Llext relocatable fixes - #11219
Draft
jsarha wants to merge 9 commits into
Draft
Llext relocatable fixes#11219jsarha wants to merge 9 commits into
jsarha wants to merge 9 commits into
Conversation
When CONFIG_LLEXT_TYPE_ELF_RELOCATABLE is active, bypass appending static address flags (-Ttext, --section-start, -Tdata) in the linker helper script. This keeps section base addresses at 0. Also adjust the offset calculator to avoid integer parsing errors when all section addresses are set to 0. Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
…table modules Implement page-level virtual memory mapping for LLEXT libraries by reserving VMA ranges from the single shared vpage allocator (zephyr/lib/vpage.c), the same allocator already used by the vregion pipeline-resource-management code. This avoids introducing a second, private virtual-address allocator/region for libraries and keeps all virtual page bookkeeping in one place. Compile section layout at load-time to allocate virtual addresses and rewrite section sh_addr headers in-place. This enables Zephyr LLEXT to naturally relocate references. Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
Enable CONFIG_LLEXT_EXPORT_BUILTINS_BY_SLID=y in llext_relocatable.conf to link relocatable LLEXT modules against build-time function signature hashing, providing load-time ABI mismatch protection. Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
- Fix off-by-one error in vpages_unreserve() when relocating last in-use element. - Use DIV_ROUND_UP() for page calculation. - Guard ALIGN_UP() on section alignment when sh_addralign <= 1. Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
Add doxygen documentation to llext functions. Signed-off-by: Jyri Sarha <jyri.sarha@linux.intel.com>
Add more comments to llext_manager_layout_sections() and llext_manager_layout_sections() to explain linking process in a bit more detail. Signed-off-by: Jyri Sarha <jyri.sarha@linux.intel.com>
Loading a relocatable (ET_REL) module could fault while copying sections into SRAM, or leave part of the image unmapped. Three separate problems in the same path, all coming from region sizes that do not describe the virtual addresses the module is actually placed at. 1. mctx->segment[] took its size straight from the llext region descriptor. llext_map_sections() builds those sizes from ELF *file* offsets, pads them by the region alignment, and only shifts sh_addr back for ET_DYN images. SOF pre-locates ET_REL modules at freshly allocated virtual addresses, so the file span says nothing about the VMA span. Regions overlapped each other and, for mfcc and mixin_mixout, reached a page past the module's own vpage reservation, re-mapping and re-permissioning a page belonging to another module. Measure the real extent of each region's non-detached sections instead, in the new llext_manager_region_extent(). 2. The destination bound passed to memcpy_s() was 'size - s_offset', mixing the region size with a file-offset delta. The inflated sizes from 1. hid this; with correct sizes it under-bounds the destination and memcpy_s() starts rejecting valid copies. Bound the copy by the space left in the mapped range instead, and log the section that failed rather than returning silently. 3. The layout page-aligns .bss into a page of its own, so it is normally not adjacent to .data. The merge only added bss_size to data_size and ignored the gap between them, so the mapping covered the .data page alone and the memset() that zeroes .bss faulted on an unmapped page. smart_amp hit this with 8 bytes of .data and 4 bytes of .bss a page apart. Map the range spanning both, which is what the comment on the .data mapping already promises. Tested on ACE30/PTL: MICSEL, PEAKVOL, MIXIN and SMATEST all load and map, and playback runs cleanly. Note that a matching mm fix is needed for the HPSRAM banks above the firmware image to be powered on at all.
The virtual page allocator is brought up by a SYS_INIT() handler, so it
starts out empty on every firmware boot, including the D0 restore that
follows DSP power gating. llext_manager_restore_from_dram() brings back
modules that are still resident at the addresses they were given before
gating, but never tells the allocator that those ranges are taken.
The first allocation after such a restore then hands out memory that a
restored module already occupies. On PTL the SRC module sits at the very
start of the region, so a DP module heap is given the same base and the
overlap is only caught when the module's memory domain is populated:
vpage_alloc ptr 0xa02b9000 pages 7 free 480/487
vregion_create: new at base 0xa02b9000 size 0x7000
check_add_partition: partition base a02b9000 (size 12288) overlaps
existing base a02b9000 (size 28672)
scheduler_dp_task_init: failed to add LLEXT to domain -22
The "free 480/487" above shows the allocator accounting for the seven
pages it just handed out and nothing else, while the five restored
extensions hold 102 pages. Reproduced by running playback and capture
concurrently, which makes the power gating cycle far more likely.
Add vpage_reserve_at() to claim a specific range and use it to account
for each restored module, so the allocator only offers addresses that
are genuinely free.
Signed-off-by: Jyri Sarha <jyri.sarha@linux.intel.com>
… relocatable modules
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes and additions to #11037