We do not have control in the default 'riscv-aia' default value. We can
try to set it to a specific value, in this case 'auto', but there's no
guarantee that the host will accept it.
Couple with this we're always doing a 'qemu_log' to inform whether we're
ended up using the host default or if we managed to set the AIA mode to
the QEMU default we wanted to set.
Change the 'riscv-aia' description to better reflect how the option
works, and remove the two informative 'qemu_log' that are now unneeded:
if no message shows, riscv-aia was set to the default or uset-set value.
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241028182037.290171-3-dbarboza@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
When failing to set the selected AIA mode, 'aia_mode' is left untouched.
This means that 'aia_mode' will not reflect the actual AIA mode,
retrieved in 'default_aia_mode',
This is benign for now, but it will impact QMP query commands that will
expose the 'aia_mode' value, retrieving the wrong value.
Set 'aia_mode' to 'default_aia_mode' if we fail to change the AIA mode
in KVM.
While we're at it, rework the log/warning messages to be a bit less
verbose. Instead of:
KVM AIA: default mode is emul
qemu-system-riscv64: warning: KVM AIA: failed to set KVM AIA mode
We can use a single warning message:
qemu-system-riscv64: warning: KVM AIA: failed to set KVM AIA mode 'auto', using default host mode 'emul'
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241028182037.290171-2-dbarboza@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
zicfiss has following instructions
- sspopchk: pops a value from shadow stack and compares with x1/x5.
If they dont match, reports a sw check exception with tval = 3.
- sspush: pushes value in x1/x5 on shadow stack
- ssrdp: reads current shadow stack
- ssamoswap: swaps contents of shadow stack atomically
sspopchk/sspush/ssrdp default to zimop if zimop implemented and SSE=0
If SSE=0, ssamoswap is illegal instruction exception.
This patch implements shadow stack operations for qemu-user and shadow
stack is not protected.
Signed-off-by: Deepak Gupta <debug@rivosinc.com>
Co-developed-by: Jim Shu <jim.shu@sifive.com>
Co-developed-by: Andy Chiu <andy.chiu@sifive.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241008225010.1861630-17-debug@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
zicfiss protects shadow stack using new page table encodings PTE.W=1,
PTE.R=0 and PTE.X=0. This encoding is reserved if zicfiss is not
implemented or if shadow stack are not enabled.
Loads on shadow stack memory are allowed while stores to shadow stack
memory leads to access faults. Shadow stack accesses to RO memory
leads to store page fault.
To implement special nature of shadow stack memory where only selected
stores (shadow stack stores from sspush) have to be allowed while rest
of regular stores disallowed, new MMU TLB index is created for shadow
stack.
Furthermore, `check_zicbom_access` (`cbo.clean/flush/inval`) may probe
shadow stack memory and must always raise store/AMO access fault because
it has store semantics. For non-shadow stack memory even though
`cbo.clean/flush/inval` have store semantics, it will not fault if read
is allowed (probably to follow `clflush` on x86). Although if read is not
allowed, eventually `probe_write` will do store page (or access) fault (if
permissions don't allow it). cbo operations on shadow stack memory must
always raise store access fault. Thus extending `get_physical_address` to
recieve `probe` parameter as well.
Signed-off-by: Deepak Gupta <debug@rivosinc.com>
Suggested-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241008225010.1861630-14-debug@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
zicfiss introduces a new state ssp ("shadow stack register") in cpu.
ssp is expressed as a new unprivileged csr (CSR_SSP=0x11) and holds
virtual address for shadow stack as programmed by software.
Shadow stack (for each mode) is enabled via bit3 in *envcfg CSRs.
Shadow stack can be enabled for a mode only if it's higher privileged
mode had it enabled for itself. M mode doesn't need enabling control,
it's always available if extension is available on cpu.
This patch also implements helper bcfi function which determines if bcfi
is enabled at current privilege or not.
Adds ssp to migration state as well.
