FPU context switching is always performed on demand through the FPU
access exception handler. Actual task switching only grants or denies
FPU access depending on the current FPU owner.
Because RISC-V doesn't have a dedicated FPU access exception, we must
catch the Illegal Instruction exception and look for actual FP opcodes.
There is no longer a need to allocate FPU storage on the stack for every
exception making esf smaller and stack overflows less likely.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Currently it is assumed that Zephyr CPU numbers match their hartid
value one for one. This assumption was relied upon to efficiently
retrieve the current CPU's `struct _cpu` pointer.
People are starting to have systems with a mix of different usage for
each CPU and such assumption may no longer be true.
Let's completely decouple the hartid from the Zephyr CPU number by
stuffing each CPU's `struct _cpu` pointer in their respective scratch
register instead. `arch_curr_cpu()` becomes more efficient as well.
Since the scratch register was previously used to store userspace's
exception stack pointer, that is now moved into `struct _cpu_arch`
which implied minor user space entry code cleanup and rationalization.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Some RISC-V SoCs implement a mechanism for hardware supported stacking /
unstacking of registers during ISR / exceptions. What happens is that on
ISR / exception entry part of the context is automatically saved by the
hardware on the stack without software intervention, and the same part
of the context is restored by the hardware usually on mret.
This is currently not yet supported by Zephyr, where the full context
must be saved by software in the full fledged ESF. This patcheset is
trying to address exactly this case.
At least three things are needed to support in a general fashion this
problem: (1) a way to store in software only the part of the ESF not
already stacked by hardware, (2) a way to restore in software only the
part of the context that is not going to be restored by hardware and (3)
a way to define a custom ESF.
Point (3) is important because the full ESF frame is now composed by a
custom part depending on the hardware (that can choose which register to
stack / unstack and the order they are saved onto the stack) and a part
defined in software for the remaining part of the context.
In this patch a new CONFIG_RISCV_SOC_HAS_ISR_STACKING is introduced that
enables the code path supporting the three points by the mean of three
macros that must be implemented by the user in a soc_stacking.h file:
SOC_ISR_SW_STACKING, SOC_ISR_SW_UNSTACKING and SOC_ISR_STACKING_ESF
(refer to the symbol help for more details).
This is an example of soc_isr_stacking.h for an hardware that doesn't do
any hardware stacking / unstacking but everything is managed in
software:
#ifndef __SOC_ISR_STACKING
#define __SOC_ISR_STACKING
#if !defined(_ASMLANGUAGE)
#define SOC_ISR_STACKING_ESF_DECLARE \
struct __esf { \
unsigned long ra; \
unsigned long t0; \
unsigned long t1; \
unsigned long t2; \
unsigned long t3; \
unsigned long t4; \
unsigned long t5; \
unsigned long t6; \
unsigned long a0; \
unsigned long a1; \
unsigned long a2; \
unsigned long a3; \
unsigned long a4; \
unsigned long a5; \
unsigned long a6; \
unsigned long a7; \
unsigned long mepc; \
unsigned long mstatus; \
unsigned long s0; \
} __aligned(16)
#else
#define SOC_ISR_SW_STACKING \
addi sp, sp, -__z_arch_esf_t_SIZEOF; \
DO_CALLER_SAVED(sr);
#define SOC_ISR_SW_UNSTACKING \
DO_CALLER_SAVED(lr);
#endif /* _ASMLANGUAGE */
#endif /* __SOC_ISR_STACKING */
Signed-off-by: Carlo Caione <ccaione@baylibre.com>
All SOC_ERET definitions expand to the mret instruction (used to return
from a trap: exception or interruption). The 'eret' instruction existed
in previous RISC-V privileged specs, but it doesn't seem to be used in
Zephyr (ref. RISC-V Privileged Architectures 3.2.2).
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
Some early RISC-V SoCs have a problem when an `mret` instruction is used
outside a trap handler.
After the latest Zephyr RISC-V huge rework, the arch_switch code is
indeed calling `mret` when not in handler mode, breaking some early
RISC-V platforms.
