[GH-ISSUE #23] Initial plan for cpu soft-virtualization

kerem commented

2026-02-27 21:04:07 +03:00

Owner

Originally created by @skmp on GitHub (Jul 11, 2023).
Original GitHub issue: https://github.com/shadps4-emu/shadPS4/issues/23

Overview

The idea is to soft-virtualize, by reserving a few registers for the host system, and leaving everything else as is with static register allocation. RIP-relocations also need to be done.

Decode x86 instructions (core, rex, mmx/sse, and vex encodings)
Use a register as context and some scratch registers
Can use similar branch lookup structures as hex-emu does
RIP-relative accesses will have to be relocated
fs and gs segments need to be virtualized

Implementing using zydis

r14 as temp
r15 as context
free more temp regs as needed
RIP REL as double deref with rip moves

Kind of patches

RIP_REL
+r
MOD_r
MOD_m
SIB_b
SIB_i

x2 for rex versions

vex encodings

Notes

Windows TLS doesn't give direct access to TLS slots in the same way linux TLS does, so we'll definitely need a context register.

Original writeup

#### Vasika - Elafria eikoniki mixani pou doulevei se leitourgia xristi gia x86/64 arxitetoniki ipodoxis - Vasismeno stous metapinakes apo to hex - Gennitiki arxitekoniki me metaprogramatismo - Eikonefsi ton kataxoriton me statiki antistixisi gia tous perisoterous, mono oti spanizei se dinamiki katanomi - Eikonefsi diefthiseon ektelesis meson paromion domon antistixisis me to hex

Epipleon

Eikonefsi entolon SSE4a pou isos spanizoun

Ektimiseis

~ 5-15K grammes gia ton pirina + metapinakes
me xrisi mono metafraseon gramikis ektelesis, ~ 10% fira sta almata rois
me kapia epipleon poliplokotita kai xrisi diagramaton rois elenxou/ektelesis, < 2% fira?, alla isos kai travlisma kata tin eikonopiisi tis filoksenoumenis arxitektonikis

Xronika

~ 1w gia PoC
~ + 2w gia aploki, gramiki ektelesi
~ + 4w gia DRE me polapla simia eisodou

Pithano sxedio ergasion

Aploiki ilopiisi me ton fortoti ksotikon tou hex + metapinakes + ola apla
Anaparagontopiisi meta gia artiotita kai aisthitiki kodika
Sindesi me ton fortoti ksotikon gia to PS4

Sxolia

@georgemoralis ti les, pos s fenete?
hex-emu gia reference

Originally created by @skmp on GitHub (Jul 11, 2023). Original GitHub issue: https://github.com/shadps4-emu/shadPS4/issues/23 ### Overview The idea is to soft-virtualize, by reserving a few registers for the host system, and leaving everything else as is with static register allocation. RIP-relocations also need to be done. - Decode x86 instructions (core, rex, mmx/sse, and vex encodings) - Use a register as context and some scratch registers - Can use similar branch lookup structures as hex-emu does - RIP-relative accesses will have to be relocated - fs and gs segments need to be virtualized #### Implementing using zydis - r14 as temp - r15 as context - free more temp regs as needed - RIP REL as double deref with rip moves #### Kind of patches - RIP_REL - +r - MOD_r - MOD_m - SIB_b - SIB_i x2 for rex versions + vex encodings #### Notes Windows TLS doesn't give direct access to TLS slots in the same way linux TLS does, so we'll definitely need a context register. <details> <summary> Original writeup </summary> #### Vasika - Elafria eikoniki mixani pou doulevei se leitourgia xristi gia x86/64 arxitetoniki ipodoxis - Vasismeno stous metapinakes apo to hex - Gennitiki arxitekoniki me metaprogramatismo - Eikonefsi ton kataxoriton me statiki antistixisi gia tous perisoterous, mono oti spanizei se dinamiki katanomi - Eikonefsi diefthiseon ektelesis meson paromion domon antistixisis me to hex #### Epipleon - Eikonefsi entolon SSE4a pou isos spanizoun #### Ektimiseis - ~ 5-15K grammes gia ton pirina + metapinakes - me xrisi mono metafraseon gramikis ektelesis, ~ 10% fira sta almata rois - me kapia epipleon poliplokotita kai xrisi diagramaton rois elenxou/ektelesis, < 2% fira?, alla isos kai travlisma kata tin eikonopiisi tis filoksenoumenis arxitektonikis #### Xronika - ~ 1w gia PoC - ~ + 2w gia aploki, gramiki ektelesi - ~ + 4w gia DRE me polapla simia eisodou ### Pithano sxedio ergasion - [ ] Aploiki ilopiisi me ton fortoti ksotikon tou hex + metapinakes + ola apla - [ ] Anaparagontopiisi meta gia artiotita kai aisthitiki kodika - [ ] Sindesi me ton fortoti ksotikon gia to PS4 ### Sxolia - @georgemoralis ti les, pos s fenete? - [hex-emu](https://gitlab.com/hex-emu/hex-emu) gia reference </details>

