load/store user fpsimd on Arm64

full context switch: add fpsimd load/store support to container
application.

Signed-off-by: Bin Lu <bin.lu@arm.com>

3 files changed, 94 insertions, 0 deletions

diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go
index fca3a5478..d483ff03c 100644
--- a/pkg/sentry/platform/ring0/kernel_arm64.go
+++ b/pkg/sentry/platform/ring0/kernel_arm64.go
@@ -59,11 +59,13 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) {
 	regs.Pstate &= ^uint64(UserFlagsClear)
 	regs.Pstate |= UserFlagsSet
 
+	LoadFloatingPoint(switchOpts.FloatingPointState)
 	SetTLS(regs.TPIDR_EL0)
 
 	kernelExitToEl0()
 
 	regs.TPIDR_EL0 = GetTLS()
+	SaveFloatingPoint(switchOpts.FloatingPointState)
 
 	vector = c.vecCode
 
diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go
index a6345010d..00e52c8af 100644
--- a/pkg/sentry/platform/ring0/lib_arm64.go
+++ b/pkg/sentry/platform/ring0/lib_arm64.go
@@ -38,6 +38,12 @@ func SaveVRegs(*byte)
 // LoadVRegs loads V0-V31 registers.
 func LoadVRegs(*byte)
 
+// LoadFloatingPoint loads floating point state.
+func LoadFloatingPoint(*byte)
+
+// SaveFloatingPoint saves floating point state.
+func SaveFloatingPoint(*byte)
+
 // GetTLS returns the value of TPIDR_EL0 register.
 func GetTLS() (value uint64)
 
diff --git a/pkg/sentry/platform/ring0/lib_arm64.s b/pkg/sentry/platform/ring0/lib_arm64.s
index b63e14b41..86bfbe46f 100644
--- a/pkg/sentry/platform/ring0/lib_arm64.s
+++ b/pkg/sentry/platform/ring0/lib_arm64.s
@@ -129,3 +129,89 @@ TEXT ·LoadVRegs(SB),NOSPLIT,$0-8
 	ISB $15
 
 	RET
+
+TEXT ·LoadFloatingPoint(SB),NOSPLIT,$0-8
+	MOVD addr+0(FP), R0
+
+	MOVD 0(R0), R1
+	MOVD R1, FPSR
+	MOVD 8(R0), R1
+	MOVD R1, NZCV
+
+	FMOVD 16*1(R0), F0
+	FMOVD 16*2(R0), F1
+	FMOVD 16*3(R0), F2
+	FMOVD 16*4(R0), F3
+	FMOVD 16*5(R0), F4
+	FMOVD 16*6(R0), F5
+	FMOVD 16*7(R0), F6
+	FMOVD 16*8(R0), F7
+	FMOVD 16*9(R0), F8
+	FMOVD 16*10(R0), F9
+	FMOVD 16*11(R0), F10
+	FMOVD 16*12(R0), F11
+	FMOVD 16*13(R0), F12
+	FMOVD 16*14(R0), F13
+	FMOVD 16*15(R0), F14
+	FMOVD 16*16(R0), F15
+	FMOVD 16*17(R0), F16
+	FMOVD 16*18(R0), F17
+	FMOVD 16*19(R0), F18
+	FMOVD 16*20(R0), F19
+	FMOVD 16*21(R0), F20
+	FMOVD 16*22(R0), F21
+	FMOVD 16*23(R0), F22
+	FMOVD 16*24(R0), F23
+	FMOVD 16*25(R0), F24
+	FMOVD 16*26(R0), F25
+	FMOVD 16*27(R0), F26
+	FMOVD 16*28(R0), F27
+	FMOVD 16*29(R0), F28
+	FMOVD 16*30(R0), F29
+	FMOVD 16*31(R0), F30
+	FMOVD 16*32(R0), F31
+
+	RET
+
+TEXT ·SaveFloatingPoint(SB),NOSPLIT,$0-8
+	MOVD addr+0(FP), R0
+
+	MOVD FPSR, R1
+	MOVD R1, 0(R0)
+	MOVD NZCV, R1
+	MOVD R1, 8(R0)
+
+	FMOVD F0, 16*1(R0)
+	FMOVD F1, 16*2(R0)
+	FMOVD F2, 16*3(R0)
+	FMOVD F3, 16*4(R0)
+	FMOVD F4, 16*5(R0)
+	FMOVD F5, 16*6(R0)
+	FMOVD F6, 16*7(R0)
+	FMOVD F7, 16*8(R0)
+	FMOVD F8, 16*9(R0)
+	FMOVD F9, 16*10(R0)
+	FMOVD F10, 16*11(R0)
+	FMOVD F11, 16*12(R0)
+	FMOVD F12, 16*13(R0)
+	FMOVD F13, 16*14(R0)
+	FMOVD F14, 16*15(R0)
+	FMOVD F15, 16*16(R0)
+	FMOVD F16, 16*17(R0)
+	FMOVD F17, 16*18(R0)
+	FMOVD F18, 16*19(R0)
+	FMOVD F19, 16*20(R0)
+	FMOVD F20, 16*21(R0)
+	FMOVD F21, 16*22(R0)
+	FMOVD F22, 16*23(R0)
+	FMOVD F23, 16*24(R0)
+	FMOVD F24, 16*25(R0)
+	FMOVD F25, 16*26(R0)
+	FMOVD F26, 16*27(R0)
+	FMOVD F27, 16*28(R0)
+	FMOVD F28, 16*29(R0)
+	FMOVD F29, 16*30(R0)
+	FMOVD F30, 16*31(R0)
+	FMOVD F31, 16*32(R0)
+
+	RET