[Qemu-devel] [PATCH v2 00/20] Invert Endian bit in SPARCv9 MMU TTE

* [Qemu-devel] [PATCH v2 00/20] Invert Endian bit in SPARCv9 MMU TTE
@ 2019-07-22 15:34 ` tony.nguyen
  0 siblings, 0 replies; 120+ messages in thread
From: tony.nguyen @ 2019-07-22 15:34 UTC (permalink / raw)
  To: qemu-devel
  Cc: peter.maydell, walling, david, palmer, mark.cave-ayland,
	Alistair.Francis, arikalo, mst, pasic, borntraeger, rth,
	atar4qemu, ehabkost, sw, alex.williamson, qemu-arm, david,
	qemu-riscv, cohuck, claudio.fontana, qemu-s390x, qemu-ppc,
	amarkovic, pbonzini, aurelien

This patchset implements the IE (Invert Endian) bit in SPARCv9 MMU TTE.

It is an attempt of the instructions outlined by Richard Henderson to Mark
Cave-Ayland.

Tested with OpenBSD on sun4u. Solaris 10 is my actual goal, but unfortunately a
separate keyboard issue remains in the way.

On 01/11/17 19:15, Mark Cave-Ayland wrote:

>On 15/08/17 19:10, Richard Henderson wrote:
>
>> [CC Peter re MemTxAttrs below]
>> 
>> On 08/15/2017 09:38 AM, Mark Cave-Ayland wrote:
>>> Working through an incorrect endian issue on qemu-system-sparc64, it has
>>> become apparent that at least one OS makes use of the IE (Invert Endian)
>>> bit in the SPARCv9 MMU TTE to map PCI memory space without the
>>> programmer having to manually endian-swap accesses.
>>>
>>> In other words, to quote the UltraSPARC specification: "if this bit is
>>> set, accesses to the associated page are processed with inverse
>>> endianness from what is specified by the instruction (big-for-little and
>>> little-for-big)".

A good explanation by Mark why the IE bit is required.

>>>
>>> Looking through various bits of code, I'm trying to get a feel for the
>>> best way to implement this in an efficient manner. From what I can see
>>> this could be solved using an additional MMU index, however I'm not
>>> overly familiar with the memory and softmmu subsystems.
>> 
>> No, it can't be solved with an MMU index.
>> 
>>> Can anyone point me in the right direction as to what would be the best
>>> way to implement this feature within QEMU?
>> 
>> It's definitely tricky.
>> 
>> We definitely need some TLB_FLAGS_MASK bit set so that we're forced through 
>> the
>> memory slow path.  There is no other way to bypass the endianness that we've
>> already encoded from the target instruction.
>> 
>> Given the tlb_set_page_with_attrs interface, I would think that we need a new
>> bit in MemTxAttrs, so that the target/sparc tlb_fill (and subroutines) can 
>> pass
>> along the TTE bit for the given page.
>> 
>> We have an existing problem in softmmu_template.h,
>> 
>>     /* ??? Note that the io helpers always read data in the target
>>        byte ordering.  We should push the LE/BE request down into io.  */
>>     res = glue(io_read, SUFFIX)(env, mmu_idx, index, addr, retaddr);
>>     res = TGT_BE(res);
>> 
>> We do not want to add a third(!) byte swap along the i/o path.  We need to
>> collapse the two that we have already before considering this one.
>> 
>> This probably takes the form of:
>> 
>> (1) Replacing the "int size" argument with "TCGMemOp memop" for
>>       a) io_{read,write}x in accel/tcg/cputlb.c,
>>       b) memory_region_dispatch_{read,write} in memory.c,
>>       c) adjust_endianness in memory.c.
>>     This carries size+sign+endianness down to the next level.
>> 
>> (2) In memory.c, adjust_endianness,
>> 
>>      if (memory_region_wrong_endianness(mr)) {
>> -        switch (size) {
>> +        memop ^= MO_BSWAP;
>> +    }
>> +    if (memop & MO_BSWAP) {
>> 
>>     For extra credit, re-arrange memory_region_wrong_endianness
>>     to something more explicit -- "wrong" isn't helpful.
>
>Finally I've had a bit of spare time to experiment with this approach,
>and from what I can see there are currently 2 issues:
>
>
>1) Using TCGMemOp in memory.c means it is no longer accelerator agnostic
>
>For the moment I've defined a separate MemOp in memory.h and provided a
>mapping function in io_{read,write}x to map from TCGMemOp to MemOp and
>then pass that into memory_region_dispatch_{read,write}.
>
>Other than not referencing TCGMemOp in the memory API, another reason
>for doing this was that I wasn't convinced that all the MO_ attributes
>were valid outside of TCG. I do, of course, strongly defer to other
>people's knowledge in this area though.
>
>
>2) The above changes to adjust_endianness() fail when
>memory_region_dispatch_{read,write} are called recursively
>
>Whilst booting qemu-system-sparc64 I see that
>memory_region_dispatch_{read,write} get called recursively - once via
>io_{read,write}x and then again via flatview_read_continue() in exec.c.
>
>The net effect of this is that we perform the bswap correctly at the
>tail of the recursion, but then as we travel back up the stack we hit
>memory_region_dispatch_{read,write} once again causing a second bswap
>which means the value is returned with the incorrect endian again.
>
>
>My understanding from your softmmu_template.h comment above is that the
>memory API should do the endian swapping internally allowing the removal
>of the final TGT_BE/TGT_LE applied to the result, or did I get this wrong?
>
>> (3) In tlb_set_page_with_attrs, notice attrs.byte_swap and set
>>     a new TLB_FORCE_SLOW bit within TLB_FLAGS_MASK.
>> 
>> (4) In io_{read,write}x, if iotlbentry->attrs.byte_swap is set,
>>     then memop ^= MO_BSWAP.

