This tag contains the following bug fixes:
- Fix the dma address that is passed to dma_mmap_coherent. We passed
an address that includes an offset that is needed by our device and
that caused dma_mmap_coherent to do an errounous mapping.
- Fix the reset process in case failures happen during the reset process.
Without this fix, if the user would have asked to perform reset after
the previous reset failed he would get a kernel panic
- WA to prevent soft lockup BUG during unmap of host memory. In case of
tens of thousands of mappings, the unmapping can take a long time that
exceeds the soft lockup timeout. This WA adds a small sleep every 32K
page unmappings to prevent that.
habanalabs: prevent soft lockup during unmap
When using Deep learning framework such as tensorflow or pytorch, there
are tens of thousands of host memory mappings. When the user frees
all those mappings at the same time, the process of unmapping and
unpinning them can take a long time, which may cause a soft lockup
bug.
To prevent this, we need to free the core to do other things during
the unmapping process. For now, we chose to do it every 32K unmappings
(each unmap is a single 4K page).
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
3 files changed
