qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
* Re: [PATCH v2 63/93] tcg/tci: Use ffi for calls
@ 2025-12-02 18:38 Philippe Mathieu-Daudé
  0 siblings, 0 replies; 16+ messages in thread
From: Philippe Mathieu-Daudé @ 2025-12-02 18:38 UTC (permalink / raw)
  To: Stefan Weil, Jeremy Huddleston Sequoia
  Cc: QEMU Developers, Peter Maydell, Richard Henderson

On 7/2/21 22:33, Stefan Weil wrote:
> On 07.02.21 21:12, Richard Henderson wrote:
>> On 2/7/21 11:52 AM, Peter Maydell wrote:
>>> On Sun, 7 Feb 2021 at 17:41, Richard Henderson
>>> <richard.henderson@linaro.org> wrote:
>>>>
>>>> On 2/7/21 8:25 AM, Stefan Weil wrote:
>>>>>> +#include "qemu-common.h"
>>>>>> +#include "tcg/tcg.h"           /* MAX_OPC_PARAM_IARGS */
>>>>>> +#include "exec/cpu_ldst.h"
>>>>>> +#include "tcg/tcg-op.h"
>>>>>> +#include "qemu/compiler.h"
>>>>>> +#include <ffi.h>
>>>>>> +
>>>>>
>>>>>
>>>>> ffi.h is not found on macOS with Homebrew.
>>>>>
>>>>> This can be fixed by using pkg-config to find the right compiler (and maybe
>>>>> also linker) flags:
>>>>>
>>>>> % pkg-config --cflags libffi
>>>>> -I/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk/usr/include/ffi
>>>>> % pkg-config --libs libffi
>>>>> -lffi
>>>>
>>>>
>>>> Which is exactly what I do in the previous patch:
>>>>
>>>>
>>>>> +++ b/meson.build
>>>>> @@ -1901,7 +1901,14 @@ specific_ss.add(when: 'CONFIG_TCG', if_true: files(
>>>>>     'tcg/tcg-op.c',
>>>>>     'tcg/tcg.c',
>>>>>   ))
>>>>> -specific_ss.add(when: 'CONFIG_TCG_INTERPRETER', if_true: files('tcg/tci.c'))
>>>>> +
>>>>> +if get_option('tcg_interpreter')
>>>>> +  libffi = dependency('libffi', version: '>=3.0',
>>>>> +                      static: enable_static, method: 'pkg-config',
>>>>> +                      required: true)
>>>>> +  specific_ss.add(libffi)
>>>>> +  specific_ss.add(files('tcg/tci.c'))
>>>>> +endif
>>>>
>>>> Did you need a PKG_CONFIG_LIBDIR set for homebrew?
>>>
>>> Is this the "meson doesn't actually add the cflags everywhere it should"
>>> bug again ?
>>
>> I guess so.  I realized after sending this reply that PKG_CONFIG_LIBDIR can't
>> be the answer, since the original configure should have failed if pkg-config
>> didn't find ffi.
>>
>> Was there a resolution to said meson bug?
>
> Meanwhile I noticed an additional detail:
>
> There exist two different pkg-config configurations for libffi on Homebrew:
>
> % pkg-config --cflags libffi
> -I/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk/usr/include/ffi
> % export PKG_CONFIG_PATH="/opt/homebrew/opt/libffi/lib/pkgconfig"
> % pkg-config --cflags libffi
> -I/opt/homebrew/Cellar/libffi/3.3_2/include
>
> By default it points to a system directory which does not exist at all
> on my Mac, so that will never work.
>
> With the right PKG_CONFIG_PATH a correct include directory is set, and
> the latest rebased tci-next branch now works for me with a compiler warning:
>
> /opt/homebrew/Cellar/libffi/3.3_2/include/ffi.h:441:5: warning:
> 'FFI_GO_CLOSURES' is not defined, evaluates to 0 [-Wundef]

This got introduced in
https://github.com/libffi/libffi/commit/e951d64c0852; should we report there?

Cc'ing Jeremy who sent patches around FFI_GO_CLOSURES:
https://github.com/libffi/libffi/pull/586/commits/8c25da7d2cdf8

Meanwhile, about 5 years have passed, I'll post a patch...


^ permalink raw reply	[flat|nested] 16+ messages in thread
* [PATCH v2 00/93] TCI fixes and cleanups
@ 2021-02-04  1:43 Richard Henderson
  2021-02-04  1:44 ` [PATCH v2 63/93] tcg/tci: Use ffi for calls Richard Henderson
  0 siblings, 1 reply; 16+ messages in thread
From: Richard Henderson @ 2021-02-04  1:43 UTC (permalink / raw)
  To: qemu-devel; +Cc: sw

Almost 7 years ago I detailed 5 major problems in tci[1], of
which three still remain:

  * Unaligned accesses to the bytecode stream, which means
    that we immediately SIGBUS on any host requiring alignment.
  * Non-portable calls to helper functions.
  * Full of useless ifdefs and TODOs.

