in SMP, signal processing cannot be nested, we use xcp.sigdeliver to identify whether there is currently a signal being processed, but this state does not match the actual situation
One possible scenario is that signal processing has already been completed, but an interrupt occurs, resulting in xcp.sigdeliver not being correctly set to NULL,
At this point, a new signal arrives, which can only be placed in the queue and cannot be processed immediately
Our solution is that signal processing and signal complete status are set in the same critical section, which can ensure status synchronization
Signed-off-by: hujun5 <hujun5@xiaomi.com>
CURRENT_REGS may change during assert handling, so pass
in the 'regs' parameter at the entry point of _assert.
Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
When l2cc is already in disable state, performing a disable operation
again will flush incorrect cache data to memory
Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
Invalidate operations at DDI0246H_l2c310_r3p3_trm:
If there is a stale entry in the L2 cache, the system enables the invalidation of
the L1 cache. But before the controller invalidates the L2 cache, it allocates a
line from the L2 cache to an L1 cache.
The robust code sequence for invalidation with a non-exclusive cache arrangement is:
1. InvalLevel2 Address ; forces the address out past level 2
2. CACHE SYNC ; Ensures completion of the L2 inval
3. InvalLevel1 Address ; This is broadcast within the cluster
4. DSB ; Ensure completion of the inval as far as Level 2.
This sequence ensures that, if there is an allocation to L1 after the L1 invalidation, the data
picked up is the new data and not stale data from the L2
Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
minidump will backtrace failure when use C code to save user context,
because the stack push operation in C code can disrupt the stack information.
Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
The text describes an issue related to the running task in code.
The running task is only used when calling the _assert function
to indicate the task that was running before an exception occurred.
However, the current code only updates the running task during
irq_dispatch, which is suitable for ARM-M architecture but not
for ARM-A or ARM-R architecture, because their context switches
are not done through irq handler. Therefore, if the following
process is followed, the value of the running task will be incorrect:
1. task1 is running, this_task()=task1
2. do_irq is executed, setting running task()=task1
3. task1 switches to task2
4. task2 is running and generates a data abort
5. In the data abort, the _assert function is called,
and the running task obtained is still task1, but
the actual task that generated the exception is task2.
Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
On a GICv2 implementation, setting GICC_CTLR.EOImode to 1 separates
the priority drop and interrupt deactivation operations.
Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
Summary:
- Support arm64 pmu api, Currently only the cycle counter function is supported.
- Using ARM64 PMU hardware capability to implement perf interface, modify all
perf interface related code.
- Support for pmu init under smp.
Signed-off-by: wangming9 <wangming9@xiaomi.com>
Current Toolchain.defs set the compile flags directly, it's OK for
target specified gcc toolchain.
But some LLVM based toolchains (Rust/Zig etc) use single toolchain to handle all supported paltform.
In this patch, arch level Toolchain.defs export standard LLVM style arch flags, and let <Lang>.defs to map them into internal style,
This will simplify the intergration of non-c language.
Signed-off-by: Huang Qi <huangqi3@xiaomi.com>
The function is not relevant any longer, remove it. Also remove
save_addrenv_t, the parameter taken by up_addrenv_restore.
Implement addrenv_select() / addrenv_restore() to handle the temporary
instantiation of address environments, e.g. when a process is being
created.
Detach the address environment handling from the group structure to the
tcb. This is preparation to fix rare cases where the system (MMU) is left
without a valid page directory, e.g. when a process exits.
This is preparation for moving address environments out of the group
structure into the tcb.
Why move ? Because the group is destroyed very early in the exit phase,
but the MMU mappings are needed until the context switch to the next
process is complete. Otherwise the MMU will lose its mappings and the
system will crash.
There were two issues with signal handling:
- With a kernel stack the "info" parameter was passed from kernel memory.
This is fixed by making a stack frame to the user stack and copying it
there.
- If the signal handler uses a system call, the kernel stack was completely
and unconditionally destroyed, resulting in a crash in the user application
There is also no need to check ustkptr, it is always NULL. Why ? Because
signal delivery is deferred when a system call is being executed.
This is unnecessary, the address environment is getting wiped anyway,
there is no need to remove the L1 references because they will get
wiped when the page directory is changed
When invalidate address is not aligned with cache line,
must align address and flush the cache line.
Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
The dependency should be vice versa; the MM_SHM should depend on the
existence of the virtual memory range allocator.
Create a new CONFIG flag CONFIG_ARCH_VMA_MAPPING, which will define that
there is a virtual memory range allocator. Make MM_SHM select that flag
Signed-off-by: Jukka Laitinen <jukkax@ssrc.tii.ae>
Registers S0-S15 (D0-D7, Q0-Q3) do not need to be preserved. They can be used for passing
arguments or returning results in standard procedure-call variants.
Registers D16-D31 (Q8-Q15), do not need to be preserved.
Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
When a function is known to preserve eight-byte alignment of the stack, armclang assigns the build
attribute Tag_ABI_align_preserved to that function. However, the armclang integrated assembler does
not automatically assign this attribute to assembly code.
Signed-off-by: xiangdong6 <xiangdong6@xiaomi.com>
In SMP mode, if all cores start at same time, all from __start(),
then only primary need do initialize, so others core should wait
primary, use 'sev' let the non-primary continue to __cpuN_start().
Signed-off-by: ligd <liguiding1@xiaomi.com>
NON-primary cpu will invalidate cpu0's cache L2, that will caused cpu0's data mismatch, and then system crash
Signed-off-by: ligd <liguiding1@xiaomi.com>
Summary:
- Currently, CONFIG_ARCH_PGPOOL_MAPPING=y is necessary for
CONFIG_BUILD_KERNEL=y.
- This commit removes the code for CONFIG_ARCH_PGPOOL_MAPPING=n
Impact:
- None
Testing:
- Tested with sabre-6quad:netknsh_smp
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