kerem closed this issue

2026-02-27 21:04:07 +03:00

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@georgemoralis commented on GitHub (Jul 11, 2023):

καλό φαινεται προχωρα :D

@georgemoralis commented on GitHub (Jul 11, 2023): καλό φαινεται προχωρα :D

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@wheremyfoodat commented on GitHub (Jul 11, 2023):

Skopeuoume na ypostiriksoyme panarxaious epeksergastes typou xoris AVX2? H mono osoys einai pio kainourgioi apo to Jaguar?

@wheremyfoodat commented on GitHub (Jul 11, 2023): Skopeuoume na ypostiriksoyme panarxaious epeksergastes typou xoris AVX2? H mono osoys einai pio kainourgioi apo to Jaguar?

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@georgemoralis commented on GitHub (Jul 11, 2023):

Δεν νομιζω οτι έχει AVX2 μεχρι AVX φτανει

@georgemoralis commented on GitHub (Jul 11, 2023): Δεν νομιζω οτι έχει AVX2 μεχρι AVX φτανει

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@georgemoralis commented on GitHub (Jul 11, 2023):

MMX instructions
SSE / Streaming SIMD Extensions
SSE2 / Streaming SIMD Extensions 2
SSE3 / Streaming SIMD Extensions 3
SSSE3 / Supplemental Streaming SIMD Extensions 3
SSE4a ?
SSE4 / SSE4.1 + SSE4.2 / Streaming SIMD Extensions 4 ?
AES / Advanced Encryption Standard instructions
AVX / Advanced Vector Extensions
BMI1 / Bit Manipulation instructions 1
F16C / 16-bit Floating-Point conversion instructions
AMD64 / AMD 64-bit technology ?
AMD-V / AMD Virtualization technology

@georgemoralis commented on GitHub (Jul 11, 2023): MMX instructions SSE / Streaming SIMD Extensions SSE2 / Streaming SIMD Extensions 2 SSE3 / Streaming SIMD Extensions 3 SSSE3 / Supplemental Streaming SIMD Extensions 3 SSE4a [ ? ](https://www.cpu-world.com/Glossary/S/SSE4a.html) SSE4 / SSE4.1 + SSE4.2 / Streaming SIMD Extensions 4 [ ? ](https://www.cpu-world.com/Glossary/S/SSE4.html) AES / Advanced Encryption Standard instructions AVX / Advanced Vector Extensions BMI1 / Bit Manipulation instructions 1 F16C / 16-bit Floating-Point conversion instructions AMD64 / AMD 64-bit technology [ ? ](https://www.cpu-world.com/Glossary/A/AMD64_technology.html) AMD-V / AMD Virtualization technology

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@georgemoralis commented on GitHub (Jul 11, 2023):

οποτε στόχος επεξεργαστής? τι λιγοτερο μπορουμε να παρουμε...

@georgemoralis commented on GitHub (Jul 11, 2023): οποτε στόχος επεξεργαστής? τι λιγοτερο μπορουμε να παρουμε...

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@wheremyfoodat commented on GitHub (Jul 11, 2023):

Βαλε το xbyak να τσεκάρει οτι εχεις AVX και F16C. Αυτό πρέπει να καλύπτει ολα τα Intel chips από Ivy Bridge και μετα.

Τo AMD-V δεν μας ενδιαφέρει γιατί αποκλείεται να δεις userspace code να φτιαχνει VM.

@wheremyfoodat commented on GitHub (Jul 11, 2023): Βαλε το xbyak να τσεκάρει οτι εχεις AVX και F16C. Αυτό πρέπει να καλύπτει ολα τα Intel chips από Ivy Bridge και μετα. Τo AMD-V δεν μας ενδιαφέρει γιατί αποκλείεται να δεις userspace code να φτιαχνει VM.

