The register context is not needed, the original idea was to provide
the user stack pointer for signal handler delivery, but the user stack
can be obtained via sp_el0 so the context registers are not needed.
SP0 is not stored upon exception entry anyways, so this code is just
completely redundant and wrong.
reason:
In SMP, when a context switch occurs, restore_critical_section is executed.
To reduce the time taken for context switching, we directly pass the required
parameters to restore_critical_section instead of acquiring them repeatedly.
Signed-off-by: hujun5 <hujun5@xiaomi.com>
The vaddr field in TLBI means: Bits[55:12] of the virtual address to match.
This basically means the page offset of the virtual address, so the input
vaddr must be shifted to the page offset.
Reference TLBI VALE1IS register description from ARMv8-A reference manual.
The 12:0 bits in table descriptors are RES0 and AF is the 10th bit, so
it is not valid to set it in this case.
Fix this by moving AF to the common MMU_MT_NORMAL_FLAGS field
Make sure the user L1 page is updated to system memory when the kernel
mappings are copied.
Also, flush the I-cache when switching address environments.
Make this_cpu is arch independent and up_cpu_index do that.
In AMP mode, up_cpu_index() may return the index of the physical core.
Signed-off-by: fangxinyong <fangxinyong@xiaomi.com>
In corner case, the pending ISR will be triggered immediately
after enable the IRQ, this PR will setting CPU affinity first
to avoid routing the unexpected IRQ to other CPUs.
Signed-off-by: chao an <anchao@lixiang.com>
Only in the non-critical region， nuttx can the respond to the irq and not hold the lock
When returning from the irq, there is no need to check whether the lock needs to be released
we also need keep restore_critical_section in svc call
test:
Configuring NuttX and compile:
$ ./tools/configure.sh -l qemu-armv8a:nsh_smp
$ make
Running with qemu
$ qemu-system-aarch64 -cpu cortex-a53 -smp 4 -nographic \
-machine virt,virtualization=on,gic-version=3 \
-net none -chardev stdio,id=con,mux=on -serial chardev:con \
-mon chardev=con,mode=readline -kernel ./nuttx
Signed-off-by: hujun5 <hujun5@xiaomi.com>
This is the initial version for kernel mode build on the arm64 platform.
It works much in the same way as the risc-v implementation so any
highlights can be read from there.
Features that have been tested working:
- Creating address environments
- Loading init (nsh) from elf file
- Booting to nsh
- Starting other processes from nsh
- ostest runs to completion
Features that are not tested / do not work:
- SHM / shared memory support
- Kernel memory mapping (MM_KMAP)
- fork/vfork
An example qemu target is provided as a separate patch:
tools/configure.sh qemu-armv8a:knsh
Why? Because this allows optimizing the user system call path in such
a way that the parameter registers don't have to be read from the saved
integer register context when the system call is executed.
"/mnt/yang/qixinwei_vela_warnings/nuttx/include/nuttx/spinlock.h", line 252: warning #76-D:
argument to macro is empty
SP_DSB();
^
"/mnt/yang/qixinwei_vela_warnings/nuttx/include/nuttx/spinlock.h", line 261: warning #76-D:
argument to macro is empty
SP_DMB();
^
"/mnt/yang/qixinwei_vela_warnings/nuttx/include/nuttx/spinlock.h", line 252: warning #76-D:
argument to macro is empty
SP_DSB();
^
"/mnt/yang/qixinwei_vela_warnings/nuttx/include/nuttx/spinlock.h", line 261: warning #76-D:
argument to macro is empty
SP_DMB();
^
"/mnt/yang/qixinwei_vela_warnings/nuttx/include/nuttx/spinlock.h", line 296: warning #76-D:
argument to macro is empty
SP_DSB();
^
Signed-off-by: yanghuatao <yanghuatao@xiaomi.com>
when the thread to backtrace is exiting, get_tcb and up_backtrace in
different critical section may cause try to dump invalid pointer, have
to ensure the nxsched_get_tcb and up_backtrace inside same critical
section procedure.
Signed-off-by: buxiasen <buxiasen@xiaomi.com>
This adds enablers for setting various clocks to some default
values. Also, this provides helpers to grant nonsecure access
to a number of clocks. Bootloader may utilize these to make
the system boot in a deterministic manner.