To my mind, this means the code is unmaintained, despite what it
says in MAINTAINERS.  Thus tci *should* be simply removed.
However, every time removal is suggested, someone comes out of the
woodwork and says we should keep it, because it's useful for $FOO.

Anyway, if we're not going to remove tci, then we have to fix it.

Previously, I rewrote tci all in one lump.  Which was obviously a
mistake, since it meant that the patch was never going to get
reviewed.  This time I've done the rewrite in tiny pieces.

Previously, I invented a moderately complex encoding scheme that
allowed any operand to encode a register or an int32_t immediate.
This time the encoding is quite simple: with only 4 exceptions
all operands encoded into 4-bit slots (registers & conditions).
I rely on the new TEMP_CONST optimization to do a decent job of
loading an immediate value into a register once (via movi), and
reusing that across the TB.

There is a disassembler built into tcg/tci.c, replacing the stub
in disas/tci.c, which reuses the same decoding helpers that are
used by the interpreter.  So finally -d out_asm is useful.

This is good enough to pass make check check-tcg with all of the
docker cross-compilers enabled.  I can boot linux with aarch64,
alpha, and s390x guests.


r~


Based-on: 20210203021550.375058-1-richard.henderson@linaro.org
("[PULL 00/24] tcg patch queue")

[1] https://lists.gnu.org/archive/html/qemu-devel/2014-05/msg02594.html

Richard Henderson (91):
  gdbstub: Fix handle_query_xfer_auxv
  tcg: Split out tcg_raise_tb_overflow
  configure: Fix --enable-tcg-interpreter
  tcg: Manage splitwx in tc_ptr_to_region_tree by hand
  tcg/tci: Make tci_tb_ptr thread-local
  tcg/tci: Inline tci_write_reg32s into the only caller
  tcg/tci: Inline tci_write_reg8 into its callers
  tcg/tci: Inline tci_write_reg16 into the only caller
  tcg/tci: Inline tci_write_reg32 into all callers
  tcg/tci: Inline tci_write_reg64 into 64-bit callers
  tcg/tci: Merge INDEX_op_ld8u_{i32,i64}
  tcg/tci: Merge INDEX_op_ld8s_{i32,i64}
  tcg/tci: Merge INDEX_op_ld16u_{i32,i64}
  tcg/tci: Merge INDEX_op_ld16s_{i32,i64}
  tcg/tci: Merge INDEX_op_{ld_i32,ld32u_i64}
  tcg/tci: Merge INDEX_op_st8_{i32,i64}
  tcg/tci: Merge INDEX_op_st16_{i32,i64}
  tcg/tci: Move stack bounds check to compile-time
  tcg/tci: Merge INDEX_op_{st_i32,st32_i64}
  tcg/tci: Use g_assert_not_reached
  tcg/tci: Remove dead code for TCG_TARGET_HAS_div2_*
  tcg/tci: Implement 64-bit division
  tcg/tci: Remove TODO as unused
  tcg/tci: Restrict TCG_TARGET_NB_REGS to 16
  tcg/tci: Fix TCG_REG_R4 misusage
  tcg/tci: Use bool in tcg_out_ri*
  tcg/tci: Remove TCG_CONST
  tcg/tci: Merge identical cases in generation
  tcg/tci: Remove tci_read_r8
  tcg/tci: Remove tci_read_r8s
  tcg/tci: Remove tci_read_r16
  tcg/tci: Remove tci_read_r16s
  tcg/tci: Remove tci_read_r32s
  tcg/tci: Reduce use of tci_read_r64
  tcg/tci: Merge basic arithmetic operations
  tcg/tci: Merge extension operations
  tcg/tci: Remove ifdefs for TCG_TARGET_HAS_ext32[us]_i64
  tcg/tci: Merge bswap operations
  tcg/tci: Merge mov, not and neg operations
  tcg/tci: Rename tci_read_r to tci_read_rval
  tcg/tci: Split out tci_args_rrs
  tcg/tci: Split out tci_args_rr
  tcg/tci: Split out tci_args_rrr
  tcg/tci: Split out tci_args_rrrc
  tcg/tci: Split out tci_args_l
  tcg/tci: Split out tci_args_rrrrrc
  tcg/tci: Split out tci_args_rrcl and tci_args_rrrrcl
  tcg/tci: Split out tci_args_ri and tci_args_rI
  tcg/tci: Reuse tci_args_l for calls.
  tcg/tci: Reuse tci_args_l for exit_tb
  tcg/tci: Reuse tci_args_l for goto_tb
  tcg/tci: Split out tci_args_rrrrrr
  tcg/tci: Split out tci_args_rrrr
  tcg/tci: Clean up deposit operations
  tcg/tci: Reduce qemu_ld/st TCGMemOpIdx operand to 32-bits
  tcg/tci: Split out tci_args_{rrm,rrrm,rrrrm}
  tcg/tci: Hoist op_size checking into tci_args_*
  tcg/tci: Remove tci_disas
  tcg/tci: Implement the disassembler properly
  tcg: Build ffi data structures for helpers
  tcg/tci: Use ffi for calls
  tcg/tci: Improve tcg_target_call_clobber_regs
  tcg/tci: Move call-return regs to end of tcg_target_reg_alloc_order
  tcg/tci: Push opcode emit into each case
  tcg/tci: Split out tcg_out_op_rrs
  tcg/tci: Split out tcg_out_op_l
  tcg/tci: Split out tcg_out_op_p
  tcg/tci: Split out tcg_out_op_rr
  tcg/tci: Split out tcg_out_op_rrr
  tcg/tci: Split out tcg_out_op_rrrc
  tcg/tci: Split out tcg_out_op_rrrrrc
  tcg/tci: Split out tcg_out_op_rrrbb
  tcg/tci: Split out tcg_out_op_rrcl
  tcg/tci: Split out tcg_out_op_rrrrrr
  tcg/tci: Split out tcg_out_op_rrrr
  tcg/tci: Split out tcg_out_op_rrrrcl
  tcg/tci: Split out tcg_out_op_{rrm,rrrm,rrrrm}
  tcg/tci: Split out tcg_out_op_v
  tcg/tci: Split out tcg_out_op_np
  tcg/tci: Split out tcg_out_op_r[iI]
  tcg/tci: Reserve r13 for a temporary
  tcg/tci: Emit setcond before brcond
  tcg/tci: Remove tci_write_reg
  tcg/tci: Change encoding to uint32_t units
  tcg/tci: Implement goto_ptr
  tcg/tci: Implement movcond
  tcg/tci: Implement andc, orc, eqv, nand, nor
  tcg/tci: Implement extract, sextract
  tcg/tci: Implement clz, ctz, ctpop
  tcg/tci: Implement mulu2, muls2
  tcg/tci: Implement add2, sub2

