OctoSPI and Memory Mapped Flash support (#11825)
* STM32H730/STM32H750 - Fix use of USE_LP_UART1 instead of USE_LPUART1.
* STM32H723 - Prepare for being able to build for using non-internal-flash
config storage.
* STM32H723 - Prepare for using non-default strings.
* STM32H723 - Disable 'use custom defaults' when using USE_EXST.
* STM32H723 - Disable CUSTOM_DEFAULTS_EXTENDED when EXST is used.
* OCTOSPI - Add initialisation code.
* Add support for RAM_CODE.
* STM32H730 - Add support for RAM_CODE via the .ram_code attribute.
* OCTOSPI - Proof-of-concept for disabling/enabling memory mapped mode on
a running system.
NOTE: The HAL libs are compiled into a memory mapped region, and this cannot be used for OctoSPI access when memory mapped mode is disabled.
* OCTOSPI - Drop HAL support after determining that it's not suitable for
the memory mapped flash use-case.
* OCTOSPI - Sometimes, when disabling memory map mode, the abort fails.
Handle this by disabling the OSPI peripheral first and re-enabling it
afterwards.
* SD/FLASH - Update comments regarding possible solutions to the catch-22
issue with SD/SPI/QUADSPI/OCTOSPI pin configurations.
* OCTOSPI - Use device instance directly.
* OCTOSPI - Prepare W25Q flash driver for octospi support.
* OCTOSPI - Add octospi flash detection.
Note: The method to detect flash chips is similar to the method used for
QUADSPI and as such the code was used as a base. However the initial
OCTOSPI implementation doesn't support the non-memory-mapped use-case so
the un-tested code is gated with `USE_OCTOSPI_EXPERIMENTAL`.
The key differences are:
* uses octospi api not quadspi api.
* flash chip clock speeds should not be changed for memory-mapped flash
chips, bootloader already set it correctly.
* only certain flash chips are capable of memory mapped support.
* W25Q - Ensure w25q128fv_readRegister returns 0 if the receive fails.
* OCTOSPI - Implement octoSpiTransmit1LINE, octoSpiReceive1LINE and
octoSpiReceive4LINES.
* OCTOSPI - Specify device from init.
* OCTOSPI - More fixes and work on init.
Current status is that memory mapped mode is disabled and flash chip is
detected, but w25q128fv_detect should not be calling w25q128fv_reset.
* FLASH - Add comment regarding wasted flash space on targets that only
use one bus type for the flash chip.
* FLASH - Split `detect` into `identify` and `configure`.
* OCTOSPI - Extract flashMemoryMappedModeEnable/Disable to separate
methods.
* FLASH - Reduce size of targets that don't support the use of multiple
flash interfaces.
* Single-flash-chip targets usually only support one type of io
interface.
* Without this, compiler warnings are generated in `flashSpiInit` for
targets that only use flash chip drivers that support quadspi or octospi
that don't even use SPI for flash.
* FLASH - Use MMFLASH_CODE/DATA to conditionally move code/data to RAM.
Only targets compiled to support memory mapped flash chips need the some
specific code in RAM. Otherwise the code/data should be in the normal
linker section.
* FLASH - W25Q Prepare for memory mapped flash usage.
* Wait/Delay functions must work with interrupts disabled.
* Code used for reading/writing must run from RAM.
* OCTOSPI - Implement remaining required methods.
* OCTOSPI - Fixes for earlier code (not last commit).
* FLASH - W25Q update timeout values from Datasheet Rev L.
* FLASH - Prepare flash driver for use when memory mapped flash is
disabled.
* System - Prepare microsISR for use when memory mapped flash is disabled.
* FLASH - Add support for CONFIG_IN_MEMORY_MAPPED_FLASH.
* Flash - Fix incorrect gating on cli flash commands.
When compiling with USE_FLASH_CHIP and without USE_FLASHFS there were
compiler warnings.
* MMFLASH - Fix release-mode build.
* FLASH - Allow SPI pins to be reconfigured when using CONFIG_IN_MEMORY_MAPPED_FLASH.
MMFLASH only works via QuadSPI/OctoSPI peripherals.
* EXST - Disable the 2GB workaround which is now causing a different
error.
The error we get with 'remove-section' enabled is:
"error in private header data: sorry, cannot handle this file". The
cause of this new error in the objcopy codebase is an out of memory
condition, somehow the 2GB files and this error are related but the root
cause of both is still unknown.
* OCTOSPI - Add support for STM32H723.
* STM32H723 - Add linker scripts for EXST usage.
* NucleoH723ZG - Add build script to demonstrate OCTOSPI and Memory Mapped
flash support.
* FLASH - WUse the size in bits to set the size of the buffer.
* FLASH - Fix typo in W25N driver defines.
Was using W28N instead of W25N
* OCTOSPI - Fix missing semilcolon when compiling without
USE_FLASH_MEMORY_MAPPED.
* OCTPSPI - Fix missing call to 'memoryMappedModeInit'.
* SPRacingH7RF - Add example build script to allow for testing prior to
unified target / cloud-build support.
Call targetConfiguration() once before config is loaded and again afterwards in case the config needs to be changed to load from SD card etc
Drop SPI clock during binding
Remove debug
Add per device SPI DMA enable
Fix sdioPinConfigure() declaration warning
Reduce clock speed during SPI RX initialisation
* Enable QSPI flash chip drivers on NUCLEOH743 for some CI visibility.
* Rework QSPI flash detection.
* Supports chips that are in QSPI mode by default (factory).
* Supports chips that are in QSPI mode due to configuration (e.g. Quad
Enable in Non-Volatile on W25Q128FV)
* Supports chips that need 8 dummy bytes (e.g. W25Q128JV) and ones that
don't (e.g. W25N01G)
Flash operations can specify how long to wait before the next operation
is issued.
Prior to this the amount of time waited, and when, was wrong.
e.g.
m25p16_eraseCompletely - possibly waits ages TO START, starts, exit.
m25p16_pageProgramContinue - waits DEFAULT_TIMEOUT_MILLIS to START,
starts, exits.
m25p16_pageProgramContinue would fail to write to the flash as the
device was still busy erasing and didn't wait long enough.
what happens now is:
m25p16_eraseCompletely - waits using the current timeout, starts, sets
timeout to be `now + BULK_ERASE_TIMEOUT_MILLIS`, exits.
m25p16_pageProgramContinue - waits using the current
`BULK_ERASE_TIMEOUT_MILLIS`, starts, exists, sets timeout to be `now +
DEFAULT_TIMEOUT_MILLIS`.
Since the timeout is stored in the flashDevice_t the solution also works
for multi-die devices which use an instance of flashDevice_t for each
die.
If there is only a single partition, its type is FLASHFS, and it uses the entire geometry, then perform a full rather than a sector-based erase.
The full erase is significantly faster than erasing by individual sectors.
- Update flash w25n01g driver to support QuadSPI.
- flash_m25p16 update for QuadSPI support
- w25m driver update for QuadSPI support
- Use 100Mhz (ULTRAFAST) clock for QuadSPI w25n01
- Conditionalize QUADSPI code in w25n01g driver
- Use handle instead of handle_u