Thanks for v1 feedback.

v2:
- Moved size+sign+endianness attributes from TCGMemOp into MemOp.
  In v1 TCGMemOp was re-purposed entirely into MemOp.
- Replaced MemOp MO_{8|16|32|64} with TCGMemOp MO_{UB|UW|UL|UQ} alias.
  This is to avoid warnings on comparing and coercing different enums.
- Renamed get_memop to get_tcgmemop for clarity.
- MEMOP is now SIZE_MEMOP, which is just ctzl(size). 
- Split patch 3/4 so one memory_region_dispatch_{read|write} interface
  is converted per patch.
- Do not reuse TLB_RECHECK, use new TLB_FORCE_SLOW instead.
- Split patch 4/4 so adding the MemTxAddrs parameters and converting
  tlb_set_page() to tlb_set_page_with_attrs() is separate from usage.
- CC'd maintainers.

Tony Nguyen (20):
  tcg: Replace MO_8 with MO_UB alias
  tcg: Replace MO_16 with MO_UW alias
  tcg: Replace MO_32 with MO_UL alias
  tcg: Replace MO_64 with MO_UQ alias
  tcg: Move size+sign+endian from TCGMemOp to MemOp
  tcg: Rename get_memop to get_tcgmemop
  memory: Access MemoryRegion with MemOp
  target/mips: Access MemoryRegion with MemOp
  hw/s390x: Access MemoryRegion with MemOp
  hw/intc/armv7m_nic: Access MemoryRegion with MemOp
  hw/virtio: Access MemoryRegion with MemOp
  hw/vfio: Access MemoryRegion with MemOp
  exec: Access MemoryRegion with MemOp
  cputlb: Access MemoryRegion with MemOp
  memory: Access MemoryRegion with MemOp semantics
  memory: Single byte swap along the I/O path
  cpu: TLB_FLAGS_MASK bit to force memory slow path
  cputlb: Byte swap memory transaction attribute
  target/sparc: Add TLB entry with attributes
  target/sparc: sun4u Invert Endian TTE bit

 MAINTAINERS                         |   1 +
 accel/tcg/cputlb.c                  |  75 +++---
 exec.c                              |   6 +-
 hw/intc/armv7m_nvic.c               |  12 +-
 hw/s390x/s390-pci-inst.c            |   8 +-
 hw/vfio/pci-quirks.c                |   5 +-
 hw/virtio/virtio-pci.c              |   7 +-
 include/exec/cpu-all.h              |  10 +-
 include/exec/memattrs.h             |   2 +
 include/exec/memop.h                |  30 +++
 include/exec/memory.h               |   9 +-
 memory.c                            |  37 +--
 memory_ldst.inc.c                   |  18 +-
 target/arm/sve_helper.c             |  16 +-
 target/arm/translate-a64.c          | 516 ++++++++++++++++++------------------
 target/arm/translate-sve.c          |  74 +++---
 target/arm/translate-vfp.inc.c      |  10 +-
 target/arm/translate.c              | 134 +++++-----
 target/i386/translate.c             | 506 +++++++++++++++++------------------
 target/mips/op_helper.c             |   5 +-
 target/mips/translate.c             |   8 +-
 target/ppc/translate.c              |  28 +-
 target/ppc/translate/fp-impl.inc.c  |   4 +-
 target/ppc/translate/vmx-impl.inc.c | 118 ++++-----
 target/ppc/translate/vsx-impl.inc.c |  22 +-
 target/s390x/translate.c            |  10 +-
 target/s390x/translate_vx.inc.c     |  12 +-
 target/s390x/vec.h                  |  16 +-
 target/sparc/cpu.h                  |   2 +
 target/sparc/mmu_helper.c           |  40 +--
 target/sparc/translate.c            |   4 +-
 tcg/aarch64/tcg-target.inc.c        |  82 +++---
 tcg/arm/tcg-target.inc.c            |  38 +--
 tcg/i386/tcg-target.inc.c           | 176 ++++++------
 tcg/mips/tcg-target.inc.c           |  38 +--
 tcg/optimize.c                      |   2 +-
 tcg/ppc/tcg-target.inc.c            |  28 +-
 tcg/riscv/tcg-target.inc.c          |  26 +-
 tcg/s390/tcg-target.inc.c           |  18 +-
 tcg/sparc/tcg-target.inc.c          |  18 +-
 tcg/tcg-op-gvec.c                   | 342 ++++++++++++------------
 tcg/tcg-op-vec.c                    |  30 +--
 tcg/tcg-op.c                        |  66 ++---
 tcg/tcg.c                           |   2 +-
 tcg/tcg.h                           |  34 ++-
 tcg/tci.c                           |   8 +-
 46 files changed, 1365 insertions(+), 1288 deletions(-)
 create mode 100644 include/exec/memop.h

-- 
1.8.3.1

^ permalink raw reply	[flat|nested] 120+ messages in thread