3.18+: soft-float userland unusable due to .MIPS.abiflags patch

3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Manuel Lauss]

Hi Paul,

Your patch commit 90cee759f08a6b7a8daab9977d3e163ebbcac220
("MIPS: ELF: Set FP mode according to .MIPS.abiflags") completely
breaks my pure soft-float o32 userland:

[...]
Freeing unused kernel memory: 244K (80993000 - 809d0000)
Failed to execute /usr/lib/systemd/systemd (error -84).  Attempting defaults...
Starting init: /sbin/init exists but couldn't execute it (error -84)
sh: cannot set terminal process group (-1): Inappropriate ioctl for device
sh: no job control in this shell
sh-4.3# ls
sh: /bin/ls: Accessing a corrupted shared library
sh-4.3#

I've recently rebuilt bash, ncurses, readline and glibc-2.20 (with
binutils 2.25+)
to track down another userland issue, so that may explain why at least
sh is able to run.

Reverting the patch fixes it for me.


Thanks!
      Manuel

Re: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Manuel Lauss]

On Fri, Jan 16, 2015 at 6:59 PM, Manuel Lauss <manuel.lauss@gmail.com> wrote:
> Hi Paul,
>
> Your patch commit 90cee759f08a6b7a8daab9977d3e163ebbcac220
> ("MIPS: ELF: Set FP mode according to .MIPS.abiflags") completely
> breaks my pure soft-float o32 userland:
>
> [...]
> Freeing unused kernel memory: 244K (80993000 - 809d0000)
> Failed to execute /usr/lib/systemd/systemd (error -84).  Attempting defaults...
> Starting init: /sbin/init exists but couldn't execute it (error -84)
> sh: cannot set terminal process group (-1): Inappropriate ioctl for device
> sh: no job control in this shell
> sh-4.3# ls
> sh: /bin/ls: Accessing a corrupted shared library
> sh-4.3#
>
> I've recently rebuilt bash, ncurses, readline and glibc-2.20 (with
> binutils 2.25+)
> to track down another userland issue, so that may explain why at least
> sh is able to run.

Although ls and all its dependencies have the following soft-float abi tag:
Displaying notes found at file offset 0x00000164 with length 0x00000020:
  Owner                 Data size       Description
  GNU                  0x00000010       NT_GNU_ABI_TAG (ABI version tag)
    OS: Linux, ABI: 2.6.32
Attribute Section: gnu
File Attributes
  Tag_GNU_MIPS_ABI_FP: Soft float


Manuel

RE: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Matthew Fortune]

> On Fri, Jan 16, 2015 at 6:59 PM, Manuel Lauss <manuel.lauss@gmail.com>
> wrote:
> > Hi Paul,
> >
> > Your patch commit 90cee759f08a6b7a8daab9977d3e163ebbcac220
> > ("MIPS: ELF: Set FP mode according to .MIPS.abiflags") completely
> > breaks my pure soft-float o32 userland:
> >
> > [...]
> > Freeing unused kernel memory: 244K (80993000 - 809d0000) Failed to
> > execute /usr/lib/systemd/systemd (error -84).  Attempting defaults...
> > Starting init: /sbin/init exists but couldn't execute it (error -84)
> > sh: cannot set terminal process group (-1): Inappropriate ioctl for
> > device
> > sh: no job control in this shell
> > sh-4.3# ls
> > sh: /bin/ls: Accessing a corrupted shared library sh-4.3#
> >
> > I've recently rebuilt bash, ncurses, readline and glibc-2.20 (with
> > binutils 2.25+) to track down another userland issue, so that may
> > explain why at least sh is able to run.
> 
> Although ls and all its dependencies have the following soft-float abi
> tag:
> Displaying notes found at file offset 0x00000164 with length 0x00000020:
>   Owner                 Data size       Description
>   GNU                  0x00000010       NT_GNU_ABI_TAG (ABI version tag)
>     OS: Linux, ABI: 2.6.32
> Attribute Section: gnu
> File Attributes
>   Tag_GNU_MIPS_ABI_FP: Soft float

Can you provide the output of readelf -hl for 'ls', 'init' and whatever is
listed as the interpreter for those executables in the header output shown
like:

  INTERP         0x0000000000000238 0x0000000000400238 0x0000000000400238
                 0x000000000000001c 0x000000000000001c  R      1
      [Requesting program interpreter: /lib64/ld-linux-x86-64.so.2]


Thanks,
Matthew

Re: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Manuel Lauss]

