mirrors/qemu
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
734a659ad2
qemu/target/arm/translate-m-nocp.c

789 lines
24 KiB
C
Raw Normal View History

target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
/*
* ARM translation: M-profile NOCP special-case instructions
*
* Copyright (c) 2020 Linaro, Ltd.
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "qemu/osdep.h"
#include "tcg/tcg-op.h"
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
#include "tcg/tcg-op-gvec.h"
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
#include "translate.h"
#include "translate-a32.h"
#include "decode-m-nocp.c.inc"
/*
* Decode VLLDM and VLSTM are nonstandard because:
* * if there is no FPU then these insns must NOP in
* Secure state and UNDEF in Nonsecure state
* * if there is an FPU then these insns do not have
* the usual behaviour that vfp_access_check() provides of
* being controlled by CPACR/NSACR enable bits or the
* lazy-stacking logic.
*/
static bool trans_VLLDM_VLSTM(DisasContext *s, arg_VLLDM_VLSTM *a)
{
TCGv_i32 fptr;
if (!arm_dc_feature(s, ARM_FEATURE_M) ||
!arm_dc_feature(s, ARM_FEATURE_V8)) {
return false;
}
if (a->op) {
/*
* T2 encoding ({D0-D31} reglist): v8.1M and up. We choose not
* to take the IMPDEF option to make memory accesses to the stack
* slots that correspond to the D16-D31 registers (discarding
* read data and writing UNKNOWN values), so for us the T2
* encoding behaves identically to the T1 encoding.
*/
if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
return false;
}
} else {
/*
* T1 encoding ({D0-D15} reglist); undef if we have 32 Dregs.
* This is currently architecturally impossible, but we add the
* check to stay in line with the pseudocode. Note that we must
* emit code for the UNDEF so it takes precedence over the NOCP.
*/
if (dc_isar_feature(aa32_simd_r32, s)) {
unallocated_encoding(s);
return true;
}
}
/*
* If not secure, UNDEF. We must emit code for this
* rather than returning false so that this takes
* precedence over the m-nocp.decode NOCP fallback.
*/
if (!s->v8m_secure) {
unallocated_encoding(s);
return true;
}
target/arm: Add handling for PSR.ECI/ICI On A-profile, PSR bits [15:10][26:25] are always the IT state bits. On M-profile, some of the reserved encodings of the IT state are used to instead indicate partial progress through instructions that were interrupted partway through by an exception and can be resumed. These resumable instructions fall into two categories: (1) load/store multiple instructions, where these bits are called "ICI" and specify the register in the ldm/stm list where execution should resume. (Specifically: LDM, STM, VLDM, VSTM, VLLDM, VLSTM, CLRM, VSCCLRM.) (2) MVE instructions subject to beatwise execution, where these bits are called "ECI" and specify which beats in this and possibly also the following MVE insn have been executed. There are also a few insns (LE, LETP, and BKPT) which do not use the ICI/ECI bits but must leave them alone. Otherwise, we should raise an INVSTATE UsageFault for any attempt to execute an insn with non-zero ICI/ECI bits. So far we have been able to ignore ECI/ICI, because the architecture allows the IMPDEF choice of "always restart load/store multiple from the beginning regardless of ICI state", so the only thing we have been missing is that we don't raise the INVSTATE fault for bad guest code. However, MVE requires that we honour ECI bits and do not rexecute beats of an insn that have already been executed. Add the support in the decoder for handling ECI/ICI: * identify the ECI/ICI case in the CONDEXEC TB flags * when a load/store multiple insn succeeds, it updates the ECI/ICI state (both in DisasContext and in the CPU state), and sets a flag to say that the ECI/ICI state was handled * if we find that the insn we just decoded did not handle the ECI/ICI state, we delete all the code that we just generated for it and instead emit the code to raise the INVFAULT. This allows us to avoid having to update every non-MVE non-LDM/STM insn to make it check for "is ECI/ICI set?". We continue with our existing IMPDEF choice of not caring about the ICI state for the load/store multiples and simply restarting them from the beginning. Because we don't allow interrupts in the middle of an insn, the only way we would see this state is if the guest set ICI manually on return from an exception handler, so it's a corner case which doesn't merit optimisation. ICI update for LDM/STM is simple -- it always zeroes the state. ECI update for MVE beatwise insns will be a little more complex, since the ECI state may include information for the following insn. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210614151007.4545-5-peter.maydell@linaro.org
2021-06-14 18:09:14 +03:00
s->eci_handled = true;
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
/* If no fpu, NOP. */
if (!dc_isar_feature(aa32_vfp, s)) {
target/arm: Add handling for PSR.ECI/ICI On A-profile, PSR bits [15:10][26:25] are always the IT state bits. On M-profile, some of the reserved encodings of the IT state are used to instead indicate partial progress through instructions that were interrupted partway through by an exception and can be resumed. These resumable instructions fall into two categories: (1) load/store multiple instructions, where these bits are called "ICI" and specify the register in the ldm/stm list where execution should resume. (Specifically: LDM, STM, VLDM, VSTM, VLLDM, VLSTM, CLRM, VSCCLRM.) (2) MVE instructions subject to beatwise execution, where these bits are called "ECI" and specify which beats in this and possibly also the following MVE insn have been executed. There are also a few insns (LE, LETP, and BKPT) which do not use the ICI/ECI bits but must leave them alone. Otherwise, we should raise an INVSTATE UsageFault for any attempt to execute an insn with non-zero ICI/ECI bits. So far we have been able to ignore ECI/ICI, because the architecture allows the IMPDEF choice of "always restart load/store multiple from the beginning regardless of ICI state", so the only thing we have been missing is that we don't raise the INVSTATE fault for bad guest code. However, MVE requires that we honour ECI bits and do not rexecute beats of an insn that have already been executed. Add the support in the decoder for handling ECI/ICI: * identify the ECI/ICI case in the CONDEXEC TB flags * when a load/store multiple insn succeeds, it updates the ECI/ICI state (both in DisasContext and in the CPU state), and sets a flag to say that the ECI/ICI state was handled * if we find that the insn we just decoded did not handle the ECI/ICI state, we delete all the code that we just generated for it and instead emit the code to raise the INVFAULT. This allows us to avoid having to update every non-MVE non-LDM/STM insn to make it check for "is ECI/ICI set?". We continue with our existing IMPDEF choice of not caring about the ICI state for the load/store multiples and simply restarting them from the beginning. Because we don't allow interrupts in the middle of an insn, the only way we would see this state is if the guest set ICI manually on return from an exception handler, so it's a corner case which doesn't merit optimisation. ICI update for LDM/STM is simple -- it always zeroes the state. ECI update for MVE beatwise insns will be a little more complex, since the ECI state may include information for the following insn. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210614151007.4545-5-peter.maydell@linaro.org