Optionally restore the old behavior by adding a new
CONFIG_RISCV_ALWAYS_SWITCH_THROUGH_ECALL symbol.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Signed-off-by: Carlo Caione <ccaione@baylibre.com>
This is really useful only for one case i.e. when testing against zero.
Do that test inline where it is needed and make the rest of the code
independent from the actual numerical value being tested to make code
maintenance easier if/when new cases are added.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
This reverts the bulk of commit c8bfc2afda ("riscv: make
arch_is_user_context() SMP compatible") and replaces it with a flag
stored in the thread local storage (TLS) area, therefore making TLS
mandatory for userspace support on RISC-V.
This has many advantages:
- The tp (x4) register is already dedicated by the standard for this
purpose, making TLS support almost free.
- This is very efficient, requiring only a single instruction to clear
and 2 instructions to set.
- This makes the SMP case much more efficient. No need for funky
exception code any longer.
- SMP and non-SMP now use the same implementation making maintenance
easier.
- The is_user_mode variable no longer requires a dedicated PMP mapping
and therefore freeing one PMP slot for other purposes.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
5f65dbcc9dab3d39473b05397e05.
The tp (x4) register is neither caller nor callee saved according to
the RISC-V standard calling convention. It only has to be set on thread
context switching and is otherwise read-only.
To protect the kernel against a possible rogue user thread, the tp is
also re-set on exception entry from u-mode.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
For some reasons RISCV is the only arch where the vector table entry is
called __irq_wrapper instead of _isr_wrapper. This is not only a
cosmetic change but Zephyr expects the common ISR handler to be called
_isr_wrapper (for example when generating the IRQ vector table).
Change it.
find ./ -type f -exec sed -i 's/__irq_wrapper/_isr_wrapper/g' {} \;
Signed-off-by: Carlo Caione <ccaione@baylibre.com>
In preparation for the support of RV32E optimize a bit the t* registers
usage limiting that to t{0-2}.
Signed-off-by: Carlo Caione <ccaione@baylibre.com>
After the introduction of arch_switch() in #43085, ECALL is no longer
used for context switching by default, so remove the comment stating so.
Signed-off-by: Carles Cufi <carles.cufi@nordicsemi.no>
Assembler files were not migrated with the new <zephyr/...> prefix.
Note that the conversion has been scripted, refer to #45388 for more
details.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
Add the appropriate hooks effectively replacing the old implementation
with the new one.
Also the stackguard wasn't properly enforced especially with the
usermode combination. This is now fixed.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Set TP in exception context so that it gets loaded into the CPU when
first running the thread. Set TP for secondary cores to related idle TLS
area.
Signed-off-by: Keith Packard <keithp@keithp.com>
This is painful. There is no way for u-mode code to know if we're
currently executing in u-mode without generating a fault, besides
stealing a general purpose register away from the standard ABI
that is. And a global variable doesn't work on SMP as this must be
per-CPU and we could be migrated to another CPU just at the right
moment to peek at the wrong CPU variable (and u-mode can't disable
preemption either).
So, given that we'll have to pay the price of an exception entry
anyway, let's at least make it free to privileged threads by using
the mscratch register as the non-user context indicator (it must
be zero in m-mode for exception entry to work properly). In the
case of u-mode we'll simulate a proper return value in the
exception trap code. Let's settle on the return value in t0
and omit the volatile to give the compiler a chance to cache
the result.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
To do so efficiently on systems without the mul instruction, we use
shifts and adds which is faster and sometimes smaller than a plain loop.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Stop using &_kernel as this is not SMP friendly. Let's use s0 (after
preserving its content) to hold ¤t_cpu instead so it won't have
to be reloaded each time it is needed. This will be even more relevant
when SMP support is added.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Rely on mstatus rather than thread->base.user_options since it is always
up to date (updated by z_riscv_switch) to simplify the code and be SMP
proof. Also carry over SF_INIT to the mstatus being restored in case
it was changed in the mean time.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
The move to arch_switch() is a prerequisite for SMP support.
Make it optimal without the need for an ECALL roundtrip on every
context switch. Performance numbers from tests/benchmarks/sched:
Before:
unpend 107 ready 102 switch 188 pend 218 tot 615 (avg 615)
After:
unpend 107 ready 102 switch 170 pend 217 tot 596 (avg 595)
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
This is a per-thread register that gets updated only when context
switching. No need to load and save it on every exception entry.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Get rid of all those global variables and IRQ locking.