On Fri, Jan 16, 2015 at 8:12 PM, Matthew Fortune
<Matthew.Fortune@imgtec.com> wrote:
>> On Fri, Jan 16, 2015 at 6:59 PM, Manuel Lauss <manuel.lauss@gmail.com>
>> wrote:
>> > Hi Paul,
>> >
>> > Your patch commit 90cee759f08a6b7a8daab9977d3e163ebbcac220
>> > ("MIPS: ELF: Set FP mode according to .MIPS.abiflags") completely
>> > breaks my pure soft-float o32 userland:
>> >
>> > [...]
>> > Freeing unused kernel memory: 244K (80993000 - 809d0000) Failed to
>> > execute /usr/lib/systemd/systemd (error -84).  Attempting defaults...
>> > Starting init: /sbin/init exists but couldn't execute it (error -84)
>> > sh: cannot set terminal process group (-1): Inappropriate ioctl for
>> > device
>> > sh: no job control in this shell
>> > sh-4.3# ls
>> > sh: /bin/ls: Accessing a corrupted shared library sh-4.3#
>> >
>> > I've recently rebuilt bash, ncurses, readline and glibc-2.20 (with
>> > binutils 2.25+) to track down another userland issue, so that may
>> > explain why at least sh is able to run.
>>
>> Although ls and all its dependencies have the following soft-float abi
>> tag:
>> Displaying notes found at file offset 0x00000164 with length 0x00000020:
>>   Owner                 Data size       Description
>>   GNU                  0x00000010       NT_GNU_ABI_TAG (ABI version tag)
>>     OS: Linux, ABI: 2.6.32
>> Attribute Section: gnu
>> File Attributes
>>   Tag_GNU_MIPS_ABI_FP: Soft float
>
> Can you provide the output of readelf -hl for 'ls', 'init' and whatever is
> listed as the interpreter for those executables in the header output shown
> like:
>
>   INTERP         0x0000000000000238 0x0000000000400238 0x0000000000400238
>                  0x000000000000001c 0x000000000000001c  R      1
>       [Requesting program interpreter: /lib64/ld-linux-x86-64.so.2]


Here's the output, identical on all tested binaries and libs:

# readelf -hl /bin/ls
ELF Header:
  Magic:   7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00
  Class:                             ELF32
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              EXEC (Executable file)
  Machine:                           MIPS R3000
  Version:                           0x1
  Entry point address:               0x403f20
  Start of program headers:          52 (bytes into file)
  Start of section headers:          131784 (bytes into file)
  Flags:                             0x50001005, noreorder, cpic, o32, mips32
  Size of this header:               52 (bytes)
  Size of program headers:           32 (bytes)
  Number of program headers:         9
  Size of section headers:           40 (bytes)
  Number of section headers:         33
  Section header string table index: 32

Program Headers:
  Type           Offset   VirtAddr   PhysAddr   FileSiz MemSiz  Flg Align
  PHDR           0x000034 0x00400034 0x00400034 0x00120 0x00120 R E 0x4
  INTERP         0x000154 0x00400154 0x00400154 0x0000d 0x0000d R   0x1
      [Requesting program interpreter: /lib/ld.so.1]
  REGINFO        0x000184 0x00400184 0x00400184 0x00018 0x00018 R   0x4
  LOAD           0x000000 0x00400000 0x00400000 0x1d290 0x1d290 R E 0x10000
  LOAD           0x01e000 0x0042e000 0x0042e000 0x00378 0x00fb0 RW  0x10000
  DYNAMIC        0x00019c 0x0040019c 0x0040019c 0x00140 0x00140 R   0x4
  NOTE           0x000164 0x00400164 0x00400164 0x00020 0x00020 R   0x4
  GNU_EH_FRAME   0x01cb30 0x0041cb30 0x0041cb30 0x0002c 0x0002c R   0x4
  NULL           0x000000 0x00000000 0x00000000 0x00000 0x00000     0x4

 Section to Segment mapping:
  Segment Sections...
   00
   01     .interp
   02     .reginfo
   03     .interp .note.ABI-tag .reginfo .dynamic .hash .dynsym
.dynstr .gnu.version .gnu.version_r .rel.dyn .rel.plt .init .text
.MIPS.stubs .fini .rodata .eh_frame_hdr .eh_frame .plt
   04     .init_array .fini_array .jcr .got.plt .data .rld_map .got .sdata .bss
   05     .dynamic
   06     .note.ABI-tag
   07     .eh_frame_hdr
   08