Stefan Weil (2):
  tcg/tci: Implement INDEX_op_ld16s_i32
  tcg/tci: Implement INDEX_op_ld8s_i64

 configure                              |    5 +-
 meson.build                            |    9 +-
 include/exec/exec-all.h                |    2 +-
 include/exec/helper-ffi.h              |  115 ++
 include/exec/helper-tcg.h              |   24 +-
 include/tcg/tcg-opc.h                  |    6 +-
 include/tcg/tcg.h                      |    1 +
 target/hppa/helper.h                   |    2 +
 target/i386/ops_sse_header.h           |    6 +
 target/m68k/helper.h                   |    1 +
 target/ppc/helper.h                    |    3 +
 tcg/tci/tcg-target-con-set.h           |    8 +-
 tcg/tci/tcg-target.h                   |  118 +-
 disas/tci.c                            |   61 -
 gdbstub.c                              |   17 +-
 tcg/tcg-common.c                       |    4 -
 tcg/tcg.c                              |  117 +-
 tcg/tci.c                              | 1695 +++++++++++++-----------
 tcg/tci/tcg-target.c.inc               |  989 +++++++-------
 tcg/tci/README                         |   20 +-
 tests/docker/dockerfiles/fedora.docker |    1 +
 21 files changed, 1734 insertions(+), 1470 deletions(-)
 create mode 100644 include/exec/helper-ffi.h
 delete mode 100644 disas/tci.c

-- 
2.25.1



^ permalink raw reply	[flat|nested] 16+ messages in thread

end of thread, other threads:[~2025-12-02 18:38 UTC | newest]

Thread overview: 16+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2025-12-02 18:38 [PATCH v2 63/93] tcg/tci: Use ffi for calls Philippe Mathieu-Daudé
  -- strict thread matches above, loose matches on Subject: below --
2021-02-04  1:43 [PATCH v2 00/93] TCI fixes and cleanups Richard Henderson
2021-02-04  1:44 ` [PATCH v2 63/93] tcg/tci: Use ffi for calls Richard Henderson
2021-02-07 16:25   ` Stefan Weil
2021-02-07 17:39     ` Richard Henderson
2021-02-07 19:52       ` Peter Maydell
2021-02-07 20:12         ` Richard Henderson
2021-02-07 21:33           ` Stefan Weil
2021-02-08  9:20           ` Peter Maydell
2021-02-08  9:35             ` Paolo Bonzini
2021-02-08 13:07               ` Stefan Weil
2021-02-08 17:39                 ` Richard Henderson
2021-02-08 19:04                   ` Stefan Weil
2021-02-08 22:55                     ` Richard Henderson
2021-02-09 20:46                       ` Richard Henderson
2021-02-09 21:15                         ` Stefan Weil
2021-02-09 21:54                           ` Stefan Weil

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).