Use the regular IRQ exit path to let tests validate preemption properly.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Complete revamp of the exception entry code, including syscall handling.
Proper syscall frame exception trigger. Many correctness fixes, hacks
removal, etc. etc.
I tried to make this into several commits, but this stuff is all
inter-related and a pain to split.
The diffstat summary:
14 files changed, 250 insertions(+), 802 deletions(-)
Binary size (before):
text data bss dec hex filename
1104 0 0 1104 450 isr.S.obj
64 0 0 64 40 userspace.S.obj
Binary size (after):
text data bss dec hex filename
600 0 0 600 258 isr.S.obj
36 0 0 36 24 userspace.S.obj
Run of samples/userspace/syscall_perf (before):
*** Booting Zephyr OS build zephyr-v3.0.0-325-g3748accae018 ***
Main Thread started; qemu_riscv32
Supervisor thread started
User thread started
Supervisor thread(0x80010048): 384 cycles 509 instructions
User thread(0x80010140): 77312 cycles 77437 instructions
Run of samples/userspace/syscall_perf (after):
*** Booting Zephyr OS build zephyr-v3.0.0-326-g4c877a2753b3 ***
Main Thread started; qemu_riscv32
Supervisor thread started
User thread started
Supervisor thread(0x80010048): 384 cycles 509 instructions
User thread(0x80010138): 7040 cycles 7165 instructions
Yes, that's more than a 10x speed-up!
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Same rationale as preceding commit. Let's create pseudo-instructions in
assembly scope to make the code more uniform and readable.
Furthermore the definition of COPY_ESF_FP() was wrong as the width of
floating point registers vary not according to CONFIG_64BIT but
CONFIG_CPU_HAS_FPU_DOUBLE_PRECISION. It is therefore wrong to use
lr/sr (previously RV_OP_LOADREG/RV_OP_STOREREG) and a regular temporary
register to transfer such content.
Note: There are far more efficient ways to copy FP context around but
such optimisations will come separately.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Those are prominent enough that having RV_OP_LOADREG and RV_OP_STOREREG
shouting at you all over the place is rather unpleasant and bad taste.
Let's create pseudo-instructions of our own with assembler macros
rather than preprocessor defines and only in assembly scope.
This makes the asm code way more uniform and readable.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
The thread->base.user_options field is an uint8_t. Access it using lb.
A "copy" of it is made into __esf.fp_state. Make that field an uint8_t
too and access it with lb/sb.
_callee_saved.fcsr is an uint32_t. Access it with lw/sw.
Ditto for is_user_mode.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
This reverts commit 8686ab5472.
The purpose of this commit will be reintroduced later on top of
a cleaner codebase.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
This reverts commit be28de692c.
The purpose of this commit will be reintroduced later on top of
a cleaner codebase.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Enable `arch_switch()` as preparation for SMP support. This patch
doesn't try to keep support for old style context swap - only switch
based swap is supported, to keep things simple.
A fair amount of refactoring was done in this patch, specially regarding
the code that decides what to do about the ISR. In RISC-V, ECALL
instructions are used to signalize several events, such as user space
system calls, forced syscall, IRQ offload, return from syscall and
context switch. All those handled by the ISR - which also handles
interrupts. After refactor, this "dispatching" step is done at the
beginning of ISR (just after saving generic registers).
As with other platforms, the thread object itself is used as the thread
"switch handle" for the context swap.
Signed-off-by: Ederson de Souza <ederson.desouza@intel.com>
isr.S code currently gets CPU information from global `_kernel` assuming
there's only one CPU. In order to prepare for upcoming SMP support,
change code to actually get current CPU information.
Signed-off-by: Ederson de Souza <ederson.desouza@intel.com>
Non-standard `jalr rd, rs` pseudo-instructions are used.
This commit changes them to `ret` for standard return pseudo-instruction
or `jalr rd, rs, 0` for no offset jump register and link.
Fixes#41100.