# readelf -hl /usr/lib/systemd/systemd
ELF Header:
  Magic:   7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00
  Class:                             ELF32
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              DYN (Shared object file)
  Machine:                           MIPS R3000
  Version:                           0x1
  Entry point address:               0x165a0
  Start of program headers:          52 (bytes into file)
  Start of section headers:          1574780 (bytes into file)
  Flags:                             0x50001005, noreorder, cpic, o32, mips32
  Size of this header:               52 (bytes)
  Size of program headers:           32 (bytes)
  Number of program headers:         10
  Size of section headers:           40 (bytes)
  Number of section headers:         32
  Section header string table index: 31

Program Headers:
  Type           Offset   VirtAddr   PhysAddr   FileSiz MemSiz  Flg Align
  PHDR           0x000034 0x00000034 0x00000034 0x00140 0x00140 R   0x4
  INTERP         0x000174 0x00000174 0x00000174 0x0000d 0x0000d R   0x1
      [Requesting program interpreter: /lib/ld.so.1]
  LOAD           0x000000 0x00000000 0x00000000 0x148bf8 0x148bf8 R E 0x10000
  LOAD           0x14ef68 0x0015ef68 0x0015ef68 0x1a788 0x1b06c RW  0x10000
  NULL           0x000000 0x00000000 0x00000000 0x00000 0x00000     0
  DYNAMIC        0x15feb0 0x0016feb0 0x0016feb0 0x00150 0x00150 RW  0x4
  NOTE           0x000184 0x00000184 0x00000184 0x00020 0x00020 R   0x4
  GNU_EH_FRAME   0x148bd4 0x00148bd4 0x00148bd4 0x00024 0x00024 R   0x4
  TLS            0x14ef68 0x0015ef68 0x0015ef68 0x00008 0x00061 R   0x8
  GNU_RELRO      0x14ef68 0x0015ef68 0x0015ef68 0x11098 0x11098 RW  0x8

Re: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Manuel Lauss]

On Fri, Jan 16, 2015 at 9:01 PM, Manuel Lauss <manuel.lauss@gmail.com> wrote:
> On Fri, Jan 16, 2015 at 8:12 PM, Matthew Fortune
> <Matthew.Fortune@imgtec.com> wrote:
>>> On Fri, Jan 16, 2015 at 6:59 PM, Manuel Lauss <manuel.lauss@gmail.com>
>>> wrote:
>>> > Hi Paul,
>>> >
>>> > Your patch commit 90cee759f08a6b7a8daab9977d3e163ebbcac220
>>> > ("MIPS: ELF: Set FP mode according to .MIPS.abiflags") completely
>>> > breaks my pure soft-float o32 userland:
>>> >
>>> > [...]
>>> > Freeing unused kernel memory: 244K (80993000 - 809d0000) Failed to
>>> > execute /usr/lib/systemd/systemd (error -84).  Attempting defaults...
>>> > Starting init: /sbin/init exists but couldn't execute it (error -84)
>>> > sh: cannot set terminal process group (-1): Inappropriate ioctl for
>>> > device
>>> > sh: no job control in this shell
>>> > sh-4.3# ls
>>> > sh: /bin/ls: Accessing a corrupted shared library sh-4.3#
>>> >
>>> > I've recently rebuilt bash, ncurses, readline and glibc-2.20 (with
>>> > binutils 2.25+) to track down another userland issue, so that may
>>> > explain why at least sh is able to run.
>>>
>>> Although ls and all its dependencies have the following soft-float abi
>>> tag:
>>> Displaying notes found at file offset 0x00000164 with length 0x00000020:
>>>   Owner                 Data size       Description
>>>   GNU                  0x00000010       NT_GNU_ABI_TAG (ABI version tag)
>>>     OS: Linux, ABI: 2.6.32
>>> Attribute Section: gnu
>>> File Attributes
>>>   Tag_GNU_MIPS_ABI_FP: Soft float
>>
>> Can you provide the output of readelf -hl for 'ls', 'init' and whatever is
>> listed as the interpreter for those executables in the header output shown
>> like:
>>
>>   INTERP         0x0000000000000238 0x0000000000400238 0x0000000000400238
>>                  0x000000000000001c 0x000000000000001c  R      1
>>       [Requesting program interpreter: /lib64/ld-linux-x86-64.so.2]


I noticed that /bin/sh does have a .MIPS.abiflags section while systemd
and ls do not.  Maybe that's the issue here?