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
In the algorithm there is a subtraction (int - unsigned), which results (potentially overflowed)
unsigned.
Passing this to macro ABS and the assigning to int doesn't work ( unsigned is always >= 0 ).
Fix this by replacing (dangerous) ABS macro with stdlib's standard "int abs(int)"
and change the substraction to (int - int).
Signed-off-by: Jukka Laitinen <jukkax@ssrc.tii.ae>
This is an initial FlexSPI SPI NOR MTD driver for IMX9
This supprts M25P SPI NOR on FlexSPI for now, and can later be extended to other
SPINOR devices if needed. The following configurations are needed to use this driver:
CONFIG_IMX9_FLEXSPI_NOR=y
CONFIG_MTD_M25P=y
In addition, board initialization logic needs to call the imx9_flexspi_nor_initialize
to receive a pointer to the mtd device.
Signed-off-by: Jukka Laitinen <jukkax@ssrc.tii.ae>
Co-authored-by: Jouni Ukkonen <jouni.ukkonen@unikie.com>
During the boot phase, when we transition from tee smp to ap smp, we can use a busy waitflag to wait for the completion of the initialization of ap's core0
test:
We can use qemu for testing.
compiling
make distclean -j20; ./tools/configure.sh -l qemu-armv8a:nsh_smp ;make -j20
running
qemu-system-aarch64 -cpu cortex-a53 -smp 4 -nographic -machine virt,virtualization=on,gic-version=3 -net none -chardev stdio,id=con,mux=on -serial chardev:con -mon chardev=con,mode=readline -kernel ./nuttx
Signed-off-by: hujun5 <hujun5@xiaomi.com>
Only in the non-critical region， nuttx can the respond to the irq and not hold the lock
When returning from the irq, there is no need to check whether the lock needs to be restored
test:
We can use qemu for testing.
compiling
make distclean -j20; ./tools/configure.sh -l qemu-armv8a:nsh_smp ;make -j20
running
qemu-system-aarch64 -cpu cortex-a53 -smp 4 -nographic -machine virt,virtualization=on,gic-version=3 -net none -chardev stdio,id=con,mux=on -serial chardev:con -mon chardev=con,mode=readline -kernel ./nuttx
Signed-off-by: hujun5 <hujun5@xiaomi.com>
reduce the time consumed by function call
test:
We can use qemu for testing.
compiling
make distclean -j20; ./tools/configure.sh -l qemu-armv8a:nsh_smp ;make -j20
running
qemu-system-aarch64 -cpu cortex-a53 -smp 4 -nographic -machine virt,virtualization=on,gic-version=3 -net none -chardev stdio,id=con,mux=on -serial chardev:con -mon chardev=con,mode=readline -kernel ./nuttx
Signed-off-by: hujun5 <hujun5@xiaomi.com>
cpu0 cpu1:
user_main
signest_test
sched_unlock
nxsched_merge_pending
nxsched_add_readytorun
up_cpu_pause
arm_sigdeliver
enter_critical_section
Reason:
In the SMP, cpu0 is already in the critical section and waiting for cpu1 to enter the suspended state.
However, when cpu1 executes arm_sigdeliver, it is in the irq-disabled state but not in the critical section.
At this point, cpu1 is unable to respond to interrupts and
is continuously attempting to enter the critical section, resulting in a deadlock.
Resolve:
adjust the logic, do not entering the critical section when interrupt-disabled.
test:
We can use qemu for testing.
compiling
make distclean -j20; ./tools/configure.sh -l qemu-armv8a:nsh_smp ;make -j20
running
qemu-system-aarch64 -cpu cortex-a53 -smp 4 -nographic -machine virt,virtualization=on,gic-version=3 -net none -chardev stdio,id=con,mux=on -serial chardev:con -mon chardev=con,mode=readline -kernel ./nuttx
Signed-off-by: hujun5 <hujun5@xiaomi.com>
when repeatedly enabling and disabling string-controlled configurations,
the generated toolchain configuration may be incorrect.
Signed-off-by: xuxin19 <xuxin19@xiaomi.com>
Search and replace ARCH_BOOT_EL3 with more generic
ARCH_ARM64_EXCEPTION_LEVEL that holds the EL level
in an integer variable.
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
This provides a capable bootloader that may be run from OCRAM.