Signed-off-by: Deepak Gupta <debug@rivosinc.com>
Co-developed-by: Jim Shu <jim.shu@sifive.com>
Co-developed-by: Andy Chiu <andy.chiu@sifive.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241008225010.1861630-12-debug@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Implements setting lp expected when `jalr` is encountered and implements
`lpad` instruction of zicfilp. `lpad` instruction is taken out of
auipc x0, <imm_20>. This is an existing HINTNOP space. If `lpad` is
target of an indirect branch, cpu checks for 20 bit value in x7 upper
with 20 bit value embedded in `lpad`. If they don't match, cpu raises a
sw check exception with tval = 2.
Signed-off-by: Deepak Gupta <debug@rivosinc.com>
Co-developed-by: Jim Shu <jim.shu@sifive.com>
Co-developed-by: Andy Chiu <andy.chiu@sifive.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241008225010.1861630-8-debug@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
zicfilp protects forward control flow (if enabled) by enforcing all
indirect call and jmp must land on a landing pad instruction `lpad`. If
target of an indirect call or jmp is not `lpad` then cpu/hart must raise
a sw check exception with tval = 2.
This patch implements the mechanism using TCG. Target architecture branch
instruction must define the end of a TB. Using this property, during
translation of branch instruction, TB flag = FCFI_LP_EXPECTED can be set.
Translation of target TB can check if FCFI_LP_EXPECTED flag is set and a
flag (fcfi_lp_expected) can be set in DisasContext. If `lpad` gets
translated, fcfi_lp_expected flag in DisasContext can be cleared. Else
it'll fault.
Signed-off-by: Deepak Gupta <debug@rivosinc.com>
Co-developed-by: Jim Shu <jim.shu@sifive.com>
Co-developed-by: Andy Chiu <andy.chiu@sifive.com>
Suggested-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241008225010.1861630-7-debug@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
sw check exception support was recently added. This patch further augments
sw check exception by providing support for additional code which is
provided in *tval. Adds `sw_check_code` field in cpuarchstate. Whenever
sw check exception is raised *tval gets the value deposited in
`sw_check_code`.
Signed-off-by: Deepak Gupta <debug@rivosinc.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241008225010.1861630-6-debug@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
elp state is recorded in *status on trap entry (less privilege to higher
privilege) and restored in elp from *status on trap exit (higher to less
privilege).
Additionally this patch introduces a forward cfi helper function to
determine if current privilege has forward cfi is enabled or not based on
*envcfg (for U, VU, S, VU, HS) or mseccfg csr (for M).
Signed-off-by: Deepak Gupta <debug@rivosinc.com>
Co-developed-by: Jim Shu <jim.shu@sifive.com>
Co-developed-by: Andy Chiu <andy.chiu@sifive.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241008225010.1861630-5-debug@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
zicfilp introduces a new state elp ("expected landing pad") in cpu.
During normal execution, elp is idle (NO_LP_EXPECTED) i.e not expecting
landing pad. On an indirect call, elp moves LP_EXPECTED. When elp is
LP_EXPECTED, only a subsquent landing pad instruction can set state back
to NO_LP_EXPECTED. On reset, elp is set to NO_LP_EXPECTED.
zicfilp is enabled via bit2 in *envcfg CSRs. Enabling control for M-mode
is in mseccfg CSR at bit position 10.
On trap, elp state is saved away in *status.
Adds elp to the migration state as well.
Signed-off-by: Deepak Gupta <debug@rivosinc.com>
Co-developed-by: Jim Shu <jim.shu@sifive.com>
Co-developed-by: Andy Chiu <andy.chiu@sifive.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241008225010.1861630-4-debug@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Execution environment config CSR controlling user env and current
privilege state shouldn't be limited to qemu-system only. *envcfg
CSRs control enabling of features in next lesser mode. In some cases
bits *envcfg CSR can be lit up by kernel as part of kernel policy or
software (user app) can choose to opt-in by issuing a system call
(e.g. prctl). In case of qemu-user, it should be no different because
qemu is providing underlying execution environment facility and thus
either should provide some default value in *envcfg CSRs or react to
system calls (prctls) initiated from application. priv is set to PRV_U
and menvcfg/senvcfg set to 0 for qemu-user on reest.
`henvcfg` has been left for qemu-system only because it is not expected
that someone will use qemu-user where application is expected to have
hypervisor underneath which is controlling its execution environment. If
such a need arises then `henvcfg` could be exposed as well.