 Section to Segment mapping:
  Segment Sections...
   00
   01     .interp
   02     .MIPS.abiflags

 Manuel

RE: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Matthew Fortune]

Manuel Lauss <manuel.lauss@gmail.com> writes:
> On Fri, Jan 16, 2015 at 9:01 PM, Manuel Lauss <manuel.lauss@gmail.com>
> wrote:
> > On Fri, Jan 16, 2015 at 8:12 PM, Matthew Fortune
> > <Matthew.Fortune@imgtec.com> wrote:
> >>> On Fri, Jan 16, 2015 at 6:59 PM, Manuel Lauss
> >>> <manuel.lauss@gmail.com>
> >>> wrote:
> >>> > Hi Paul,
> >>> >
> >>> > Your patch commit 90cee759f08a6b7a8daab9977d3e163ebbcac220
> >>> > ("MIPS: ELF: Set FP mode according to .MIPS.abiflags") completely
> >>> > breaks my pure soft-float o32 userland:
> >>> >
> >>> > [...]
> >>> > Freeing unused kernel memory: 244K (80993000 - 809d0000) Failed to
> >>> > execute /usr/lib/systemd/systemd (error -84).  Attempting
> defaults...
> >>> > Starting init: /sbin/init exists but couldn't execute it (error
> >>> > -84)
> >>> > sh: cannot set terminal process group (-1): Inappropriate ioctl
> >>> > for device
> >>> > sh: no job control in this shell
> >>> > sh-4.3# ls
> >>> > sh: /bin/ls: Accessing a corrupted shared library sh-4.3#
> >>> >
> >>> > I've recently rebuilt bash, ncurses, readline and glibc-2.20 (with
> >>> > binutils 2.25+) to track down another userland issue, so that may
> >>> > explain why at least sh is able to run.
> >>>
> >>> Although ls and all its dependencies have the following soft-float
> >>> abi
> >>> tag:
> >>> Displaying notes found at file offset 0x00000164 with length
> 0x00000020:
> >>>   Owner                 Data size       Description
> >>>   GNU                  0x00000010       NT_GNU_ABI_TAG (ABI version
> tag)
> >>>     OS: Linux, ABI: 2.6.32
> >>> Attribute Section: gnu
> >>> File Attributes
> >>>   Tag_GNU_MIPS_ABI_FP: Soft float
> >>
> >> Can you provide the output of readelf -hl for 'ls', 'init' and
> >> whatever is listed as the interpreter for those executables in the
> >> header output shown
> >> like:
> >>
> >>   INTERP         0x0000000000000238 0x0000000000400238
> 0x0000000000400238
> >>                  0x000000000000001c 0x000000000000001c  R      1
> >>       [Requesting program interpreter: /lib64/ld-linux-x86-64.so.2]
> 
> 
> I noticed that /bin/sh does have a .MIPS.abiflags section while systemd
> and ls do not.  Maybe that's the issue here?
> 
>  Section to Segment mapping:
>   Segment Sections...
>    00
>    01     .interp
>    02     .MIPS.abiflags

It is the ABIFLAGS program header that I am focussed on but the presence of
a .MIPS.abiflags section is a strong indicator that there will be an
ABIFLAGS pheader. The key is in the readelf output for the interpreter i.e.
/lib/ld.so.1. I would expect it has the ABIFLAGS header too given you
said that you recently rebuilt glibc.

You are right that it is the .MIPS.abiflags patch that is causing your
trouble. For a long time I had to put a restriction in the ABI plan that
soft-float binaries without an ABIFLAGS pheader could not be linked against
soft-float binaries with an ABIFLAGS pheader. We have since found a way to
relax that restriction without reducing the effectiveness of the new
compatibility checks. I would need to check the code in the kernel but I 
suspect that is the issue. Markos has done a significant update to this
piece of code which he posted earlier today. That updated version should
allow the combination of soft-float without ABIFLAGS and soft-float with
ABIFLAGS.

Matthew

Re: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Aaro Koskinen]

On Fri, Jan 16, 2015 at 08:36:12PM +0000, Matthew Fortune wrote:
> You are right that it is the .MIPS.abiflags patch that is causing your
> trouble. For a long time I had to put a restriction in the ABI plan that
> soft-float binaries without an ABIFLAGS pheader could not be linked against
> soft-float binaries with an ABIFLAGS pheader. We have since found a way to
> relax that restriction without reducing the effectiveness of the new
> compatibility checks. I would need to check the code in the kernel but I 
> suspect that is the issue. Markos has done a significant update to this
> piece of code which he posted earlier today. That updated version should
> allow the combination of soft-float without ABIFLAGS and soft-float with
> ABIFLAGS.

