rely on something provided by a kernel header. While at it,
turn these function-like macros into inline functions.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Most of these tests should be runtime tests. This also finally means
that on a MIPS III systems MIPS IV opcodes are going to result in an
exception as they're supposed to.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The header file had no include guards; this only happened to work because
the file only contains macro definitions and protypes.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The upper bits of an FP register are architecturally defined as
unpredictable following an instructions which only writes the lower
bits. The prior behaviour of the kernel is to leave them unmodified.
This patch modifies that to clear the upper bits to zero. This is what
the MSA architecture reference manual specifies should happen for its
wider registers and is still permissible for scalar FP instructions
given the bits unpredictability there.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: sergei.shtylyov@cogentembedded.com
Patchwork: https://patchwork.linux-mips.org/patch/6435/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
These functions aren't directly related to the FPU emulator at all, they
simply copy between a thread's saved context & a sigcontext. Thus move
them to the appropriate signal files & rename them accordingly. This
makes it clearer that the functions don't require the FPU emulator in
any way.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Reviewed-by: Qais Yousef <qais.yousef@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/6422/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
This patch replaces the fpureg_t typedef with a "union fpureg" enabling
easier access to 32 & 64 bit values. This allows the access macros used
in cp1emu.c to be simplified somewhat. It will also make it easier to
expand the width of the FP registers as will be done in a future
patch in order to support the 128 bit registers introduced with MSA.
No behavioural change is intended by this patch.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Reviewed-by: Qais Yousef <qais.yousef@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/6532/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
When running applications which contain the instruction "prefx" on FPU-less
CPUs, a message "Illegal instruction" will be seen. This instruction is
supposed to be ignored by the FPU emulator. However, its current detection
and function field encoding are incorrect. This patch fix the issue.
Signed-off-by: Deng-Cheng Zhu <dengcheng.zhu@imgtec.com>
Reviewed-by: Leonid Yegoshin <Leonid.Yegoshin@imgtec.com>
Reviewed-by: Paul Burton <paul.burton@imgtec.com>
Acked-by: David Daney <david.daney@cavium.com>
Cc: linux-mips@linux-mips.org
Cc: Steven.Hill@imgtec.com
Patchwork: https://patchwork.linux-mips.org/patch/6608/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