Relevant discussion:
https://lore.kernel.org/all/CAKmqyKOTVWPFep2msTQVdUmJErkH+bqCcKEQ4hAnyDFPdWKe0Q@mail.gmail.com/
Signed-off-by: Deepak Gupta <debug@rivosinc.com>
Suggested-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241008225010.1861630-2-debug@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
The RISC-V unprivileged specification "31.3.11. State of Vector
Extension at Reset" has a note that recommends vtype.vill be set on
reset as part of ensuring that the vector extension have a consistent
state at reset.
This change now makes QEMU consistent with Spike which sets vtype.vill
on reset.
Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
Reviewed-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Message-ID: <20240930165258.72258-1-rbradford@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
The register VXSAT should be RW only to the first bit.
The remaining bits should be 0.
The RISC-V Instruction Set Manual Volume I: Unprivileged Architecture
The vxsat CSR has a single read-write least-significant bit (vxsat[0])
that indicates if a fixed-point instruction has had to saturate an output
value to fit into a destination format. Bits vxsat[XLEN-1:1]
should be written as zeros.
Signed-off-by: Evgenii Prokopiev <evgenii.prokopiev@syntacore.com>
Reviewed-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20241002084436.89347-1-evgenii.prokopiev@syntacore.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
If QEMU has been configured with "--without-default-devices", the build
is currently failing with:
/usr/bin/ld: libqemu-riscv32-softmmu.a.p/target_riscv_cpu_helper.c.o:
in function `riscv_cpu_do_interrupt':
.../qemu/target/riscv/cpu_helper.c:1678:(.text+0x2214): undefined
reference to `do_common_semihosting'
We always want semihosting to be enabled if TCG is available, so change
the "imply" statements in the Kconfig file to "select", and make sure to
avoid calling into do_common_semihosting() if TCG is not available.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20240906094858.718105-1-thuth@redhat.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
C doesn't extend the sign bit for unsigned types since there isn't a
sign bit to extend. This means a promotion of a u32 to a u64 results
in the upper 32 bits of the u64 being zero. If that result is then
used as a mask on another u64 the upper 32 bits will be cleared. rv32
physical addresses may be up to 34 bits wide, so we don't want to
clear the high bits while page aligning the address. The fix is to
use hwaddr for the mask, which, even on rv32, is 64-bits wide.
Fixes: af3fc195e3 ("target/riscv: Change the TLB page size depends on PMP entries.")
Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <20240909083241.43836-2-ajones@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
The Svvptc extension describes a uarch that does not cache invalid TLB
entries: that's the case for qemu so there is nothing particular to
implement other than the introduction of this extension.
Since qemu already exposes Svvptc behaviour, let's enable it by default
since it allows to drastically reduce the number of sfence.vma emitted
by S-mode.
Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20240828083651.203861-1-alexghiti@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
According to RISC-V Debug specification, the optional textra32 and
textra64 trigger CSRs can be used to configure additional matching
conditions for the triggers. For example, if the textra.MHSELECT field
is set to 4 (mcontext), this trigger will only match or fire if the low
bits of mcontext/hcontext equal textra.MHVALUE field.
This commit adds the aforementioned matching condition as common trigger
matching conditions. Currently, the only legal values of textra.MHSELECT
are 0 (ignore) and 4 (mcontext). When textra.MHSELECT is 0, we pass the
checking. When textra.MHSELECT is 4, we compare textra.MHVALUE with
mcontext CSR. The remaining fields, such as textra.SBYTEMASK,
textra.SVALUE, and textra.SSELECT, are hardwired to zero for now. Thus,
we skip checking them here.
Signed-off-by: Alvin Chang <alvinga@andestech.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20240826024657.262553-3-alvinga@andestech.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
This commit allows program to write textra trigger CSR for type 2, 3, 6
triggers. In this preliminary patch, the textra.MHVALUE and the
textra.MHSELECT fields are allowed to be configured. Other fields, such
as textra.SBYTEMASK, textra.SVALUE, and textra.SSELECT, are hardwired to
zero for now.
