Re: [Qemu-devel] [PATCH v2] configure: enable --s390-pgste linker option

[Christian Borntraeger]

On 08/23/2017 09:28 AM, Christian Ehrhardt wrote:
> 
> 
> On Wed, Aug 23, 2017 at 8:53 AM, Christian Borntraeger <address@hidden 
> <mailto:address@hidden>> wrote:
> 
>     KVM guests on s390 need a different page table layout than normal
>     processes (2kb page table + 2kb page status extensions vs 2kb page table
>     only). As of today this has to be enabled via the vm.allocate_pgste
>     sysctl.
> 
>     Newer kernels (>= 4.12) on s390 check for an S390_PGSTE program header
>     and enable the pgste page table extensions in that case. This makes the
>     vm.allocate_pgste sysctl unnecessary. We enable this program header for
>     the s390 system emulation (qemu-system-s390x) if we build on s390
>     - for s390 system emulation
>     - the linker supports --s390-pgste (binutils >= 2.29)
>     - KVM is enabled
> 
>     This will allow distributions to disable the global vm.allocate_pgste
>     sysctl, which will improve the page table allocation for non KVM
>     processes as only 2kb chunks are necessary.
> 
> 
> Hi Christian,
> it is great to see context pgste come to life.
> Currently vm.allocate_pgste defaults to 0 in the kernel but as you stated 
> mostly enabled for KVM support in Distros.
> So when someone wants to disable it he has to drop the enabling (e.g. 
> /etc/sysctl.d/10-arch-specific.conf for us).
> 
> I want to be sure on the proper phasing of this - we can drop the "enabling" 
> of global pgste once for a release we:
> - do not expect/support a kernel <4.12 to run there
> - will have only qemu versions >= the one carrying this change (and have it 
> properly enabled)
> - binutils >= 2.29 to get the linking right

Yes. So I guess that for the Ubuntu case you could remove the sysctl thing for 
18.04 assuming that
this will hit qemu 2.11 and 18.04 will use 2.11.


> 
> But furthermore if we have a qemu with this enabled, there is no drawback and 
> we could still run it in:
> - former releases with older kernels

Yes.

> - former releases with older build environments

Yes.

> That program header would just be ignored and we just would have to keep the 
> sysctl enabled there right?

Yes.

> 
> Also for the time we want to check on the proper header, you surely have a 
> one liner you can share that you run against the binary to check if it was 
> generated correctly?
> Maybe even one that you can run against a pid if the status is correct?

readelf -l on the binary
$ readelf -l REPOS/qemu/build/s390x-softmmu/qemu-system-s390x 

Elf file type is EXEC (Executable file)
Entry point 0x101f758
There are 11 program headers, starting at offset 64

Program Headers:
  Type           Offset             VirtAddr           PhysAddr
                 FileSiz            MemSiz              Flags  Align
  PHDR           0x0000000000000040 0x0000000001000040 0x0000000001000040
                 0x0000000000000268 0x0000000000000268  R E    0x8
  INTERP         0x00000000000002a8 0x00000000010002a8 0x00000000010002a8
                 0x000000000000000f 0x000000000000000f  R      0x1
      [Requesting program interpreter: /lib/ld64.so.1]
  LOAD           0x0000000000000000 0x0000000001000000 0x0000000001000000
                 0x00000000004852f0 0x00000000004852f0  R E    0x1000
  LOAD           0x0000000000485450 0x0000000001486450 0x0000000001486450
                 0x000000000003dcc8 0x0000000000485840  RW     0x1000
  DYNAMIC        0x0000000000485b80 0x0000000001486b80 0x0000000001486b80
                 0x0000000000000480 0x0000000000000480  RW     0x8
  NOTE           0x00000000000002b8 0x00000000010002b8 0x00000000010002b8
                 0x0000000000000044 0x0000000000000044  R      0x4
  TLS            0x0000000000485450 0x0000000001486450 0x0000000001486450
                 0x0000000000000000 0x0000000000000230  R      0x8
  GNU_EH_FRAME   0x00000000003dc638 0x00000000013dc638 0x00000000013dc638
                 0x0000000000017a74 0x0000000000017a74  R      0x4
  GNU_STACK      0x0000000000000000 0x0000000000000000 0x0000000000000000
                 0x0000000000000000 0x0000000000000000  RW     0x10
  GNU_RELRO      0x0000000000485450 0x0000000001486450 0x0000000001486450
                 0x0000000000000bb0 0x0000000000000bb0  R      0x1
  S390_PGSTE     0x0000000000000000 0x0000000000000000 0x0000000000000000   
<----
                 0x0000000000000000 0x0000000000000000         0x8          
<----

[...]

Older binutils will report something like

  LOPROC+0       0x0000000000000000 0x0000000000000000 0x0000000000000000
                 0x0000000000000000 0x0000000000000000         8

instead of S390_PGSTE.