Are you referring to the series with 70 patches? I think a fix that passes
stable kernel rules is needed.

A.

RE: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Matthew Fortune]

Aaro Koskinen <aaro.koskinen@iki.fi> writes:
> On Fri, Jan 16, 2015 at 08:36:12PM +0000, Matthew Fortune wrote:
> > You are right that it is the .MIPS.abiflags patch that is causing your
> > trouble. For a long time I had to put a restriction in the ABI plan
> > that soft-float binaries without an ABIFLAGS pheader could not be
> > linked against soft-float binaries with an ABIFLAGS pheader. We have
> > since found a way to relax that restriction without reducing the
> > effectiveness of the new compatibility checks. I would need to check
> > the code in the kernel but I suspect that is the issue. Markos has
> > done a significant update to this piece of code which he posted
> > earlier today. That updated version should allow the combination of
> > soft-float without ABIFLAGS and soft-float with ABIFLAGS.
> 
> Are you referring to the series with 70 patches? I think a fix that
> passes stable kernel rules is needed.

Yes it was just one patch though for this issue:
[PATCH RFC v2 68/70] MIPS: kernel: elf: Improve the overall ABI and FPU
mode checks

I wasn't trying to suggest how to fix the existing code just explaining
how it came to be and what has been done about it for next release.
(I'm not a kernel developer I'm just interested as I did most of the
design work for the new ABI extensions.)

I guess there are three options:
a) revert the patch - That would remove the new ABI safety measures from
   3.19 which is a shame given it has MSA support in it (I think anyway).
   equally given that the new prctl FPU mode options did not make 3.19
   then I suppose it doesn't lose too much either as the two features
   go hand in hand to some extent.
b) Fix the broken case - I doubt it will be too challenging to fix up
   the implementation to allow the soft-float ABIFLAGS + no ABIFLAGS
   combination.
c) Apply Markos' updated patch (with the references to r6 removed).

I'll leave it to others to decide which approach is best.

Thanks,
Matthew

Re: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Manuel Lauss]

On Sat, Jan 17, 2015 at 8:00 PM, Matthew Fortune
<Matthew.Fortune@imgtec.com> wrote:
> Aaro Koskinen <aaro.koskinen@iki.fi> writes:
>> On Fri, Jan 16, 2015 at 08:36:12PM +0000, Matthew Fortune wrote:
>> > You are right that it is the .MIPS.abiflags patch that is causing your
>> > trouble. For a long time I had to put a restriction in the ABI plan
>> > that soft-float binaries without an ABIFLAGS pheader could not be
>> > linked against soft-float binaries with an ABIFLAGS pheader. We have
>> > since found a way to relax that restriction without reducing the
>> > effectiveness of the new compatibility checks. I would need to check
>> > the code in the kernel but I suspect that is the issue. Markos has
>> > done a significant update to this piece of code which he posted
>> > earlier today. That updated version should allow the combination of
>> > soft-float without ABIFLAGS and soft-float with ABIFLAGS.
>>
>> Are you referring to the series with 70 patches? I think a fix that
>> passes stable kernel rules is needed.
>
> Yes it was just one patch though for this issue:
> [PATCH RFC v2 68/70] MIPS: kernel: elf: Improve the overall ABI and FPU
> mode checks
>
> I wasn't trying to suggest how to fix the existing code just explaining
> how it came to be and what has been done about it for next release.
> (I'm not a kernel developer I'm just interested as I did most of the
> design work for the new ABI extensions.)
>
> I guess there are three options:
> a) revert the patch - That would remove the new ABI safety measures from
>    3.19 which is a shame given it has MSA support in it (I think anyway).
>    equally given that the new prctl FPU mode options did not make 3.19
>    then I suppose it doesn't lose too much either as the two features
>    go hand in hand to some extent.

I favor this one.  I don't know how many systems with MSA are in the wild,
and if there are any, I'm sure they're using some mti/imgtec-supplied kernel
anyway.  Another thing I noticed last time is that companies shipping MIPS
products rarely upgrade their toolchains, so I'm sure the ABI safety measures
can wait for another release, but then function with all configurations
in the wild.