2021-06-14 18:09:14 +03:00
clear_eci_state(s);
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
return true;
}
fptr = load_reg(s, a->rn);
if (a->l) {
gen_helper_v7m_vlldm(cpu_env, fptr);
} else {
gen_helper_v7m_vlstm(cpu_env, fptr);
}
tcg_temp_free_i32(fptr);
target/arm: Add handling for PSR.ECI/ICI On A-profile, PSR bits [15:10][26:25] are always the IT state bits. On M-profile, some of the reserved encodings of the IT state are used to instead indicate partial progress through instructions that were interrupted partway through by an exception and can be resumed. These resumable instructions fall into two categories: (1) load/store multiple instructions, where these bits are called "ICI" and specify the register in the ldm/stm list where execution should resume. (Specifically: LDM, STM, VLDM, VSTM, VLLDM, VLSTM, CLRM, VSCCLRM.) (2) MVE instructions subject to beatwise execution, where these bits are called "ECI" and specify which beats in this and possibly also the following MVE insn have been executed. There are also a few insns (LE, LETP, and BKPT) which do not use the ICI/ECI bits but must leave them alone. Otherwise, we should raise an INVSTATE UsageFault for any attempt to execute an insn with non-zero ICI/ECI bits. So far we have been able to ignore ECI/ICI, because the architecture allows the IMPDEF choice of "always restart load/store multiple from the beginning regardless of ICI state", so the only thing we have been missing is that we don't raise the INVSTATE fault for bad guest code. However, MVE requires that we honour ECI bits and do not rexecute beats of an insn that have already been executed. Add the support in the decoder for handling ECI/ICI: * identify the ECI/ICI case in the CONDEXEC TB flags * when a load/store multiple insn succeeds, it updates the ECI/ICI state (both in DisasContext and in the CPU state), and sets a flag to say that the ECI/ICI state was handled * if we find that the insn we just decoded did not handle the ECI/ICI state, we delete all the code that we just generated for it and instead emit the code to raise the INVFAULT. This allows us to avoid having to update every non-MVE non-LDM/STM insn to make it check for "is ECI/ICI set?". We continue with our existing IMPDEF choice of not caring about the ICI state for the load/store multiples and simply restarting them from the beginning. Because we don't allow interrupts in the middle of an insn, the only way we would see this state is if the guest set ICI manually on return from an exception handler, so it's a corner case which doesn't merit optimisation. ICI update for LDM/STM is simple -- it always zeroes the state. ECI update for MVE beatwise insns will be a little more complex, since the ECI state may include information for the following insn. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210614151007.4545-5-peter.maydell@linaro.org
2021-06-14 18:09:14 +03:00
clear_eci_state(s);
target/arm: Add TB flag for "MVE insns not predicated" Our current codegen for MVE always calls out to helper functions, because some byte lanes might be predicated. The common case is that in fact there is no predication active and all lanes should be updated together, so we can produce better code by detecting that and using the TCG generic vector infrastructure. Add a TB flag that is set when we can guarantee that there is no active MVE predication, and a bool in the DisasContext. Subsequent patches will use this flag to generate improved code for some instructions. In most cases when the predication state changes we simply end the TB after that instruction. For the code called from vfp_access_check() that handles lazy state preservation and creating a new FP context, we can usually avoid having to try to end the TB because luckily the new value of the flag following the register changes in those sequences doesn't depend on any runtime decisions. We do have to end the TB if the guest has enabled lazy FP state preservation but not automatic state preservation, but this is an odd corner case that is not going to be common in real-world code. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210913095440.13462-4-peter.maydell@linaro.org
2021-09-13 12:54:31 +03:00
/*
* End the TB, because we have updated FP control bits,
* and possibly VPR or LTPSIZE.
*/
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
s->base.is_jmp = DISAS_UPDATE_EXIT;
return true;
}
static bool trans_VSCCLRM(DisasContext *s, arg_VSCCLRM *a)
{
int btmreg, topreg;
TCGv_i64 zero;
TCGv_i32 aspen, sfpa;
if (!dc_isar_feature(aa32_m_sec_state, s)) {
/* Before v8.1M, fall through in decode to NOCP check */
return false;
}
/* Explicitly UNDEF because this takes precedence over NOCP */
if (!arm_dc_feature(s, ARM_FEATURE_M_MAIN) || !s->v8m_secure) {
unallocated_encoding(s);
return true;
}
target/arm: Add handling for PSR.ECI/ICI On A-profile, PSR bits [15:10][26:25] are always the IT state bits. On M-profile, some of the reserved encodings of the IT state are used to instead indicate partial progress through instructions that were interrupted partway through by an exception and can be resumed. These resumable instructions fall into two categories: (1) load/store multiple instructions, where these bits are called "ICI" and specify the register in the ldm/stm list where execution should resume. (Specifically: LDM, STM, VLDM, VSTM, VLLDM, VLSTM, CLRM, VSCCLRM.) (2) MVE instructions subject to beatwise execution, where these bits are called "ECI" and specify which beats in this and possibly also the following MVE insn have been executed. There are also a few insns (LE, LETP, and BKPT) which do not use the ICI/ECI bits but must leave them alone. Otherwise, we should raise an INVSTATE UsageFault for any attempt to execute an insn with non-zero ICI/ECI bits. So far we have been able to ignore ECI/ICI, because the architecture allows the IMPDEF choice of "always restart load/store multiple from the beginning regardless of ICI state", so the only thing we have been missing is that we don't raise the INVSTATE fault for bad guest code. However, MVE requires that we honour ECI bits and do not rexecute beats of an insn that have already been executed. Add the support in the decoder for handling ECI/ICI: * identify the ECI/ICI case in the CONDEXEC TB flags * when a load/store multiple insn succeeds, it updates the ECI/ICI state (both in DisasContext and in the CPU state), and sets a flag to say that the ECI/ICI state was handled * if we find that the insn we just decoded did not handle the ECI/ICI state, we delete all the code that we just generated for it and instead emit the code to raise the INVFAULT. This allows us to avoid having to update every non-MVE non-LDM/STM insn to make it check for "is ECI/ICI set?". We continue with our existing IMPDEF choice of not caring about the ICI state for the load/store multiples and simply restarting them from the beginning. Because we don't allow interrupts in the middle of an insn, the only way we would see this state is if the guest set ICI manually on return from an exception handler, so it's a corner case which doesn't merit optimisation. ICI update for LDM/STM is simple -- it always zeroes the state. ECI update for MVE beatwise insns will be a little more complex, since the ECI state may include information for the following insn. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210614151007.4545-5-peter.maydell@linaro.org