For textra.MHSELECT field, the only legal values are 0 (ignore) and 4
(mcontext). Writing 1~3 into textra.MHSELECT will be changed to 0, and
writing 5~7 into textra.MHSELECT will be changed to 4. This behavior is
aligned to RISC-V SPIKE simulator.
Signed-off-by: Alvin Chang <alvinga@andestech.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20240826024657.262553-2-alvinga@andestech.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
While the spec doesn't state it, setting timecmp to UINT64_MAX is
another way to stop a timer, as it's considered setting the next
timer event to occur at infinity. And, even if the time CSR does
eventually reach UINT64_MAX, the very next tick will bring it back to
zero, once again less than timecmp. For this reason
riscv_timer_write_timecmp() special cases UINT64_MAX. However, if a
previously set timecmp has not yet expired, then setting timecmp to
UINT64_MAX to disable / stop it would not work, as the special case
left the previous QEMU timer active, which would then still deliver
an interrupt at that previous timecmp time. Ensure the stopped timer
will not still deliver an interrupt by also deleting the QEMU timer
in the UINT64_MAX special case.
Fixes: ae0edf2188 ("target/riscv: No need to re-start QEMU timer when timecmp == UINT64_MAX")
Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20240829084002.1805006-2-ajones@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Just as the hart bit setting of the AIA should be calculated as
ceil(log2(max_hart_id + 1)) the group bit setting should be
calculated as ceil(log2(max_group_id + 1)). The hart bits are
implemented by passing max_hart_id to find_last_bit() and adding
one to the result. Do the same for the group bit setting.
Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Message-ID: <20240821075040.498945-2-ajones@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Gitlab issue [1] reports a misleading error when trying to run a 'rv64'
cpu with 'zfinx' and without 'f':
$ ./build/qemu-system-riscv64 -nographic -M virt -cpu rv64,zfinx=true,f=false
qemu-system-riscv64: Zfinx cannot be supported together with F extension
The user explicitly disabled F and the error message mentions a conflict
with Zfinx and F.
The problem isn't the error reporting, but the logic used when applying
the implied ZFA rule that enables RVF unconditionally, without honoring
user choice (i.e. keep F disabled).
Change cpu_enable_implied_rule() to check if the user deliberately
disabled a MISA bit. In this case we shouldn't either re-enable the bit
nor apply any implied rules related to it.
After this change the error message now shows:
$ ./build/qemu-system-riscv64 -nographic -M virt -cpu rv64,zfinx=true,f=false
qemu-system-riscv64: Zfa extension requires F extension
Disabling 'zfa':
$ ./build/qemu-system-riscv64 -nographic -M virt -cpu rv64,zfinx=true,f=false,zfa=false
qemu-system-riscv64: D extension requires F extension
And finally after disabling 'd':
$ ./build/qemu-system-riscv64 -nographic -M virt -cpu rv64,zfinx=true,f=false,zfa=false,d=false
(OpenSBI boots ...)
[1] https://gitlab.com/qemu-project/qemu/-/issues/2486
Cc: Frank Chang <frank.chang@sifive.com>
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2486
Fixes: 047da861f9 ("target/riscv: Introduce extension implied rule helpers")
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20240824173338.316666-1-dbarboza@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
RVV spec allows implementations to set vl with values within
[ceil(AVL/2),VLMAX] when VLMAX < AVL < 2*VLMAX. This commit adds a
property "rvv_vl_half_avl" to enable setting vl = ceil(AVL/2). This
behavior helps identify compiler issues and bugs.
Signed-off-by: Jason Chien <jason.chien@sifive.com>
Reviewed-by: Frank Chang <frank.chang@sifive.com>
Message-ID: <20240722175004.23666-1-jason.chien@sifive.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Coverity complained about the possible out-of-bounds access with
counter_virt/counter_virt_prev because these two arrays are
accessed with privilege mode. However, these two arrays are accessed
only when virt is enabled. Thus, the privilege mode can't be M mode.
Add the asserts anyways to detect any wrong usage of these arrays
in the future.
Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Atish Patra <atishp@rivosinc.com>
Fixes: Coverity CID 1558459
Fixes: Coverity CID 1558462
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20240724-fixes-v1-1-4a64596b0d64@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