Manuel

Re: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Paul Burton]

On Sun, Jan 18, 2015 at 11:35:31AM +0100, Manuel Lauss wrote:
> On Sat, Jan 17, 2015 at 8:00 PM, Matthew Fortune
> <Matthew.Fortune@imgtec.com> wrote:
> > Aaro Koskinen <aaro.koskinen@iki.fi> writes:
> >> On Fri, Jan 16, 2015 at 08:36:12PM +0000, Matthew Fortune wrote:
> >> > You are right that it is the .MIPS.abiflags patch that is causing your
> >> > trouble. For a long time I had to put a restriction in the ABI plan
> >> > that soft-float binaries without an ABIFLAGS pheader could not be
> >> > linked against soft-float binaries with an ABIFLAGS pheader. We have
> >> > since found a way to relax that restriction without reducing the
> >> > effectiveness of the new compatibility checks. I would need to check
> >> > the code in the kernel but I suspect that is the issue. Markos has
> >> > done a significant update to this piece of code which he posted
> >> > earlier today. That updated version should allow the combination of
> >> > soft-float without ABIFLAGS and soft-float with ABIFLAGS.
> >>
> >> Are you referring to the series with 70 patches? I think a fix that
> >> passes stable kernel rules is needed.
> >
> > Yes it was just one patch though for this issue:
> > [PATCH RFC v2 68/70] MIPS: kernel: elf: Improve the overall ABI and FPU
> > mode checks
> >
> > I wasn't trying to suggest how to fix the existing code just explaining
> > how it came to be and what has been done about it for next release.
> > (I'm not a kernel developer I'm just interested as I did most of the
> > design work for the new ABI extensions.)
> >
> > I guess there are three options:
> > a) revert the patch - That would remove the new ABI safety measures from
> >    3.19 which is a shame given it has MSA support in it (I think anyway).
> >    equally given that the new prctl FPU mode options did not make 3.19
> >    then I suppose it doesn't lose too much either as the two features
> >    go hand in hand to some extent.
> 
> I favor this one.  I don't know how many systems with MSA are in the wild,
> and if there are any, I'm sure they're using some mti/imgtec-supplied kernel
> anyway.  Another thing I noticed last time is that companies shipping MIPS
> products rarely upgrade their toolchains, so I'm sure the ABI safety measures
> can wait for another release, but then function with all configurations
> in the wild.
> 
> Manuel

An alternative would be the patch I just submitted, which makes the mode
checks conditional upon CONFIG_MIPS_O32_FP64_SUPPORT:

  http://marc.info/?l=linux-mips&m=142164553017027&w=2

Assuming this fixes your problem, and I believe it should, it would
avoid the churn of reverting the patch & readding the modified logic
again later.

Thanks,
    Paul

Re: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Markos Chandras]

On 01/19/2015 05:36 AM, Paul Burton wrote:
> On Sun, Jan 18, 2015 at 11:35:31AM +0100, Manuel Lauss wrote:
>> On Sat, Jan 17, 2015 at 8:00 PM, Matthew Fortune
>> <Matthew.Fortune@imgtec.com> wrote:
>>> Aaro Koskinen <aaro.koskinen@iki.fi> writes:
>>>> On Fri, Jan 16, 2015 at 08:36:12PM +0000, Matthew Fortune wrote:
>>>>> You are right that it is the .MIPS.abiflags patch that is causing your
>>>>> trouble. For a long time I had to put a restriction in the ABI plan
>>>>> that soft-float binaries without an ABIFLAGS pheader could not be
>>>>> linked against soft-float binaries with an ABIFLAGS pheader. We have
>>>>> since found a way to relax that restriction without reducing the
>>>>> effectiveness of the new compatibility checks. I would need to check
>>>>> the code in the kernel but I suspect that is the issue. Markos has
>>>>> done a significant update to this piece of code which he posted
>>>>> earlier today. That updated version should allow the combination of
>>>>> soft-float without ABIFLAGS and soft-float with ABIFLAGS.
>>>>
>>>> Are you referring to the series with 70 patches? I think a fix that
>>>> passes stable kernel rules is needed.
>>>
>>> Yes it was just one patch though for this issue:
>>> [PATCH RFC v2 68/70] MIPS: kernel: elf: Improve the overall ABI and FPU
>>> mode checks
>>>
>>> I wasn't trying to suggest how to fix the existing code just explaining
>>> how it came to be and what has been done about it for next release.
>>> (I'm not a kernel developer I'm just interested as I did most of the
>>> design work for the new ABI extensions.)
>>>
>>> I guess there are three options:
>>> a) revert the patch - That would remove the new ABI safety measures from
>>>    3.19 which is a shame given it has MSA support in it (I think anyway).
>>>    equally given that the new prctl FPU mode options did not make 3.19
>>>    then I suppose it doesn't lose too much either as the two features
>>>    go hand in hand to some extent.
>>
>> I favor this one.  I don't know how many systems with MSA are in the wild,
>> and if there are any, I'm sure they're using some mti/imgtec-supplied kernel
>> anyway.  Another thing I noticed last time is that companies shipping MIPS
>> products rarely upgrade their toolchains, so I'm sure the ABI safety measures
>> can wait for another release, but then function with all configurations
>> in the wild.
>>
>> Manuel
> 
> An alternative would be the patch I just submitted, which makes the mode
> checks conditional upon CONFIG_MIPS_O32_FP64_SUPPORT:
> 
>   http://marc.info/?l=linux-mips&m=142164553017027&w=2
> 
> Assuming this fixes your problem, and I believe it should, it would
> avoid the churn of reverting the patch & readding the modified logic
> again later.
> 
> Thanks,
>     Paul
> 
There is also this patch from James for 3.19 final