2021-06-14 18:09:14 +03:00
s->eci_handled = true;
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
if (!dc_isar_feature(aa32_vfp_simd, s)) {
/* NOP if we have neither FP nor MVE */
target/arm: Add handling for PSR.ECI/ICI On A-profile, PSR bits [15:10][26:25] are always the IT state bits. On M-profile, some of the reserved encodings of the IT state are used to instead indicate partial progress through instructions that were interrupted partway through by an exception and can be resumed. These resumable instructions fall into two categories: (1) load/store multiple instructions, where these bits are called "ICI" and specify the register in the ldm/stm list where execution should resume. (Specifically: LDM, STM, VLDM, VSTM, VLLDM, VLSTM, CLRM, VSCCLRM.) (2) MVE instructions subject to beatwise execution, where these bits are called "ECI" and specify which beats in this and possibly also the following MVE insn have been executed. There are also a few insns (LE, LETP, and BKPT) which do not use the ICI/ECI bits but must leave them alone. Otherwise, we should raise an INVSTATE UsageFault for any attempt to execute an insn with non-zero ICI/ECI bits. So far we have been able to ignore ECI/ICI, because the architecture allows the IMPDEF choice of "always restart load/store multiple from the beginning regardless of ICI state", so the only thing we have been missing is that we don't raise the INVSTATE fault for bad guest code. However, MVE requires that we honour ECI bits and do not rexecute beats of an insn that have already been executed. Add the support in the decoder for handling ECI/ICI: * identify the ECI/ICI case in the CONDEXEC TB flags * when a load/store multiple insn succeeds, it updates the ECI/ICI state (both in DisasContext and in the CPU state), and sets a flag to say that the ECI/ICI state was handled * if we find that the insn we just decoded did not handle the ECI/ICI state, we delete all the code that we just generated for it and instead emit the code to raise the INVFAULT. This allows us to avoid having to update every non-MVE non-LDM/STM insn to make it check for "is ECI/ICI set?". We continue with our existing IMPDEF choice of not caring about the ICI state for the load/store multiples and simply restarting them from the beginning. Because we don't allow interrupts in the middle of an insn, the only way we would see this state is if the guest set ICI manually on return from an exception handler, so it's a corner case which doesn't merit optimisation. ICI update for LDM/STM is simple -- it always zeroes the state. ECI update for MVE beatwise insns will be a little more complex, since the ECI state may include information for the following insn. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210614151007.4545-5-peter.maydell@linaro.org
2021-06-14 18:09:14 +03:00
clear_eci_state(s);
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
return true;
}
/*
* If FPCCR.ASPEN != 0 && CONTROL_S.SFPA == 0 then there is no
* active floating point context so we must NOP (without doing
* any lazy state preservation or the NOCP check).
*/
aspen = load_cpu_field(v7m.fpccr[M_REG_S]);
sfpa = load_cpu_field(v7m.control[M_REG_S]);
tcg_gen_andi_i32(aspen, aspen, R_V7M_FPCCR_ASPEN_MASK);
tcg_gen_xori_i32(aspen, aspen, R_V7M_FPCCR_ASPEN_MASK);
tcg_gen_andi_i32(sfpa, sfpa, R_V7M_CONTROL_SFPA_MASK);
tcg_gen_or_i32(sfpa, sfpa, aspen);
arm_gen_condlabel(s);
tcg_gen_brcondi_i32(TCG_COND_EQ, sfpa, 0, s->condlabel);
if (s->fp_excp_el != 0) {
target/arm: Rename gen_exception_insn to gen_exception_insn_el Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20220609202901.1177572-12-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2022-06-10 16:32:32 +03:00
gen_exception_insn_el(s, s->pc_curr, EXCP_NOCP,
syn_uncategorized(), s->fp_excp_el);
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
return true;
}
topreg = a->vd + a->imm - 1;
btmreg = a->vd;
/* Convert to Sreg numbers if the insn specified in Dregs */
if (a->size == 3) {
topreg = topreg * 2 + 1;
btmreg *= 2;
}
if (topreg > 63 || (topreg > 31 && !(topreg & 1))) {
/* UNPREDICTABLE: we choose to undef */
unallocated_encoding(s);
return true;
}
/* Silently ignore requests to clear D16-D31 if they don't exist */
if (topreg > 31 && !dc_isar_feature(aa32_simd_r32, s)) {
topreg = 31;
}
if (!vfp_access_check(s)) {
return true;
}
/* Zero the Sregs from btmreg to topreg inclusive. */
target/arm: Use tcg_constant in translate-m-nocp.c Use tcg_constant_{i32,i64} as appropriate throughout. This fixes a bug in trans_VSCCLRM() where we were leaking a TCGv. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2022-04-17 20:43:44 +03:00
zero = tcg_constant_i64(0);
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
if (btmreg & 1) {
write_neon_element64(zero, btmreg >> 1, 1, MO_32);
btmreg++;
}
for (; btmreg + 1 <= topreg; btmreg += 2) {
write_neon_element64(zero, btmreg >> 1, 0, MO_64);
}
if (btmreg == topreg) {
write_neon_element64(zero, btmreg >> 1, 0, MO_32);
btmreg++;
}
assert(btmreg == topreg + 1);
target/arm: Handle VPR semantics in existing code When MVE is supported, the VPR register has a place on the exception stack frame in a previously reserved slot just above the FPSCR. It must also be zeroed in various situations when we invalidate FPU context. Update the code which handles the stack frames (exception entry and exit code, VLLDM, and VLSTM) to save/restore VPR. Update code which invalidates FP registers (mostly also exception entry and exit code, but also VSCCLRM and the code in full_vfp_access_check() that corresponds to the ExecuteFPCheck() pseudocode) to zero VPR. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210614151007.4545-4-peter.maydell@linaro.org
2021-06-14 18:09:13 +03:00
if (dc_isar_feature(aa32_mve, s)) {
target/arm: Use tcg_constant in translate-m-nocp.c Use tcg_constant_{i32,i64} as appropriate throughout. This fixes a bug in trans_VSCCLRM() where we were leaking a TCGv. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2022-04-17 20:43:44 +03:00
