Currently, the max satp mode is set with the only constraint that it must be
implemented in QEMU, i.e. set in valid_vm_1_10_[32|64].
But we actually need to add another level of constraint: what the hw is
actually capable of, because currently, a linux booting on a sifive-u54
boots in sv57 mode which is incompatible with the cpu's sv39 max
capability.
So add a new bitmap to RISCVSATPMap which contains this capability and
initialize it in every XXX_cpu_init.
Finally:
- valid_vm_1_10_[32|64] constrains which satp mode the CPU can use
- the CPU hw capabilities constrains what the user may select
- the user's selection then constrains what's available to the guest
OS.
Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Frank Chang <frank.chang@sifive.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-ID: <20230303131252.892893-5-alexghiti@rivosinc.com>
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
RISC-V specifies multiple sizes for addressable memory and Linux probes for
the machine's support at startup via the satp CSR register (done in
csr.c:validate_vm).
As per the specification, sv64 must support sv57, which in turn must
support sv48...etc. So we can restrict machine support by simply setting the
"highest" supported mode and the bare mode is always supported.
You can set the satp mode using the new properties "sv32", "sv39", "sv48",
"sv57" and "sv64" as follows:
-cpu rv64,sv57=on # Linux will boot using sv57 scheme
-cpu rv64,sv39=on # Linux will boot using sv39 scheme
-cpu rv64,sv57=off # Linux will boot using sv48 scheme
-cpu rv64 # Linux will boot using sv57 scheme by default
We take the highest level set by the user:
-cpu rv64,sv48=on,sv57=on # Linux will boot using sv57 scheme
We make sure that invalid configurations are rejected:
-cpu rv64,sv39=off,sv48=on # sv39 must be supported if higher modes are
# enabled
We accept "redundant" configurations:
-cpu rv64,sv48=on,sv57=off # Linux will boot using sv48 scheme
And contradictory configurations:
-cpu rv64,sv48=on,sv48=off # Linux will boot using sv39 scheme
Co-Developed-by: Ludovic Henry <ludovic@rivosinc.com>
Signed-off-by: Ludovic Henry <ludovic@rivosinc.com>
Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Bin Meng <bmeng@tinylab.org>
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Frank Chang <frank.chang@sifive.com>
Message-ID: <20230303131252.892893-4-alexghiti@rivosinc.com>
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Zicbom is the Cache-Block Management extension defined in the already
ratified RISC-V Base Cache Management Operation (CBO) ISA extension [1].
The extension contains three instructions: cbo.clean, cbo.flush and
cbo.inval. All of them must be implemented in the same group as LQ and
cbo.zero due to overlapping patterns.
All these instructions can throw a Illegal Instruction/Virtual
Instruction exception, similar to the existing cbo.zero. The same
check_zicbo_envcfg() is used to handle these exceptions.
Aside from that, these instructions also need to handle page faults and
guest page faults. This is done in a new check_zicbom_access() helper.
As with Zicboz, the cache block size for Zicbom is also configurable.
Note that the spec determines that Zicbo[mp] and Zicboz can have
different cache sizes (Section 2.7 of [1]), so we also include a
'cbom_blocksize' to go along with the existing 'cboz_blocksize'. They
are set to the same size, so unless users want to play around with the
settings both sizes will be the same.
[1] https://github.com/riscv/riscv-CMOs/blob/master/specifications/cmobase-v1.0.1.pdf
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Weiwei Li <liweiwei@iscas.ac.cn>
Co-developed-by: Philipp Tomsich <philipp.tomsich@vrull.eu>
Signed-off-by: Christoph Muellner <cmuellner@linux.com>
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Message-ID: <20230224132536.552293-4-dbarboza@ventanamicro.com>
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
The RISC-V base cache management operation (CBO) ISA extension has been
ratified. It defines three extensions: Cache-Block Management, Cache-Block
Prefetch and Cache-Block Zero. More information about the spec can be
found at [1].
Let's start by implementing the Cache-Block Zero extension, Zicboz. It
uses the cbo.zero instruction that, as with all CBO instructions that
will be added later, needs to be implemented in an overlap group with
the LQ instruction due to overlapping patterns.
cbo.zero throws a Illegal Instruction/Virtual Instruction exception
depending on CSR state. This is also the case for the remaining cbo
instructions we're going to add next, so create a check_zicbo_envcfg()
that will be used by all Zicbo[mz] instructions.
