[BACKPORT] arch/arm/stm32h7: invalidate dcache after aligned SDMMC RX DMA#389
Merged
dakejahl merged 1 commit intoJul 24, 2026
Merged
Conversation
… DMA The aligned direct-DMA receive path only invalidated the destination buffer before the transfer in stm32_dmarecvsetup(). On the Cortex-M7 the cache can speculatively prefetch into that cacheable buffer between the pre-DMA invalidate and DMA completion, leaving stale lines that shadow the data just written by the IDMA, so the CPU reads a previously cached sector instead of the freshly received data. Invalidate again in stm32_recvdma() once the aligned transfer completes, before the buffer is consumed. The buffer and length are cache-line aligned on this path, so no adjacent memory is affected. This matches the STM32 AN4839 guidance that a cache invalidate is required after DMA completion and before the CPU reads the updated region, not only before the transfer starts. A related instance of the same "invalidate too early" defect on STM32H7 SPI DMA is tracked in apache/nuttx#11594. Root-caused on a PX4 FMUv6C (STM32H743) board where MAVLink ULog downloads were intermittently corrupted: forensic diffing showed corrupted windows were exactly 32 bytes (the D-cache line size), cache-line aligned, and byte-for-byte equal to the previous 512-byte SD sector cached in the FAT single-sector buffer. Disabling the D-cache made the corruption disappear, isolating the defect to cache coherency. After this fix, downloaded files matched the source file byte-for-byte (sha256 identical) across a 5.8 MB log spanning thousands of sectors. Backport of apache/nuttx#19506 (upstream fix, not yet merged at the time of this backport). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Signed-off-by: Yang-Rui Li <yang77567789@gmail.com>
Contributor
Author
|
For reviewers who'd like to independently verify the corruption pattern described above, I've attached
Running: reproduces the diffs referenced in the hex dump above — every difference falls in a 32-byte, cache-line-aligned window whose content matches |
dakejahl
pushed a commit
to PX4/PX4-Autopilot
that referenced
this pull request
Jul 24, 2026
See PX4/NuttX#389 Assisted-by: Claude:claude-sonnet-5 Signed-off-by: Yang-Rui Li <yang77567789@gmail.com>
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.
Backport of apache/nuttx#19506 (open, not yet merged upstream at the
time of this backport).
Fixes a D-cache coherency bug in the STM32H7 SDMMC aligned RX DMA
path: the aligned completion branch in stm32_recvdma() never
invalidated the destination buffer after DMA completion, only
before it started in stm32_dmarecvsetup(). On the Cortex-M7 the
cache can speculatively prefetch into that buffer between the
pre-DMA invalidate and DMA completion, leaving stale lines that
shadow the freshly-written data.
Reproduced and hardware-verified on a PX4 FMUv6C board: MAVLink
ULog downloads were corrupted in 32-byte, cache-line-aligned
windows that were byte-for-byte identical to the previous 512-byte
SD sector — the exact signature of a stale cache line. See the
upstream PR for full forensic detail and verification (including a
byte-for-byte identical 5.8 MB download after the fix).