store_cpu_field(tcg_constant_i32(0), v7m.vpr);
target/arm: Handle VPR semantics in existing code When MVE is supported, the VPR register has a place on the exception stack frame in a previously reserved slot just above the FPSCR. It must also be zeroed in various situations when we invalidate FPU context. Update the code which handles the stack frames (exception entry and exit code, VLLDM, and VLSTM) to save/restore VPR. Update code which invalidates FP registers (mostly also exception entry and exit code, but also VSCCLRM and the code in full_vfp_access_check() that corresponds to the ExecuteFPCheck() pseudocode) to zero VPR. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210614151007.4545-4-peter.maydell@linaro.org
2021-06-14 18:09:13 +03:00
}
target/arm: Add handling for PSR.ECI/ICI On A-profile, PSR bits [15:10][26:25] are always the IT state bits. On M-profile, some of the reserved encodings of the IT state are used to instead indicate partial progress through instructions that were interrupted partway through by an exception and can be resumed. These resumable instructions fall into two categories: (1) load/store multiple instructions, where these bits are called "ICI" and specify the register in the ldm/stm list where execution should resume. (Specifically: LDM, STM, VLDM, VSTM, VLLDM, VLSTM, CLRM, VSCCLRM.) (2) MVE instructions subject to beatwise execution, where these bits are called "ECI" and specify which beats in this and possibly also the following MVE insn have been executed. There are also a few insns (LE, LETP, and BKPT) which do not use the ICI/ECI bits but must leave them alone. Otherwise, we should raise an INVSTATE UsageFault for any attempt to execute an insn with non-zero ICI/ECI bits. So far we have been able to ignore ECI/ICI, because the architecture allows the IMPDEF choice of "always restart load/store multiple from the beginning regardless of ICI state", so the only thing we have been missing is that we don't raise the INVSTATE fault for bad guest code. However, MVE requires that we honour ECI bits and do not rexecute beats of an insn that have already been executed. Add the support in the decoder for handling ECI/ICI: * identify the ECI/ICI case in the CONDEXEC TB flags * when a load/store multiple insn succeeds, it updates the ECI/ICI state (both in DisasContext and in the CPU state), and sets a flag to say that the ECI/ICI state was handled * if we find that the insn we just decoded did not handle the ECI/ICI state, we delete all the code that we just generated for it and instead emit the code to raise the INVFAULT. This allows us to avoid having to update every non-MVE non-LDM/STM insn to make it check for "is ECI/ICI set?". We continue with our existing IMPDEF choice of not caring about the ICI state for the load/store multiples and simply restarting them from the beginning. Because we don't allow interrupts in the middle of an insn, the only way we would see this state is if the guest set ICI manually on return from an exception handler, so it's a corner case which doesn't merit optimisation. ICI update for LDM/STM is simple -- it always zeroes the state. ECI update for MVE beatwise insns will be a little more complex, since the ECI state may include information for the following insn. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210614151007.4545-5-peter.maydell@linaro.org
2021-06-14 18:09:14 +03:00
clear_eci_state(s);
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
return true;
}
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
/*
* M-profile provides two different sets of instructions that can
* access floating point system registers: VMSR/VMRS (which move
* to/from a general purpose register) and VLDR/VSTR sysreg (which
* move directly to/from memory). In some cases there are also side
* effects which must happen after any write to memory (which could
* cause an exception). So we implement the common logic for the
* sysreg access in gen_M_fp_sysreg_write() and gen_M_fp_sysreg_read(),
* which take pointers to callback functions which will perform the
* actual "read/write general purpose register" and "read/write
* memory" operations.
*/
/*
* Emit code to store the sysreg to its final destination; frees the
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
* TCG temp 'value' it is passed. do_access is true to do the store,
* and false to skip it and only perform side-effects like base
* register writeback.
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
*/
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
typedef void fp_sysreg_storefn(DisasContext *s, void *opaque, TCGv_i32 value,
bool do_access);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
/*
* Emit code to load the value to be copied to the sysreg; returns
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
* a new TCG temporary. do_access is true to do the store,
* and false to skip it and only perform side-effects like base
* register writeback.
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
*/
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
typedef TCGv_i32 fp_sysreg_loadfn(DisasContext *s, void *opaque,
bool do_access);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
/* Common decode/access checks for fp sysreg read/write */
typedef enum FPSysRegCheckResult {
FPSysRegCheckFailed, /* caller should return false */
FPSysRegCheckDone, /* caller should return true */
FPSysRegCheckContinue, /* caller should continue generating code */
} FPSysRegCheckResult;
static FPSysRegCheckResult fp_sysreg_checks(DisasContext *s, int regno)
{
if (!dc_isar_feature(aa32_fpsp_v2, s) && !dc_isar_feature(aa32_mve, s)) {
return FPSysRegCheckFailed;
}
switch (regno) {
case ARM_VFP_FPSCR:
case QEMU_VFP_FPSCR_NZCV:
break;
case ARM_VFP_FPSCR_NZCVQC:
if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
return FPSysRegCheckFailed;
}
break;
case ARM_VFP_FPCXT_S:
case ARM_VFP_FPCXT_NS:
if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
return FPSysRegCheckFailed;
}
if (!s->v8m_secure) {
return FPSysRegCheckFailed;
}
break;
case ARM_VFP_VPR:
case ARM_VFP_P0:
if (!dc_isar_feature(aa32_mve, s)) {
return FPSysRegCheckFailed;
}
break;
default:
return FPSysRegCheckFailed;
}
/*
* FPCXT_NS is a special case: it has specific handling for
* "current FP state is inactive", and must do the PreserveFPState()
* but not the usual full set of actions done by ExecuteFPCheck().
* So we don't call vfp_access_check() and the callers must handle this.
*/
if (regno != ARM_VFP_FPCXT_NS && !vfp_access_check(s)) {
return FPSysRegCheckDone;
}
return FPSysRegCheckContinue;
}
static void gen_branch_fpInactive(DisasContext *s, TCGCond cond,
TCGLabel *label)
{
/*
* FPCXT_NS is a special case: it has specific handling for
* "current FP state is inactive", and must do the PreserveFPState()
* but not the usual full set of actions done by ExecuteFPCheck().
