As part of inter-procedural optimizations (IPA), the compiler may
perform tasks such as constant propagation for functions. This involves
generating a specialized version of a given function with a new symbol
name that includes a suffix. For example, during constant propagation,
the compiler might create a specialized version named
`spiflash_start_core.constprop.0` for the `spiflash_start_core`
function. Additionally, the compiler may generate multiple clones of a
single function. Currently, when ldgen performs symbol placement, it
does not account for these compiler-generated functions, leading to
their incorrect or unexpected placement in memory (markers).
Consider a linker fragment with:
```
[mapping:spi_flash]
archive: libspi_flash.a
entries:
esp_flash_api: spiflash_start_core (noflash)
```
The `spiflash_start_core` function should be placed in IRAM. However,
the compiler might generate an optimized version of this function with a
`.constprop.0` suffix, resulting in a
`.text.spiflash_start_core.constprop.0` input section. Currently, ldgen
does not handle this situation, leading to misplaced symbols.
Since `.` is not allowed in C identifiers, it should be safe to consider
all input sections for a symbol with any `.` suffix as representing that
symbol. This means considering the symbol suffixes should not cause any
ambiguity.
This change automatically places all input sections, including those
with possible suffixes for a given symbol, into the specified memory. In
other words, specifying a function name like `spiflash_start_core` in a
linker fragment automatically includes input section names matching
`spiflash_start_core(\..*)?$`.
Signed-off-by: Frantisek Hrbata <frantisek.hrbata@espressif.com>
Changes generated by pre-commit related to ruff-format and an update to
the copyright date.
Signed-off-by: Frantisek Hrbata <frantisek.hrbata@espressif.com>
Add the `--debug` option to print the entity tree to stdout. This can be
useful for viewing the entire entity tree state to identify potential
issues or incorrect placements.
Signed-off-by: Frantisek Hrbata <frantisek.hrbata@espressif.com>
Introduce a new `--mutable-libraries-file` option that accepts a file
containing the filenames of mutable libraries, each listed on a separate
line. Mutable libraries are component libraries expected to change
during development. In contrast, immutable component libraries are not
expected to change. In the generated linker script, the input sections
of mutable libraries are grouped together rather than being mixed with
those of immutable libraries. The goal is to create large continuous
areas in the ELF file's output sections that remain unchanged for
immutable libraries during application recompilation, allowing these
areas to be skipped during flashing.
The build system identifies the mutable libraries and passes them to
ldgen using the `--mutable-libraries-file` option. It maintains
information about component sources, one of which is
`project_components`. This source type identifies components that are
directly related to the project being developed and are very likely to
change.
Mappings for mutable libraries are explicitly created for all sections
in the default scheme. This happens before the entity
`(archive:object_file:symbol/input_section)` node tree with placements is
generated and is equivalent to having these mappings in the mapping
linker fragment. All placements for mappings, whether newly added or
already existing as defined in linker fragments, associated with mutable
libraries are flagged as `mutable` in the entity node tree. This flag
ensures that these placements are included in the final linker script.
Currently, ldgen only emits placements that are either significant or
forced. A placement is considered significant if, for example, it is not
already covered by a placement in parent node. For instance, `*(.iram1
.iram1.*)` placement already includes `*libapp_trace.a:(.iram1
.iram1.*)`, so the latter is not emitted by default. The `mutable` flag
ensures that placements for mutable libraries are emitted in the linker
script and placed at dedicated location.
The locations where placements for mutable libraries are specified in
the linker script are identified by a new `mutable` marker, for example,
`mutable[flash_text]`. The placements for immutable libraries remain in
the existing `mapping` marker, for example, `mapping[flash_text]`. The
`mutable` marker for each target is placed after the `mapping` marker.
Signed-off-by: Frantisek Hrbata <frantisek.hrbata@espressif.com>
When a symbol needs to be placed to a different target than the one
designated for the object file, the object file is expanded, which
includes the following steps:
1. Creating a new placement for the symbol's input section with the
specified target.
2. Excluding the object placement from the orignal target.
3. Creating a new intermediate placement for the object for the original
target, where its input sections are expanded, excluding the input
section for the symbol.