[1] https://github.com/riscv/riscv-CMOs/blob/master/specifications/cmobase-v1.0.1.pdf
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Co-developed-by: Philipp Tomsich <philipp.tomsich@vrull.eu>
Signed-off-by: Christoph Muellner <cmuellner@linux.com>
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Weiwei Li <liweiwei@iscas.ac.cn>
Message-ID: <20230224132536.552293-3-dbarboza@ventanamicro.com>
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Weiwei Li <liweiwei@iscas.ac.cn> says:
This patchset adds support svadu extension. It also fixes some
relationship between *envcfg fields and Svpbmt/Sstc extensions.
Specification for Svadu extension can be found in:
https://github.com/riscv/riscv-svadu
* b4-shazam-merge:
target/riscv: Export Svadu property
target/riscv: Add *envcfg.HADE related check in address translation
target/riscv: Add *envcfg.PBMTE related check in address translation
target/riscv: Add csr support for svadu
target/riscv: Fix the relationship of PBMTE/STCE fields between menvcfg and henvcfg
target/riscv: Fix the relationship between menvcfg.PBMTE/STCE and Svpbmt/Sstc extensions
Message-ID: <20230224040852.37109-1-liweiwei@iscas.ac.cn>
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
RISC-V defines a handful of extensions related to floating point, along
with various relationships between these and other extensions. This
patch set adds support for the Zvfh, Zvhfmin, and Zve64d extensions;
along with a handful of fixes and cleanups related to the other
floating-point extension relationships.
* b4-shazam-merge
target/riscv: Expose properties for Zv* extensions
target/riscv: Simplify check for EEW = 64 in trans_rvv.c.inc
target/riscv: Fix check for vector load/store instructions when EEW=64
target/riscv: Add support for Zvfh/zvfhmin extensions
target/riscv: Remove rebundunt check for zve32f and zve64f
target/riscv: Replace check for F/D to Zve32f/Zve64d in trans_rvv.c.inc
target/riscv: Simplify check for Zve32f and Zve64f
target/riscv: Indent fixes in cpu.c
target/riscv: Add propertie check for Zvfh{min} extensions
target/riscv: Fix relationship between V, Zve*, F and D
target/riscv: Add cfg properties for Zv* extensions
target/riscv: Simplify the check for Zfhmin and Zhinxmin
target/riscv: Fix the relationship between Zhinxmin and Zhinx
target/riscv: Fix the relationship between Zfhmin and Zfh
Message-ID: <20230215020539.4788-1-liweiwei@iscas.ac.cn>
[Palmer: commit text]
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
RISCV_FEATURE_MMU is set whether cpu->cfg.mmu is set, so let's just use
the flag directly instead.
With this change the enum is also removed. It is worth noticing that
this enum, and all the RISCV_FEATURES_* that were contained in it,
predates the existence of the cpu->cfg object. Today, using cpu->cfg is
an easier way to retrieve all the features and extensions enabled in the
hart.
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Weiwei Li <liweiwei@iscas.ac.cn>
Reviewed-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: LIU Zhiwei <zhiwei_liu@linux.alibaba.com>
Message-ID: <20230222185205.355361-10-dbarboza@ventanamicro.com>
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
At this moment, and apparently since ever, we have no way of enabling
RISCV_FEATURE_MISA. This means that all the code from write_misa(), all
the nuts and bolts that handles how to properly write this CSR, has
always been a no-op as well because write_misa() will always exit
earlier.
This seems to be benign in the majority of cases. Booting an Ubuntu
'virt' guest and logging all the calls to 'write_misa' shows that no
writes to MISA CSR was attempted. Writing MISA, i.e. enabling/disabling
RISC-V extensions after the machine is powered on, seems to be a niche
use.
After discussions in the mailing list, most notably in [1], we reached
the consensus that this code is not suited to be exposed to users
because it's not well tested, but at the same time removing it is a bit
extreme because we would like to fix it, and it's easier to do so with
the code available to use instead of fetching it from git log.
The approach taken here is to get rid of RISCV_FEATURE_MISA altogether
and use a new experimental flag called x-misa-w. The default value is
false, meaning that we're keeping the existing behavior of doing nothing
if a write_misa() is attempted. As with any existing experimental flag,
x-misa-w is also a temporary flag that we need to remove once we fix
write_misa().
[1] https://lists.gnu.org/archive/html/qemu-devel/2023-02/msg05092.html
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Weiwei Li <liweiwei@iscas.ac.cn>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Weiwei Li<liweiwei@iscas.ac.cn>
Reviewed-by: LIU Zhiwei <zhiwei_liu@linux.alibaba.com>
Message-ID: <20230222185205.355361-4-dbarboza@ventanamicro.com>
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
This patch adds support for the XTheadCmo ISA extension.