* We don't have a TB flag that matches the fpInactive check, so we
* do it at runtime as we don't expect FPCXT_NS accesses to be frequent.
*
* Emit code that checks fpInactive and does a conditional
* branch to label based on it:
* if cond is TCG_COND_NE then branch if fpInactive != 0 (ie if inactive)
* if cond is TCG_COND_EQ then branch if fpInactive == 0 (ie if active)
*/
assert(cond == TCG_COND_EQ || cond == TCG_COND_NE);
/* fpInactive = FPCCR_NS.ASPEN == 1 && CONTROL.FPCA == 0 */
TCGv_i32 aspen, fpca;
aspen = load_cpu_field(v7m.fpccr[M_REG_NS]);
fpca = load_cpu_field(v7m.control[M_REG_S]);
tcg_gen_andi_i32(aspen, aspen, R_V7M_FPCCR_ASPEN_MASK);
tcg_gen_xori_i32(aspen, aspen, R_V7M_FPCCR_ASPEN_MASK);
tcg_gen_andi_i32(fpca, fpca, R_V7M_CONTROL_FPCA_MASK);
tcg_gen_or_i32(fpca, fpca, aspen);
tcg_gen_brcondi_i32(tcg_invert_cond(cond), fpca, 0, label);
tcg_temp_free_i32(aspen);
tcg_temp_free_i32(fpca);
}
static bool gen_M_fp_sysreg_write(DisasContext *s, int regno,
fp_sysreg_loadfn *loadfn,
void *opaque)
{
/* Do a write to an M-profile floating point system register */
TCGv_i32 tmp;
TCGLabel *lab_end = NULL;
switch (fp_sysreg_checks(s, regno)) {
case FPSysRegCheckFailed:
return false;
case FPSysRegCheckDone:
return true;
case FPSysRegCheckContinue:
break;
}
switch (regno) {
case ARM_VFP_FPSCR:
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
tmp = loadfn(s, opaque, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
gen_helper_vfp_set_fpscr(cpu_env, tmp);
tcg_temp_free_i32(tmp);
gen_lookup_tb(s);
break;
case ARM_VFP_FPSCR_NZCVQC:
{
TCGv_i32 fpscr;
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
tmp = loadfn(s, opaque, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
if (dc_isar_feature(aa32_mve, s)) {
/* QC is only present for MVE; otherwise RES0 */
TCGv_i32 qc = tcg_temp_new_i32();
tcg_gen_andi_i32(qc, tmp, FPCR_QC);
/*
* The 4 vfp.qc[] fields need only be "zero" vs "non-zero";
* here writing the same value into all elements is simplest.
*/
tcg_gen_gvec_dup_i32(MO_32, offsetof(CPUARMState, vfp.qc),
16, 16, qc);
}
tcg_gen_andi_i32(tmp, tmp, FPCR_NZCV_MASK);
fpscr = load_cpu_field(vfp.xregs[ARM_VFP_FPSCR]);
tcg_gen_andi_i32(fpscr, fpscr, ~FPCR_NZCV_MASK);
tcg_gen_or_i32(fpscr, fpscr, tmp);
store_cpu_field(fpscr, vfp.xregs[ARM_VFP_FPSCR]);
tcg_temp_free_i32(tmp);
break;
}
case ARM_VFP_FPCXT_NS:
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
{
TCGLabel *lab_active = gen_new_label();
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
lab_end = gen_new_label();
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
gen_branch_fpInactive(s, TCG_COND_EQ, lab_active);
/*
* fpInactive case: write is a NOP, so only do side effects
* like register writeback before we branch to end
*/
loadfn(s, opaque, false);
tcg_gen_br(lab_end);
gen_set_label(lab_active);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
/*
* !fpInactive: if FPU disabled, take NOCP exception;
* otherwise PreserveFPState(), and then FPCXT_NS writes
* behave the same as FPCXT_S writes.
*/
target/arm: Handle FPU check for FPCXT_NS insns via vfp_access_check_m() Instead of open-coding the "take NOCP exception if FPU disabled, otherwise call gen_preserve_fp_state()" code in the accessors for FPCXT_NS, add an argument to vfp_access_check_m() which tells it to skip the gen_update_fp_context() call, so we can use it for the FPCXT_NS case. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-8-peter.maydell@linaro.org
2021-06-18 17:10:19 +03:00
if (!vfp_access_check_m(s, true)) {
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
/*
* This was only a conditional exception, so override
target/arm: Rename gen_exception_insn to gen_exception_insn_el Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20220609202901.1177572-12-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2022-06-10 16:32:32 +03:00
* gen_exception_insn_el()'s default to DISAS_NORETURN
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
*/
s->base.is_jmp = DISAS_NEXT;
break;
}
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
}
/* fall through */
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
case ARM_VFP_FPCXT_S:
{
TCGv_i32 sfpa, control;
/*
* Set FPSCR and CONTROL.SFPA from value; the new FPSCR takes
* bits [27:0] from value and zeroes bits [31:28].
