← Home

CPP RCTX: Cache Prefetch Prediction Restriction by Context

Purpose

Cache Prefetch Prediction Restriction by Context applies to all Cache Allocation Resources that predict cache allocations based on information gathered within the target execution context or contexts.

The actions of code in the target execution context or contexts appearing in program order before the instruction cannot exploitatively control cache prefetch predictions occurring after the instruction is complete and synchronized.

This instruction applies to all:

This instruction is guaranteed to be complete following a DSB that covers both read and write behavior on the same PE as executed the original restriction instruction, and a subsequent context synchronization event is required to ensure that the effect of the completion of the instructions is synchronized to the current execution.

Note

This instruction does not require the invalidation of Cache Allocation Resources so long as the behavior described for completion of this instruction is met by the implementation.

On some implementations the instruction is likely to take a significant number of cycles to execute. This instruction is expected to be used very rarely, such as on the roll-over of an ASID or VMID, but should not be used on every context switch.

Configuration

This instruction is present only when FEAT_SPECRES is implemented. Otherwise, direct accesses to CPP RCTX are UNDEFINED.

Attributes

CPP RCTX is a 64-bit System instruction.

Field descriptions

6362616059585756555453525150494847464544434241403938373635343332
313029282726252423222120191817161514131211109876543210
RES0GVMIDVMID
RES0NSENSELRES0GASIDASID

Bits [63:49]

Reserved, RES0.

GVMID, bit [48]

Execution of this instruction applies to all VMIDs or a specified VMID.

GVMIDMeaning
0b0

Applies to specified VMID for an EL0 or EL1 target execution context.

0b1

Applies to all VMIDs for an EL0 or EL1 target execution context.

For target execution contexts other than EL0 and EL1, this field is RES0.

If the instruction is executed at EL0 or EL1, this field has an Effective value of 0.

If EL2 is not implemented or not enabled for the target Security state, this field is RES0.

VMID, bits [47:32]

Only applies when bit[48] is 0 and the target execution context is either:

Otherwise this field is RES0.

When the instruction is executed at EL1, this field is treated as the current VMID.

When the instruction is executed at EL0 and the Effective value of HCR_EL2.{E2H, TGE} is not {1, 1}, this field is treated as the current VMID.

When the instruction is executed at EL0 and the Effective value of HCR_EL2.{E2H, TGE} is {1, 1}, this field is ignored.

If EL2 is not implemented or not enabled for the target Security state, this field is RES0.

If the implementation supports 16 bits of VMID, then the upper 8 bits of the VMID must be written to 0 by software when the context being affected only uses 8 bits.

Bits [31:28]

Reserved, RES0.

NSE, bit [27]

When FEAT_RME is implemented:

Together with the NS field, selects the Security state.

For a description of the values derived by evaluating NS and NSE together, see CPP_RCTX.NS.



Otherwise:

Reserved, RES0.

NS, bit [26]

When FEAT_RME is implemented:

Together with the NSE field, selects the Security state. Defined values are:

NSENSMeaning
0b00b0When Secure state is implemented, Secure. Otherwise reserved.
0b00b1Non-secure.
0b10b0Root.
0b10b1Realm.

Some Effective values are determined by the current Security state:

This instruction is treated as a NOP when executed at EL3 and either:



Otherwise:

Security State. Defined values are:

NSMeaning
0b0

Secure state.

0b1

Non-secure state.

When executed in Non-secure state, the Effective value of NS is 1.

EL, bits [25:24]

Exception Level. Indicates the Exception level of the target execution context.

ELMeaning
0b00

EL0.

0b01

EL1.

0b10

EL2.

0b11

EL3.

If the instruction is executed at an Exception level lower than the specified level, or is specified to apply to a combination of Exception level and Security state that is not implemented, this instruction is treated as a NOP.

Bits [23:17]

Reserved, RES0.

GASID, bit [16]

Execution of this instruction applies to all ASIDs or a specified ASID.

GASIDMeaning
0b0

Applies to specified ASID for an EL0 target execution context.

0b1

Applies to all ASIDs for an EL0 target execution context.

For target execution contexts other than EL0, this field is RES0.

If the instruction is executed at EL0, this field has an Effective value of 0.

ASID, bits [15:0]

Only applies for an EL0 target execution context and when bit[16] is 0.

Otherwise, this field is RES0.

When the instruction is executed at EL0, this field is treated as the current ASID.

If the implementation supports 16 bits of ASID, then the upper 8 bits of the ASID must be written to 0 by software when the context being affected only uses 8 bits.

Executing CPP RCTX

Accesses to this instruction use the following encodings in the System instruction encoding space:

CPP RCTX, <Xt>

op0op1CRnCRmop2
0b010b0110b01110b00110b111

if PSTATE.EL == EL0 then if !ELIsInHost(EL0) && SCTLR_EL1.EnRCTX == '0' then if EL2Enabled() && HCR_EL2.TGE == '1' then AArch64.SystemAccessTrap(EL2, 0x18); else AArch64.SystemAccessTrap(EL1, 0x18); elsif EL2Enabled() && !ELIsInHost(EL0) && IsFeatureImplemented(FEAT_FGT) && (!HaveEL(EL3) || SCR_EL3.FGTEn == '1') && HFGITR_EL2.CPPRCTX == '1' then AArch64.SystemAccessTrap(EL2, 0x18); elsif ELIsInHost(EL0) && SCTLR_EL2.EnRCTX == '0' then AArch64.SystemAccessTrap(EL2, 0x18); else AArch64.RestrictPrediction(X[t, 64], RestrictType_CachePrefetch); elsif PSTATE.EL == EL1 then if EffectiveHCR_EL2_NVx() IN {'xx1'} then AArch64.SystemAccessTrap(EL2, 0x18); elsif EL2Enabled() && IsFeatureImplemented(FEAT_FGT) && (!HaveEL(EL3) || SCR_EL3.FGTEn == '1') && HFGITR_EL2.CPPRCTX == '1' then AArch64.SystemAccessTrap(EL2, 0x18); else AArch64.RestrictPrediction(X[t, 64], RestrictType_CachePrefetch); elsif PSTATE.EL == EL2 then AArch64.RestrictPrediction(X[t, 64], RestrictType_CachePrefetch); elsif PSTATE.EL == EL3 then AArch64.RestrictPrediction(X[t, 64], RestrictType_CachePrefetch);