Signed-off-by: Henry Hsieh <r901042004@yahoo.com.tw>
This commit add 2 minor fixes of IRQ handling:
1. Save caller registers before calling z_riscv_configure_stack_guard()
in RISC-V assembly.
2. reschedule and no_reschdule code paths use different interrupt
return path after supporting of CONFIG_PMP_STACK_GUARD. back-to-back
interrupt checking is in the reschedule code path so that it should
jump to interrupt return path of reschedule.
Signed-off-by: Jim Shu <cwshu09@gmail.com>
Although CONFIG_USERSPACE is enabled, there are supervisor threads who
don't have privileged stack using exception handler. Only let user
threads to switch to privileged stack in exception handler.
Signed-off-by: Jim Shu <cwshu@andestech.com>
In some cases the 'reschedule' code path is executed when the current
thread is the same as the next thread in the ready Q. If this happens,
the swap_return_value of the thread is ifalsely being reset to -EAGAIN.
This commit prevents the rescheduling code to run if the current thread
is the same as the thread in the ready Q.
Signed-off-by: Wolfgang Reißnegger <gnagflow@fb.com>
RISC-V global pointer (GP) register is neither caller nor callee
register, and it's a constant value in the single ELF file. Thus, we
don't need to save/restore GP at ISR enter/exit. Remove it to optimize
context switch performance.
Signed-off-by: Jim Shu <cwshu@andestech.com>
Since the tracing of thread being switched in/out has the same
instrumentation points, we can roll the tracing function calls
into the one for thread stats gathering functions.
This avoids duplicating code to call another function.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
The IRQ handler has had a major changes to manage syscall, reschedule
and interrupt from user thread and stack guard.
Add userspace support:
- Use a global variable to know if the current execution is user or
machine. The location of this variable is read only for all user
thread and read/write for kernel thread.
- Memory shared is supported.
- Use dynamic allocation to optimize PMP slot usage. If the area size
is a power of 2, only one PMP slot is used, else 2 are used.
Add stack guard support:
- Use MPRV bit to force PMP rules to machine mode execution.
- IRQ stack have a locked stack guard to avoid re-write PMP
configuration registers for each interruption and then win some
cycle.
- The IRQ stack is used as "temporary" stack at the beginning of IRQ
handler to save current ESF. That avoid to trigger write fault on
thread stack during store ESF which that call IRQ handler to
infinity.
- A stack guard is also setup for privileged stack of a user thread.
Thread:
- A PMP setup is specific to each thread. PMP setup are saved in each
thread structure to improve reschedule performance.
Signed-off-by: Alexandre Mergnat <amergnat@baylibre.com>
Reviewed-by: Nicolas Royer <nroyer@baylibre.com>
This code had one purpose only, feed timing information into a test and
was not used by anything else. The custom trace points unfortunatly were
not accurate and this test was delivering informatin that conflicted
with other tests we have due to placement of such trace points in the
architecture and kernel code.
For such measurements we are planning to use the tracing functionality
in a special mode that would be used for metrics without polluting the
architecture and kernel code with additional tracing and timing code.
Furthermore, much of the assembly code used had issues.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
This commit renames the Kconfig `FP_SHARING` symbol to `FPU_SHARING`,
since this symbol specifically refers to the hardware FPU sharing
support by means of FPU context preservation, and the "FP" prefix is
not fully descriptive of that; leaving room for ambiguity.
Signed-off-by: Stephanos Ioannidis <root@stephanos.io>
This commit renames the Kconfig `FLOAT` symbol to `FPU`, since this
symbol only indicates that the hardware Floating Point Unit (FPU) is
used and does not imply and/or indicate the general availability of
toolchain-level floating point support (i.e. this symbol is not
selected when building for an FPU-less platform that supports floating
point operations through the toolchain-provided software floating point
library).
Moreover, given that the symbol that indicates the availability of FPU
is named `CPU_HAS_FPU`, it only makes sense to use "FPU" in the name of
the symbol that enables the FPU.
Signed-off-by: Stephanos Ioannidis <root@stephanos.io>
This change adds full shared floating point support for the RISCV
architecture with minimal impact on threads with floating point
support not enabled.
Signed-off-by: Corey Wharton <coreyw7@fb.com>