http://patchwork.linux-mips.org/patch/8932/

so with these two patches we should be good for 3.19.

-- 
markos

Re: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Aaro Koskinen]

Hi,

On Mon, Jan 19, 2015 at 09:15:58AM +0000, Markos Chandras wrote:
> On 01/19/2015 05:36 AM, Paul Burton wrote:
> > On Sun, Jan 18, 2015 at 11:35:31AM +0100, Manuel Lauss wrote:
> >> On Sat, Jan 17, 2015 at 8:00 PM, Matthew Fortune
> >> <Matthew.Fortune@imgtec.com> wrote:
> >>> Aaro Koskinen <aaro.koskinen@iki.fi> writes:
> >>>> On Fri, Jan 16, 2015 at 08:36:12PM +0000, Matthew Fortune wrote:
> >>>>> You are right that it is the .MIPS.abiflags patch that is causing your
> >>>>> trouble. For a long time I had to put a restriction in the ABI plan
> >>>>> that soft-float binaries without an ABIFLAGS pheader could not be
> >>>>> linked against soft-float binaries with an ABIFLAGS pheader. We have
> >>>>> since found a way to relax that restriction without reducing the
> >>>>> effectiveness of the new compatibility checks. I would need to check
> >>>>> the code in the kernel but I suspect that is the issue. Markos has
> >>>>> done a significant update to this piece of code which he posted
> >>>>> earlier today. That updated version should allow the combination of
> >>>>> soft-float without ABIFLAGS and soft-float with ABIFLAGS.
> >>>>
> >>>> Are you referring to the series with 70 patches? I think a fix that
> >>>> passes stable kernel rules is needed.
> >>>
> >>> Yes it was just one patch though for this issue:
> >>> [PATCH RFC v2 68/70] MIPS: kernel: elf: Improve the overall ABI and FPU
> >>> mode checks
> >>>
> >>> I wasn't trying to suggest how to fix the existing code just explaining
> >>> how it came to be and what has been done about it for next release.
> >>> (I'm not a kernel developer I'm just interested as I did most of the
> >>> design work for the new ABI extensions.)
> >>>
> >>> I guess there are three options:
> >>> a) revert the patch - That would remove the new ABI safety measures from
> >>>    3.19 which is a shame given it has MSA support in it (I think anyway).
> >>>    equally given that the new prctl FPU mode options did not make 3.19
> >>>    then I suppose it doesn't lose too much either as the two features
> >>>    go hand in hand to some extent.
> >>
> >> I favor this one.  I don't know how many systems with MSA are in the wild,
> >> and if there are any, I'm sure they're using some mti/imgtec-supplied kernel
> >> anyway.  Another thing I noticed last time is that companies shipping MIPS
> >> products rarely upgrade their toolchains, so I'm sure the ABI safety measures
> >> can wait for another release, but then function with all configurations
> >> in the wild.
> >>
> >> Manuel
> > 
> > An alternative would be the patch I just submitted, which makes the mode
> > checks conditional upon CONFIG_MIPS_O32_FP64_SUPPORT:
> > 
> >   http://marc.info/?l=linux-mips&m=142164553017027&w=2
> > 
> > Assuming this fixes your problem, and I believe it should, it would
> > avoid the churn of reverting the patch & readding the modified logic
> > again later.
> > 
> > Thanks,
> >     Paul
> > 
> There is also this patch from James for 3.19 final
> 
> http://patchwork.linux-mips.org/patch/8932/
> 
> so with these two patches we should be good for 3.19.