*/
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
tmp = loadfn(s, opaque, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
sfpa = tcg_temp_new_i32();
tcg_gen_shri_i32(sfpa, tmp, 31);
control = load_cpu_field(v7m.control[M_REG_S]);
tcg_gen_deposit_i32(control, control, sfpa,
R_V7M_CONTROL_SFPA_SHIFT, 1);
store_cpu_field(control, v7m.control[M_REG_S]);
tcg_gen_andi_i32(tmp, tmp, ~FPCR_NZCV_MASK);
gen_helper_vfp_set_fpscr(cpu_env, tmp);
target/arm: Add TB flag for "MVE insns not predicated" Our current codegen for MVE always calls out to helper functions, because some byte lanes might be predicated. The common case is that in fact there is no predication active and all lanes should be updated together, so we can produce better code by detecting that and using the TCG generic vector infrastructure. Add a TB flag that is set when we can guarantee that there is no active MVE predication, and a bool in the DisasContext. Subsequent patches will use this flag to generate improved code for some instructions. In most cases when the predication state changes we simply end the TB after that instruction. For the code called from vfp_access_check() that handles lazy state preservation and creating a new FP context, we can usually avoid having to try to end the TB because luckily the new value of the flag following the register changes in those sequences doesn't depend on any runtime decisions. We do have to end the TB if the guest has enabled lazy FP state preservation but not automatic state preservation, but this is an odd corner case that is not going to be common in real-world code. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210913095440.13462-4-peter.maydell@linaro.org
2021-09-13 12:54:31 +03:00
s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
tcg_temp_free_i32(tmp);
tcg_temp_free_i32(sfpa);
break;
}
case ARM_VFP_VPR:
/* Behaves as NOP if not privileged */
if (IS_USER(s)) {
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
loadfn(s, opaque, false);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
break;
}
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
tmp = loadfn(s, opaque, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
store_cpu_field(tmp, v7m.vpr);
target/arm: Add TB flag for "MVE insns not predicated" Our current codegen for MVE always calls out to helper functions, because some byte lanes might be predicated. The common case is that in fact there is no predication active and all lanes should be updated together, so we can produce better code by detecting that and using the TCG generic vector infrastructure. Add a TB flag that is set when we can guarantee that there is no active MVE predication, and a bool in the DisasContext. Subsequent patches will use this flag to generate improved code for some instructions. In most cases when the predication state changes we simply end the TB after that instruction. For the code called from vfp_access_check() that handles lazy state preservation and creating a new FP context, we can usually avoid having to try to end the TB because luckily the new value of the flag following the register changes in those sequences doesn't depend on any runtime decisions. We do have to end the TB if the guest has enabled lazy FP state preservation but not automatic state preservation, but this is an odd corner case that is not going to be common in real-world code. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210913095440.13462-4-peter.maydell@linaro.org
2021-09-13 12:54:31 +03:00
s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
break;
case ARM_VFP_P0:
{
TCGv_i32 vpr;
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
tmp = loadfn(s, opaque, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
vpr = load_cpu_field(v7m.vpr);
tcg_gen_deposit_i32(vpr, vpr, tmp,
R_V7M_VPR_P0_SHIFT, R_V7M_VPR_P0_LENGTH);
store_cpu_field(vpr, v7m.vpr);
target/arm: Add TB flag for "MVE insns not predicated" Our current codegen for MVE always calls out to helper functions, because some byte lanes might be predicated. The common case is that in fact there is no predication active and all lanes should be updated together, so we can produce better code by detecting that and using the TCG generic vector infrastructure. Add a TB flag that is set when we can guarantee that there is no active MVE predication, and a bool in the DisasContext. Subsequent patches will use this flag to generate improved code for some instructions. In most cases when the predication state changes we simply end the TB after that instruction. For the code called from vfp_access_check() that handles lazy state preservation and creating a new FP context, we can usually avoid having to try to end the TB because luckily the new value of the flag following the register changes in those sequences doesn't depend on any runtime decisions. We do have to end the TB if the guest has enabled lazy FP state preservation but not automatic state preservation, but this is an odd corner case that is not going to be common in real-world code. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210913095440.13462-4-peter.maydell@linaro.org
2021-09-13 12:54:31 +03:00
s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
tcg_temp_free_i32(tmp);
break;
}
default:
g_assert_not_reached();
}
if (lab_end) {
gen_set_label(lab_end);
}
return true;
}
static bool gen_M_fp_sysreg_read(DisasContext *s, int regno,
fp_sysreg_storefn *storefn,
void *opaque)
{
/* Do a read from an M-profile floating point system register */
TCGv_i32 tmp;
TCGLabel *lab_end = NULL;
bool lookup_tb = false;
switch (fp_sysreg_checks(s, regno)) {
case FPSysRegCheckFailed:
return false;
case FPSysRegCheckDone:
return true;
case FPSysRegCheckContinue:
break;
}
if (regno == ARM_VFP_FPSCR_NZCVQC && !dc_isar_feature(aa32_mve, s)) {
/* QC is RES0 without MVE, so NZCVQC simplifies to NZCV */
regno = QEMU_VFP_FPSCR_NZCV;
}
switch (regno) {
case ARM_VFP_FPSCR:
tmp = tcg_temp_new_i32();
gen_helper_vfp_get_fpscr(tmp, cpu_env);
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
storefn(s, opaque, tmp, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
break;
case ARM_VFP_FPSCR_NZCVQC:
tmp = tcg_temp_new_i32();
gen_helper_vfp_get_fpscr(tmp, cpu_env);
tcg_gen_andi_i32(tmp, tmp, FPCR_NZCVQC_MASK);
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
storefn(s, opaque, tmp, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
break;
case QEMU_VFP_FPSCR_NZCV:
/*
* Read just NZCV; this is a special case to avoid the
* helper call for the "VMRS to CPSR.NZCV" insn.
*/
tmp = load_cpu_field(vfp.xregs[ARM_VFP_FPSCR]);
tcg_gen_andi_i32(tmp, tmp, FPCR_NZCV_MASK);
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
storefn(s, opaque, tmp, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
break;
case ARM_VFP_FPCXT_S:
{
TCGv_i32 control, sfpa, fpscr;
/* Bits [27:0] from FPSCR, bit [31] from CONTROL.SFPA */
tmp = tcg_temp_new_i32();
sfpa = tcg_temp_new_i32();
gen_helper_vfp_get_fpscr(tmp, cpu_env);
tcg_gen_andi_i32(tmp, tmp, ~FPCR_NZCV_MASK);
control = load_cpu_field(v7m.control[M_REG_S]);
tcg_gen_andi_i32(sfpa, control, R_V7M_CONTROL_SFPA_MASK);
tcg_gen_shli_i32(sfpa, sfpa, 31 - R_V7M_CONTROL_SFPA_SHIFT);
tcg_gen_or_i32(tmp, tmp, sfpa);
tcg_temp_free_i32(sfpa);
/*
* Store result before updating FPSCR etc, in case
* it is a memory write which causes an exception.
*/
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
storefn(s, opaque, tmp, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
/*
* Now we must reset FPSCR from FPDSCR_NS, and clear
* CONTROL.SFPA; so we'll end the TB here.