Let's illustrate the object expansion process with the following example:
[sections:rodata]
entries:
.rodata+
.sdata2+
.srodata+
[scheme:default]
entries:
text -> flash_text
rodata -> flash_rodata
[scheme:noflash]
entries:
text -> iram0_text
rodata -> dram0_data
[mapping:soc_pm]
archive: libsoc.a
entries:
gpio_periph: GPIO_HOLD_MASK (noflash)
gpio_periph section headers:
[Nr] Name Type Addr Off Size ES Flg Lk Inf Al
[ 0] NULL 00000000 000000 000000 00 0 0 0
[ 1] .text PROGBITS 00000000 000034 000000 00 AX 0 0 2
[ 2] .data PROGBITS 00000000 000034 000000 00 WA 0 0 1
[ 3] .bss NOBITS 00000000 000034 000000 00 WA 0 0 1
[ 4] .rodata.GPIO_HOLD_MASK PROGBITS 00000000 000034 000058 00 A 0 0 4
[ 5] .rodata.GPIO_PIN_MUX_REG PROGBITS 00000000 00008c 000058 00 A 0 0 4
[ 6] .debug_info PROGBITS 00000000 0000e4 0000d8 00 0 0 1
[ 7] .rela.debug_info RELA 00000000 0009d4 000108 0c I 16 6 4
[ 8] .debug_abbrev PROGBITS 00000000 0001bc 000070 00 0 0 1
[ 9] .debug_aranges PROGBITS 00000000 00022c 000018 00 0 0 1
[10] .rela.debug_aranges RELA 00000000 000adc 00000c 0c I 16 9 4
[11] .debug_line PROGBITS 00000000 000244 0001ab 00 0 0 1
[12] .debug_str PROGBITS 00000000 0003ef 00022d 01 MS 0 0 1
[13] .comment PROGBITS 00000000 00061c 000030 01 MS 0 0 1
[14] .note.GNU-stack PROGBITS 00000000 00064c 000000 00 0 0 1
[15] .riscv.attributes RISCV_ATTRIBUTES 00000000 00064c 000044 00 0 0 1
[16] .symtab SYMTAB 00000000 000690 000260 10 17 36 4
[17] .strtab STRTAB 00000000 0008f0 0000e1 00 0 0 1
[18] .shstrtab STRTAB 00000000 000ae8 0000d1 00 0 0 1
1. Creating a new placement
.dram0.data :
{
*libsoc.a:gpio_periph.*(.rodata.GPIO_HOLD_MASK .sdata2.GPIO_HOLD_MASK .srodata.GPIO_HOLD_MASK)
}
2. Excluding the object placement
.flash.rodata :
{
*(EXCLUDE_FILE(*libsoc.a:gpio_periph.*) .rodata.* ...)
}
3. Creating a new intermediate placement
.flash.rodata :
{
*libsoc.a:gpio_periph.*(.rodata.GPIO_PIN_MUX_REG)
}
Now, let's do the same, but also move GPIO_PIN_MUX_REG to noflash with an updated mapping.
[mapping:soc_pm]
archive: libsoc.a
entries:
gpio_periph: GPIO_HOLD_MASK (noflash)
gpio_periph: GPIO_PIN_MUX_REG (noflash)
1. Creating a new placement
.dram0.data :
{
*libsoc.a:gpio_periph.*(.rodata.GPIO_HOLD_MASK .sdata2.GPIO_HOLD_MASK .srodata.GPIO_HOLD_MASK)
*libsoc.a:gpio_periph.*(.rodata.GPIO_PIN_MUX_REG .sdata2.GPIO_PIN_MUX_REG
.srodata.GPIO_PIN_MUX_REG)
}
2. Excluding the object placement
.flash.rodata :
{
*(EXCLUDE_FILE(*libsoc.a:gpio_periph.*) .rodata.* ...)
}
3. Creating a new intermediate placement
.flash.rodata :
{
*libsoc.a:gpio_periph.*
}
The *libsoc.a:gpio_periph.* entity in step 3 no longer has input
sections, as there are no remaining .rodata input sections in the object
file. The linker behavior for this mapping is to include all object
input sections that have not yet been placed as described in
https://sourceware.org/binutils/docs/ld.html#Input-Section-Basics
"If you use a file name without a list of sections, then all sections in
the input file will be included in the output section. This is not
commonly done, but it may by useful on occasion."