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@wheremyfoodat commented on GitHub (Jul 11, 2023):

Λεω Xbyak επειδη 99% θα καταληξουμε να το εχουμε καπου στο repo και εχει functions που κανουν CPUID τυπου github.com/wheremyfoodat/Panda3DS@27cf8d75ca/include/PICA/dynapica/shader_rec_emitter_x64.hpp (L114-L125)

@wheremyfoodat commented on GitHub (Jul 11, 2023): Λεω Xbyak επειδη 99% θα καταληξουμε να το εχουμε καπου στο repo και εχει functions που κανουν CPUID τυπου https://github.com/wheremyfoodat/Panda3DS/blob/27cf8d75caf3a685e56a5e88d314b679025c1154/include/PICA/dynapica/shader_rec_emitter_x64.hpp#L114-L125

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@georgemoralis commented on GitHub (Jul 11, 2023):

ναι οπως ελεγα στον @skmp χρειαζεται για να patcharo καποια pragmata οπως το TLS

@georgemoralis commented on GitHub (Jul 11, 2023): ναι οπως ελεγα στον @skmp χρειαζεται για να patcharo καποια pragmata οπως το TLS

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@wheremyfoodat commented on GitHub (Jul 11, 2023):

TLS == thread-local storage?

@wheremyfoodat commented on GitHub (Jul 11, 2023): TLS == thread-local storage?

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@georgemoralis commented on GitHub (Jul 11, 2023):

ya

@georgemoralis commented on GitHub (Jul 11, 2023): ya

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@skmp commented on GitHub (Jul 11, 2023):

Skopeuoume na ypostiriksoyme panarxaious epeksergastes typou xoris AVX2? H mono osoys einai pio kainourgioi apo to Jaguar?

afto thelei pio plires emulation, opote dunno. me to backend to hex, isos, kathos k arm64 for sure.

@skmp commented on GitHub (Jul 11, 2023): > Skopeuoume na ypostiriksoyme panarxaious epeksergastes typou xoris AVX2? H mono osoys einai pio kainourgioi apo to Jaguar? afto thelei pio plires emulation, opote dunno. me to backend to hex, isos, kathos k arm64 for sure.

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@skmp commented on GitHub (Jul 11, 2023):

(updated ticket with hex reference)

@wheremyfoodat https://www.youtube.com/watch?v=ZYN-VqnvWmI gia perf numbers, an kai den exei pesei kapio optimisation effort

@skmp commented on GitHub (Jul 11, 2023): (updated ticket with hex reference) @wheremyfoodat https://www.youtube.com/watch?v=ZYN-VqnvWmI gia perf numbers, an kai den exei pesei kapio optimisation effort

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@georgemoralis commented on GitHub (Jul 12, 2023):

Skopeuoume na ypostiriksoyme panarxaious epeksergastes typou xoris AVX2? H mono osoys einai pio kainourgioi apo to Jaguar?

afto thelei pio plires emulation, opote dunno. me to backend to hex, isos, kathos k arm64 for sure.

ps4 se arm64? gamato!

@georgemoralis commented on GitHub (Jul 12, 2023): > > Skopeuoume na ypostiriksoyme panarxaious epeksergastes typou xoris AVX2? H mono osoys einai pio kainourgioi apo to Jaguar? > > afto thelei pio plires emulation, opote dunno. me to backend to hex, isos, kathos k arm64 for sure. ps4 se arm64? gamato!

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@skmp commented on GitHub (Jul 12, 2023):

me vulkan renderer standard pezei k android :3

@skmp commented on GitHub (Jul 12, 2023): me vulkan renderer standard pezei k android :3

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@georgemoralis commented on GitHub (Jul 12, 2023):

vulkan einai to plano

@georgemoralis commented on GitHub (Jul 12, 2023): vulkan einai to plano

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@skmp commented on GitHub (Aug 9, 2023):

arxiki idea ~ https://github.com/skmp/vex86/blob/main/main.cpp.

Prepei episeis relocation ta rip relative sections, ektos apo ta branches.

@skmp commented on GitHub (Aug 9, 2023): arxiki idea ~ https://github.com/skmp/vex86/blob/main/main.cpp. Prepei episeis relocation ta rip relative sections, ektos apo ta branches.

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@skmp commented on GitHub (Aug 9, 2023):

to opio distixos thelei ena temp hrfm :|

@skmp commented on GitHub (Aug 9, 2023): to opio distixos thelei ena temp hrfm :|

kerem commented

2026-02-27 21:04:08 +03:00

Author

Owner

@skmp commented on GitHub (Aug 9, 2023):