To avoid interfering with standard extensions, decoder and translation
are in its own xthead* specific files.
Future patches should be able to easily add additional T-Head extension.
The implementation does not have much functionality (besides accepting
the instructions and not qualifying them as illegal instructions if
the hart executes in the required privilege level for the instruction),
as QEMU does not model CPU caches and instructions are documented
to not raise any exceptions.
Co-developed-by: LIU Zhiwei <zhiwei_liu@linux.alibaba.com>
Signed-off-by: Christoph Müllner <christoph.muellner@vrull.eu>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20230131202013.2541053-2-christoph.muellner@vrull.eu>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
There is an informal contract between the cpu_init() functions and
riscv_cpu_realize(): if cpu->env.misa_ext is zero, assume that the
default settings were loaded via register_cpu_props() and do validations
to set env.misa_ext. If it's not zero, skip this whole process and
assume that the board somehow did everything.
At this moment, all SiFive CPUs are setting a non-zero misa_ext during
their cpu_init() and skipping a good chunk of riscv_cpu_realize(). This
causes problems when the code being skipped in riscv_cpu_realize()
contains fixes or assumptions that affects all CPUs, meaning that SiFive
CPUs are missing out.
To allow this code to not be skipped anymore, all the cpu->cfg.ext_*
attributes needs to be set during cpu_init() time. At this moment this
is being done in register_cpu_props(). The SiFive boards are setting
their own extensions during cpu_init() though, meaning that they don't
want all the defaults from register_cpu_props().
Let's move the contract between *_cpu_init() and riscv_cpu_realize() to
register_cpu_props(). Inside this function we'll check if
cpu->env.misa_ext was set and, if that's the case, set all relevant
cpu->cfg.ext_* attributes, and only that. Leave the 'misa_ext' = 0 case
as is today, i.e. loading all the defaults from riscv_cpu_extensions[].
register_cpu_props() can then be called by all the cpu_init() functions,
including the SiFive ones. This will make all CPUs behave more in line
with what riscv_cpu_realize() expects.
This will also make the cpu_init() functions even more alike, but at this
moment we would need some design changes in how we're initializing
extensions/attributes (e.g. some CPUs are setting cfg options after
register_cpu_props(), so we can't simply add the function to a common
post_init() hook) to make a common cpu_init() code across all CPUs.
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20230113175230.473975-2-dbarboza@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
This patch adds support for the Zawrs ISA extension.
Given the current (incomplete) implementation of reservation sets
there seems to be no way to provide a full emulation of the WRS
instruction (wake on reservation set invalidation or timeout or
interrupt). Therefore, we just exit the TB and return to the main loop.
The specification can be found here:
https://github.com/riscv/riscv-zawrs/blob/main/zawrs.adoc
Note, that the Zawrs extension is frozen, but not ratified yet.
Changes since v3:
* Remove "RFC" since the extension is frozen
* Rebase on master and fix integration issues
* Fix entry ordering in extension list
Changes since v2:
* Rebase on master and resolve conflicts
* Adjustments according to a specification change
* Inline REQUIRE_ZAWRS() since it has only one user
Changes since v1:
* Adding zawrs to the ISA string that is passed to the kernel
Signed-off-by: Christoph Müllner <christoph.muellner@vrull.eu>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20221005144948.3421504-1-christoph.muellner@vrull.eu>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
The max count in itrigger can be 0x3FFF, which will cause a no trivial
translation and execution overload.
When icount is enabled, QEMU provides API that can fetch guest
instruction number. Thus, we can set an timer for itrigger with
the count as deadline.
Only when timer expires or priviledge mode changes, do lazy update
to count.
Signed-off-by: LIU Zhiwei <zhiwei_liu@linux.alibaba.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20221013062946.7530-3-zhiwei_liu@linux.alibaba.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
When icount is not enabled, there is no API in QEMU that can get the
guest instruction number.
Translate the guest code in a way that each TB only has one instruction.
After executing the instruction, decrease the count by 1 until it reaches 0
where the itrigger fires.
Note that only when priviledge matches the itrigger configuration,
the count will decrease.
Signed-off-by: LIU Zhiwei <zhiwei_liu@linux.alibaba.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20221013062946.7530-2-zhiwei_liu@linux.alibaba.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