This patch is also needed to run hard-float O32 userspace compiled using
binutils 2.25 with 64-bit 3.19-rc5 (otherwise kernel fails to start init).

So please ensure that these patches end up to next 3.19-rc ASAP!

I have to say this O32 "ABI extension" seems to be very poorly introduced.
For softfloat, the current 2.25 binutils broke _all_ existing
GCC releases. And for hardfloat, newly compiled userspace appears
to be unusable with the current kernel mainline...

:-(

A.

Re: 3.18+: soft-float userland unusable due to .MIPS.abiflags patch

[Aaro Koskinen]

Hi,

On Wed, Jan 21, 2015 at 02:00:33AM +0200, Aaro Koskinen wrote:
> On Mon, Jan 19, 2015 at 09:15:58AM +0000, Markos Chandras wrote:
> > On 01/19/2015 05:36 AM, Paul Burton wrote:
> > > On Sun, Jan 18, 2015 at 11:35:31AM +0100, Manuel Lauss wrote:
> > >> On Sat, Jan 17, 2015 at 8:00 PM, Matthew Fortune
> > >> <Matthew.Fortune@imgtec.com> wrote:
> > >>> Aaro Koskinen <aaro.koskinen@iki.fi> writes:
> > >>>> On Fri, Jan 16, 2015 at 08:36:12PM +0000, Matthew Fortune wrote:
> > >>>>> You are right that it is the .MIPS.abiflags patch that is causing your
> > >>>>> trouble. For a long time I had to put a restriction in the ABI plan
> > >>>>> that soft-float binaries without an ABIFLAGS pheader could not be
> > >>>>> linked against soft-float binaries with an ABIFLAGS pheader. We have
> > >>>>> since found a way to relax that restriction without reducing the
> > >>>>> effectiveness of the new compatibility checks. I would need to check
> > >>>>> the code in the kernel but I suspect that is the issue. Markos has
> > >>>>> done a significant update to this piece of code which he posted
> > >>>>> earlier today. That updated version should allow the combination of
> > >>>>> soft-float without ABIFLAGS and soft-float with ABIFLAGS.
> > >>>>
> > >>>> Are you referring to the series with 70 patches? I think a fix that
> > >>>> passes stable kernel rules is needed.
> > >>>
> > >>> Yes it was just one patch though for this issue:
> > >>> [PATCH RFC v2 68/70] MIPS: kernel: elf: Improve the overall ABI and FPU
> > >>> mode checks
> > >>>
> > >>> I wasn't trying to suggest how to fix the existing code just explaining
> > >>> how it came to be and what has been done about it for next release.
> > >>> (I'm not a kernel developer I'm just interested as I did most of the
> > >>> design work for the new ABI extensions.)
> > >>>
> > >>> I guess there are three options:
> > >>> a) revert the patch - That would remove the new ABI safety measures from
> > >>>    3.19 which is a shame given it has MSA support in it (I think anyway).
> > >>>    equally given that the new prctl FPU mode options did not make 3.19
> > >>>    then I suppose it doesn't lose too much either as the two features
> > >>>    go hand in hand to some extent.
> > >>
> > >> I favor this one.  I don't know how many systems with MSA are in the wild,
> > >> and if there are any, I'm sure they're using some mti/imgtec-supplied kernel
> > >> anyway.  Another thing I noticed last time is that companies shipping MIPS
> > >> products rarely upgrade their toolchains, so I'm sure the ABI safety measures
> > >> can wait for another release, but then function with all configurations
> > >> in the wild.
> > >>
> > >> Manuel
> > > 
> > > An alternative would be the patch I just submitted, which makes the mode
> > > checks conditional upon CONFIG_MIPS_O32_FP64_SUPPORT:
> > > 
> > >   http://marc.info/?l=linux-mips&m=142164553017027&w=2
> > > 
> > > Assuming this fixes your problem, and I believe it should, it would
> > > avoid the churn of reverting the patch & readding the modified logic
> > > again later.
> > > 
> > > Thanks,
> > >     Paul
> > > 
> > There is also this patch from James for 3.19 final
> > 
> > http://patchwork.linux-mips.org/patch/8932/
> > 
> > so with these two patches we should be good for 3.19.
> 
> This patch is also needed to run hard-float O32 userspace compiled using
> binutils 2.25 with 64-bit 3.19-rc5 (otherwise kernel fails to start init).

Seems to be still missing in 3.19-rc6...

A.