*/
tcg_gen_andi_i32(control, control, ~R_V7M_CONTROL_SFPA_MASK);
store_cpu_field(control, v7m.control[M_REG_S]);
fpscr = load_cpu_field(v7m.fpdscr[M_REG_NS]);
gen_helper_vfp_set_fpscr(cpu_env, fpscr);
tcg_temp_free_i32(fpscr);
lookup_tb = true;
break;
}
case ARM_VFP_FPCXT_NS:
{
target/arm: Use tcg_constant in translate-m-nocp.c Use tcg_constant_{i32,i64} as appropriate throughout. This fixes a bug in trans_VSCCLRM() where we were leaking a TCGv. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2022-04-17 20:43:44 +03:00
TCGv_i32 control, sfpa, fpscr, fpdscr;
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
TCGLabel *lab_active = gen_new_label();
lookup_tb = true;
gen_branch_fpInactive(s, TCG_COND_EQ, lab_active);
/* fpInactive case: reads as FPDSCR_NS */
TCGv_i32 tmp = load_cpu_field(v7m.fpdscr[M_REG_NS]);
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
storefn(s, opaque, tmp, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
lab_end = gen_new_label();
tcg_gen_br(lab_end);
gen_set_label(lab_active);
/*
* !fpInactive: if FPU disabled, take NOCP exception;
* otherwise PreserveFPState(), and then FPCXT_NS
* reads the same as FPCXT_S.
*/
target/arm: Handle FPU check for FPCXT_NS insns via vfp_access_check_m() Instead of open-coding the "take NOCP exception if FPU disabled, otherwise call gen_preserve_fp_state()" code in the accessors for FPCXT_NS, add an argument to vfp_access_check_m() which tells it to skip the gen_update_fp_context() call, so we can use it for the FPCXT_NS case. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-8-peter.maydell@linaro.org
2021-06-18 17:10:19 +03:00
if (!vfp_access_check_m(s, true)) {
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
/*
* This was only a conditional exception, so override
target/arm: Rename gen_exception_insn to gen_exception_insn_el Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20220609202901.1177572-12-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2022-06-10 16:32:32 +03:00
* gen_exception_insn_el()'s default to DISAS_NORETURN
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
*/
s->base.is_jmp = DISAS_NEXT;
break;
}
tmp = tcg_temp_new_i32();
sfpa = tcg_temp_new_i32();
fpscr = tcg_temp_new_i32();
gen_helper_vfp_get_fpscr(fpscr, cpu_env);
tcg_gen_andi_i32(tmp, fpscr, ~FPCR_NZCV_MASK);
control = load_cpu_field(v7m.control[M_REG_S]);
tcg_gen_andi_i32(sfpa, control, R_V7M_CONTROL_SFPA_MASK);
tcg_gen_shli_i32(sfpa, sfpa, 31 - R_V7M_CONTROL_SFPA_SHIFT);
tcg_gen_or_i32(tmp, tmp, sfpa);
tcg_temp_free_i32(control);
/* Store result before updating FPSCR, in case it faults */
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
storefn(s, opaque, tmp, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
/* If SFPA is zero then set FPSCR from FPDSCR_NS */
fpdscr = load_cpu_field(v7m.fpdscr[M_REG_NS]);
target/arm: Use tcg_constant in translate-m-nocp.c Use tcg_constant_{i32,i64} as appropriate throughout. This fixes a bug in trans_VSCCLRM() where we were leaking a TCGv. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2022-04-17 20:43:44 +03:00
tcg_gen_movcond_i32(TCG_COND_EQ, fpscr, sfpa, tcg_constant_i32(0),
fpdscr, fpscr);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
gen_helper_vfp_set_fpscr(cpu_env, fpscr);
tcg_temp_free_i32(sfpa);
tcg_temp_free_i32(fpdscr);
tcg_temp_free_i32(fpscr);
break;
}
case ARM_VFP_VPR:
/* Behaves as NOP if not privileged */
if (IS_USER(s)) {
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
storefn(s, opaque, NULL, false);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
break;
}
tmp = load_cpu_field(v7m.vpr);
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
storefn(s, opaque, tmp, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
break;
case ARM_VFP_P0:
tmp = load_cpu_field(v7m.vpr);
tcg_gen_extract_i32(tmp, tmp, R_V7M_VPR_P0_SHIFT, R_V7M_VPR_P0_LENGTH);
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
storefn(s, opaque, tmp, true);
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
break;
default:
g_assert_not_reached();
}
if (lab_end) {
gen_set_label(lab_end);
}
if (lookup_tb) {
gen_lookup_tb(s);
}
return true;
}
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
static void fp_sysreg_to_gpr(DisasContext *s, void *opaque, TCGv_i32 value,
bool do_access)
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
{
arg_VMSR_VMRS *a = opaque;
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
if (!do_access) {
return;
}
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
if (a->rt == 15) {
/* Set the 4 flag bits in the CPSR */
gen_set_nzcv(value);
tcg_temp_free_i32(value);
} else {
store_reg(s, a->rt, value);
}
}
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
static TCGv_i32 gpr_to_fp_sysreg(DisasContext *s, void *opaque, bool do_access)
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
{
arg_VMSR_VMRS *a = opaque;
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
if (!do_access) {
return NULL;
}
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
return load_reg(s, a->rt);
}
static bool trans_VMSR_VMRS(DisasContext *s, arg_VMSR_VMRS *a)
{
/*
* Accesses to R15 are UNPREDICTABLE; we choose to undef.
* FPSCR -> r15 is a special case which writes to the PSR flags;
* set a->reg to a special value to tell gen_M_fp_sysreg_read()
* we only care about the top 4 bits of FPSCR there.
