Mirrors/esp-idf
1
0
Fork 0
mirror of https://github.com/espressif/esp-idf.git synced 2026-10-02 11:10:54 +03:00
Code Issues Packages Projects Releases Wiki Activity

fix(esp_system): fixes build failure with XIP PSRAM + BSS/NoInit in external memory

Browse Source 
Declare ext_ram_xip_seg as a distinct region overlapping drom_seg at the
same origin (the esp32s3/c5/c61/h4 pattern) instead of aliasing drom_seg,
so .ext_ram.dummy has its own location counter and .ext_ram.bss reclaims
the VMA space of NOLOAD rodata (.rodata_wlog_*), saving up to one MMU
page of PSRAM on esp32s31. No layout change on esp32p4.

Closes https://github.com/espressif/esp-idf/issues/18791
Related https://github.com/espressif/esp-idf/issues/14992
This commit is contained in:
Mahavir Jain
2026-07-22 16:39:25 +05:30
parent cbe15485ce
commit c566ace1f6
3 changed files with 27 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
components/esp_system/ld/ld.ext_ram.sections
+7
View File
@@ -9,6 +9,13 @@
.ext_ram.dummy (NOLOAD) :
{
HIDDEN(_ext_ram_on_same_bus = ORIGIN(ext_ram_seg) == ORIGIN(flash_rodata_seg));
/**
* This must be the first section placed in `ext_ram_seg`, and `ext_ram_seg` must be
* a distinct region (not a REGION_ALIAS of `flash_rodata_seg`) so it has its own
* location counter. Otherwise NOLOAD rodata (e.g. .rodata_wlog_*) placed after
* `_rodata_reserved_end` leaves the counter past the skip target, and moving it
* backwards here is a hard linker error.
*/
. = _ext_ram_on_same_bus ? ORIGIN(ext_ram_seg) + (_rodata_reserved_end - _flash_rodata_dummy_start) : 0;
. = ALIGN(_ext_ram_on_same_bus ? _esp_mmu_page_size : 0);
} > ext_ram_seg