The map file for such mapping now contains following input sections
.flash.rodata 0x3c0a0120 0x19b34
*libsoc.a:gpio_periph.*()
.debug_info 0x3c0b95bf 0xd8 esp-idf/soc/libsoc.a(gpio_periph.c.obj)
.debug_abbrev 0x3c0b9697 0x70 esp-idf/soc/libsoc.a(gpio_periph.c.obj)
.debug_aranges
0x3c0b9707 0x18 esp-idf/soc/libsoc.a(gpio_periph.c.obj)
.debug_line 0x3c0b971f 0x1ab esp-idf/soc/libsoc.a(gpio_periph.c.obj)
.debug_str 0x3c0b98ca 0x21a esp-idf/soc/libsoc.a(gpio_periph.c.obj)
0x22d (size before relaxing)
.comment 0x3c0b9ae4 0x30 esp-idf/soc/libsoc.a(gpio_periph.c.obj)
.note.GNU-stack
0x3c0b9ae4 0x0 esp-idf/soc/libsoc.a(gpio_periph.c.obj)
.riscv.attributes
0x3c0b9ae4 0x44 esp-idf/soc/libsoc.a(gpio_periph.c.obj)
This is incorrect, and such intermediate placement should not be
generated. This type of placement can be recognized because it is not
explicitly defined in the mapping and lacks input sections. We can
identify this in the significant function and prevent issuing commands
for such placement.
Signed-off-by: Frantisek Hrbata <frantisek.hrbata@espressif.com>
Currently, the `SORT` flag mandates the inclusion of at least the
`sort_by_first` argument in the grammar, despite the documentation[1]
indicating that `SORT` can be utilized without any arguments, defaulting
to sorting input sections by name. Fix this by modifying the grammar
to allow a default `SORT` and update a test accordingly.
[1] https://docs.espressif.com/projects/esp-idf/en/stable/esp32/api-guides/
linker-script-generation.html
Signed-off-by: Frantisek Hrbata <frantisek.hrbata@espressif.com>
This commit introduce SOC_MEM_NON_CONTIGUOUS_SRAM flag (that enebled for
esp32p4). If SOC_MEM_NON_CONTIGUOUS_SRAM is enabled:
- LDFLAGS+=--enable-non-contiguous-regions
- ldgen.py replaces "arrays[*]" from sections.ld.in with objects under
SURROUND keyword. (e.g. from linker.lf: data -> dram0_data SURROUND(foo))
- "mapping[*]" - refers to all other data
If SOC_MEM_NON_CONTIGUOUS_SRAM, sections.ld.in file should contain at
least one block of code like this (otherwise it does not make sense):
.dram0.bss (NOLOAD) :
{
arrays[dram0_bss]
mapping[dram0_bss]
} > sram_low
.dram1.bss (NOLOAD) :
{
/* do not place here arrays[dram0_bss] because it may be splited
* between segments */
mapping[dram0_bss]
} > sram_high
Currently, only `.o`, `.*.obj` and `.obj` patterns
are taken into account. It would be great to have
object files with the `.*.o` extension pattern
(e.g. `file.cpp.o`) also processed as they're quite
widespread in third-party integrations.
1. Rename ESP_SLEEP_DEEP_SLEEP_WAKEUP_DELAY to ESP_SLEEP_WAIT_FLASH_READY_EXTRA_DELAY
2. Set ESP_SLEEP_WAIT_FLASH_READY_EXTRA_DELAY visible for all targets
Moved the following kconfig options out of the target component:
* CONFIG_ESP*_DEFAULT_CPU_FREQ* -> esp_system
* ESP*_REV_MIN -> esp_hw_support
* ESP*_TIME_SYSCALL -> newlib
* ESP*_RTC_* -> esp_hw_support
Where applicable these target specific konfig names were merged into
a single common config, e.g;
CONFIG_ESP*_DEFAULT_CPU_FREQ -> CONFIG_ESP_DEFAULT_CPU_FREQ_MHZ
Moved the following kconfig options out of the target component:
* ESP32_X_BROWNOUT_* -> esp_system
* ESP32_X_DEBUG_OCDAWARE -> esp_system
* APP_NO_BLOBS -> build type (main kconfig)
This commit removes the usage of all legacy FreeRTOS data types that
are exposed via configENABLE_BACKWARD_COMPATIBILITY. Legacy types can
still be used by enabling CONFIG_FREERTOS_ENABLE_BACKWARD_COMPATIBILITY.
harshal.patil
Chen Chen
Frantisek Hrbata
C.S.M
Roland Dobai
Peter Dragun
Jan Beran
Alexey Lapshin
wanckl
Marek Fiala
David Cermak
Omar Chebib
Jakub Kocka
Jakob Hasse
Valerii Koval
wuzhenghui
You Wei
radim.karnis
Marius Vikhammer
Kapil Gupta
simon.chupin
jingli
Darian Leung
Fu Hanxi
Ivan Grokhotkov