Whenever we enter/leave a critical section, the interrupt status is
saved and, then, restored. However, for the ESP32-S3's BLE adapter,
entering/leaving a critical section is done on separate functions
that need to be registered as a callback.
The status flag was being saved as a global variable. However,
calling nested enter_critical_section would overwrite this global
variable that was storing the previous flag and, when leaving the
last critical section, the restored status would be different from
the one expected. The proposed solution for this issue is to create
a global array to store the interrupt status flags for nested calls.
This commit sets the BLE's interrupt as a IRAM-enabled interrupt,
which enables it to run during a SPI flash operation. This enables
us to create a cache to off-load semaphores and message queues
operations and treat them when the SPI flash operation is finished.
By doing that, we avoid packet losses during a SPI flash operation.
This commit provides an interface to register ISRs that run from
IRAM and keeps track of the non-IRAM interrupts. It enables, for
instance, to avoid disabling all the interrupts during a SPI flash
operation: IRAM-enabled ISRs are, then, able to run during these
operations.
When allocating a CPU interrupt, make sure to select the correct
CPU core to query for it. Simply checking for the current CPU does
not satisfy this requirement because the CPU allocation thread may
be executed by the other core: it's necessary to stick with the
intended CPU passed as an argument of the `esp32s3_setup_irq`.
The registered `task_create_wrapper` receives the `core_id`, but
the current implementation ignores this parameter while calling
`esp_task_create_pinned_to_core`. This commit fix this.
When allocating a CPU interrupt, make sure to select the correct
CPU core to query for it. Simply checking for the current CPU does
not satisfy this requirement because the CPU allocation thread may
be executed by the other core: it's necessary to stick with the
intended CPU passed as an argument of the `esp32_setup_irq`.
The SHM physically backed memory does not belong to the user process,
but the page table containing the mapping does -> delete the page table
memory regardless.
This is a collection of tweaks / optimizations to the driver to limit
CPU usage as well as interrupt processing times.
The changes are as follows:
- setfrequency is now no-op if the frequency does not change. Accessing
MPFS_SPI_CONTROL requires synchronization to the FIC domain, which
takes unnecessary time if nothing changes
- load/unload FIFO loops optimized so !buffer, priv->nbits and i==last are
only tested once (instead of for every word written in loop).
- Disable the RX interrupt only once (again, FIC domain access is slow)
- In case a spurious MPFS_SPI_DATA_RX interrupt arrives, just wipe the
whole RX FIFO, instead of trying to read it byte-by-byte
stm32_ifdown() holds critical section when calling stm32_ethreset().
That function used to call up_mdelay(10) while waiting for the ethernet
peripheral reset to complete. This resulted in excessively long
critical section time with interrupts disabled.
The actual expected delay is a few clock ticks of the 50 MHz clock domain.
This commit changes polling interval to 1us and maximum to 10us.
if not unlink shm, the shared memory object still exists in host /dev/shm after quit
if nuttx is started with administrator privileges, or if it is restarted with user
privileges, there will be a problem with the permission to open this shm file
Signed-off-by: yintao <yintao@xiaomi.com>