*/
if (a->rt == 15) {
if (a->l && a->reg == ARM_VFP_FPSCR) {
a->reg = QEMU_VFP_FPSCR_NZCV;
} else {
return false;
}
}
if (a->l) {
/* VMRS, move FP system register to gp register */
return gen_M_fp_sysreg_read(s, a->reg, fp_sysreg_to_gpr, a);
} else {
/* VMSR, move gp register to FP system register */
return gen_M_fp_sysreg_write(s, a->reg, gpr_to_fp_sysreg, a);
}
}
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
static void fp_sysreg_to_memory(DisasContext *s, void *opaque, TCGv_i32 value,
bool do_access)
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
{
arg_vldr_sysreg *a = opaque;
uint32_t offset = a->imm;
TCGv_i32 addr;
if (!a->a) {
offset = -offset;
}
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
if (!do_access && !a->w) {
return;
}
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
addr = load_reg(s, a->rn);
if (a->p) {
tcg_gen_addi_i32(addr, addr, offset);
}
if (s->v8m_stackcheck && a->rn == 13 && a->w) {
gen_helper_v8m_stackcheck(cpu_env, addr);
}
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
if (do_access) {
gen_aa32_st_i32(s, value, addr, get_mem_index(s),
MO_UL | MO_ALIGN | s->be_data);
tcg_temp_free_i32(value);
}
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
if (a->w) {
/* writeback */
if (!a->p) {
tcg_gen_addi_i32(addr, addr, offset);
}
store_reg(s, a->rn, addr);
} else {
tcg_temp_free_i32(addr);
}
}
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
static TCGv_i32 memory_to_fp_sysreg(DisasContext *s, void *opaque,
bool do_access)
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
{
arg_vldr_sysreg *a = opaque;
uint32_t offset = a->imm;
TCGv_i32 addr;
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
TCGv_i32 value = NULL;
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
if (!a->a) {
offset = -offset;
}
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
if (!do_access && !a->w) {
return NULL;
}
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
addr = load_reg(s, a->rn);
if (a->p) {
tcg_gen_addi_i32(addr, addr, offset);
}
if (s->v8m_stackcheck && a->rn == 13 && a->w) {
gen_helper_v8m_stackcheck(cpu_env, addr);
}
target/arm: Handle writeback in VLDR/VSTR sysreg with no memory access A few subcases of VLDR/VSTR sysreg succeed but do not perform a memory access: * VSTR of VPR when unprivileged * VLDR to VPR when unprivileged * VLDR to FPCXT_NS when fpInactive In these cases, even though we don't do the memory access we should still update the base register and perform the stack limit check if the insn's addressing mode specifies writeback. Our implementation failed to do this, because we handle these side-effects inside the memory_to_fp_sysreg() and fp_sysreg_to_memory() callback functions, which are only called if there's something to load or store. Fix this by adding an extra argument to the callbacks which is set to true to actually perform the access and false to only do side effects like writeback, and calling the callback with do_access = false for the three cases listed above. This produces slightly suboptimal code for the case of a write to FPCXT_NS when the FPU is inactive and the insn didn't have side effects (ie no writeback, or via VMSR), in which case we'll generate a conditional branch over an unconditional branch. But this doesn't seem to be important enough to merit requiring the callback to report back whether it generated any code or not. Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-5-peter.maydell@linaro.org
2021-06-18 17:10:16 +03:00
if (do_access) {
value = tcg_temp_new_i32();
gen_aa32_ld_i32(s, value, addr, get_mem_index(s),
MO_UL | MO_ALIGN | s->be_data);
}
target/arm: Don't NOCP fault for FPCXT_NS accesses The M-profile architecture requires that accesses to FPCXT_NS when there is no active FP state must not take a NOCP fault even if the FPU is disabled. We were not implementing this correctly, because in our decode we catch the NOCP faults early in m-nocp.decode. Fix this bug by moving all the handling of M-profile FP system register accesses from vfp.decode into m-nocp.decode and putting it above the NOCP blocks. This provides the correct behaviour: * for accesses other than FPCXT_NS the trans functions call vfp_access_check(), which will check for FPU disabled and raise a NOCP exception if necessary * for FPCXT_NS we have the special case code that doesn't call vfp_access_check() * when these trans functions want to raise an UNDEF they return false, so the decoder will fall through into the NOCP blocks. This means that NOCP correctly takes precedence over UNDEF for these insns. (This is a difference from the other insns handled by m-nocp.decode, where UNDEF takes precedence and which we implement by having those trans functions call unallocated_encoding() in the appropriate places.) [Note for backport to stable: this commit has a semantic dependency on commit 9a486856e9173af, which was not marked as cc-stable because we didn't know we'd need it for a for-stable bugfix.] Cc: qemu-stable@nongnu.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210618141019.10671-4-peter.maydell@linaro.org
2021-06-18 17:10:15 +03:00
if (a->w) {
/* writeback */
if (!a->p) {
tcg_gen_addi_i32(addr, addr, offset);
}
store_reg(s, a->rn, addr);
} else {
tcg_temp_free_i32(addr);
}
return value;
}
static bool trans_VLDR_sysreg(DisasContext *s, arg_vldr_sysreg *a)
{
if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
return false;
}
if (a->rn == 15) {
return false;
}
return gen_M_fp_sysreg_write(s, a->reg, memory_to_fp_sysreg, a);
}
static bool trans_VSTR_sysreg(DisasContext *s, arg_vldr_sysreg *a)
{
if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
return false;
}
if (a->rn == 15) {
return false;
}
return gen_M_fp_sysreg_read(s, a->reg, fp_sysreg_to_memory, a);
}
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
static bool trans_NOCP(DisasContext *s, arg_nocp *a)
{
/*
* Handle M-profile early check for disabled coprocessor:
* all we need to do here is emit the NOCP exception if
* the coprocessor is disabled. Otherwise we return false
* and the real VFP/etc decode will handle the insn.
*/
assert(arm_dc_feature(s, ARM_FEATURE_M));
if (a->cp == 11) {
a->cp = 10;
}
if (arm_dc_feature(s, ARM_FEATURE_V8_1M) &&
(a->cp == 8 || a->cp == 9 || a->cp == 14 || a->cp == 15)) {
/* in v8.1M cp 8, 9, 14, 15 also are governed by the cp10 enable */
a->cp = 10;
}
if (a->cp != 10) {
target/arm: Introduce gen_exception_insn Create a new wrapper function that passes the default exception target to gen_exception_insn_el. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20220609202901.1177572-13-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2022-06-10 16:32:32 +03:00
gen_exception_insn(s, s->pc_curr, EXCP_NOCP, syn_uncategorized());
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
return true;
}
if (s->fp_excp_el != 0) {
target/arm: Rename gen_exception_insn to gen_exception_insn_el Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20220609202901.1177572-12-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2022-06-10 16:32:32 +03:00
gen_exception_insn_el(s, s->pc_curr, EXCP_NOCP,
syn_uncategorized(), s->fp_excp_el);
target/arm: Split m-nocp trans functions into their own file Currently the trans functions for m-nocp.decode all live in translate-vfp.inc.c; move them out into their own translation unit, translate-m-nocp.c. The trans_* functions here are pure code motion with no changes. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20210430132740.10391-5-peter.maydell@linaro.org
2021-04-30 16:27:31 +03:00
return true;
}
return false;
}
static bool trans_NOCP_8_1(DisasContext *s, arg_nocp *a)
{
/* This range needs a coprocessor check for v8.1M and later only */
if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
return false;
}
return trans_NOCP(s, a);
}
Reference in New Issue Copy Permalink