The OCRAM contains regions that are always zero, so the linker
file avoids those with best effort.
iMX9 infrastructure expects:
- 0x20480000 (Start of OCRAM, AHAB)
- 0x2049a000 (NuttX or SPL)
- 0x204e0000 (ARM Trustzone, not used)
When started from SD-card, the offsets are:
- 0x1f000 with AHAB
- 0xa000 without AHAB
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
Value 1021, when read from ICC_IAR0_EL1 means:
"The GIC returns this value in response to a read of ICC_IAR0_EL1 or ICC_HPPIR0_EL1 at EL3,
to indicate that the interrupt being acknowledged is one which is expected to be handled at
Non-secure EL1 or EL2. This INTID is only returned when the PE is executing at EL3 using
AArch64 state, or when the PE is executing in AArch32 state in Monitor mode."
When this happens:
- FIQ is fired on group0
- IRQ is pending at group1
So simply check and handle the interrupt. In short, this provides interrupt support for
EL3.
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
This provides means to run NuttX completely in EL3. This may
be useful with NuttX based bootloaders that are executed from
OCRAM. Instead of SPL/U-boot combo, NuttX may replace SPL
completely.
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
We can save execution time by
inline arm64_fullcontextrestore and arm64_switchcontext
test:
We can use qemu for testing.
compiling
make distclean -j20; ./tools/configure.sh -l qemu-armv8a:nsh_smp ;make -j20
running
qemu-system-aarch64 -cpu cortex-a53 -smp 4 -nographic -machine virt,virtualization=on,gic-version=3 -net none -chardev stdio,id=con,mux=on -serial chardev:con -mon chardev=con,mode=readline -kernel ./nuttx
Signed-off-by: hujun5 <hujun5@xiaomi.com>
Optimal size of granule is 64B (the dcache line size).
We can use it now as we don't have max. 32 granules limitation anymore.
Signed-off-by: Jani Paalijarvi <jani.paalijarvi@unikie.com>
Using user allocated buffers for DMA transfers is not safe for two reasons:
- User space memory is virtual memory, DMA needs physical memory
- User memory buffer alignment cannot be guaranteed -> cache line ops
are not safe
Add a simple allocator for DMA safe memory. It will provide contiguous
blocks of memory with D-Cache line size alignment.
NOTE: The optimal granule size is the D-Cache line size (64), but due
to restrictions in the granule allocator this would result in a maximum
block size of 2K only, thus use 256B granules instead givin 8K max block
size.
Once the granule allocator is fixed this limitation can be removed.
This patch adds a way to configure PLL frequencies. The configuration is
given by board logic.
These values should only be modified by the bootloader, but we don't have
that yet so the flag is never activated.
The frequency LUT idea is not necessary as the PLL output can be calculated
by the CPU. It is better to do this as the clocks are set by the SPL (2nd
stage program loader) which means the NuttX payload would not have access
to such a LUT anyhow.
The mask PLL_DIV_RDIV_MASK is also fixed, as that was simply wrong.
Also add call to imx9_clockconfig (although it does not do anything yet).
This driver supports both eDMA3 and eDMA4 (also referred to as DMA0 / DMA1
in some contexts..)
The IP blocks are almost identical, with sufficiently minor differences
to use them via a unified driver. The price is a great amount of code
obfuscation in the hardware description layer.
This fixes names of program entry and linker script files so that to
support building kernel mode apps using CMake and export package.
flat and protected mode should be the same as before.
Signed-off-by: Yanfeng Liu <yfliu2008@qq.com>
icr is tested below with macros like GPIO_INT_LOWLEVEL et al. Those macros
are shifted left by GPIO_INTCFG_SHIFT, so the temporary icr variable
should not be shifted right.
- Add missing include guard
- Add missing C++ guard
- Fix the initialization ordering in IOMUX_PADCFG macro. Why ? Becaused of:
imx9_iomuxc.h:54:3: error: designator order for field 'iomux_padcfg_s::dsyreg' does not match declaration order in 'iomux_padcfg_s'
54 | }
|
The original assumption was that the interrupt numbers are divided
so that 16 pins from 1 port are handled by a single interrupt source.
So source 0 would handle pins 0-15 and source 1 would handle pins 16-31.
This assumption is wrong, each pin has two sources, thus there are two
interrupt lines for each pin.
The driver uses source 0, and leaves source 1 disabled.
