spi: spi-qcom-qspi: Fallback to PIO for xfers that aren't multiples of 4 bytes
authorDouglas Anderson <dianders@chromium.org>
Tue, 25 Jul 2023 18:02:26 +0000 (11:02 -0700)
committerMark Brown <broonie@kernel.org>
Wed, 26 Jul 2023 11:47:19 +0000 (12:47 +0100)
commit138d73b627c71bf2b2f61502dc6c1137b9656434
tree67c47d624e6f1a804614c255d6bc06308a3b9b54
parent916a4edf3daed845b1e5d6cf0578a7e43c6f520e
spi: spi-qcom-qspi: Fallback to PIO for xfers that aren't multiples of 4 bytes

The Qualcomm QSPI driver appears to require that any reads using DMA
are a mutliple of 4 bytes. If this isn't true then the controller will
clobber any extra bytes in memory following the last word. Let's
detect this and falback to PIO.

This fixes problems reported by slub_debug=FZPUA, which would complain
about "kmalloc Redzone overwritten". One such instance said:

  0xffffff80c29d541a-0xffffff80c29d541b @offset=21530. First byte 0x0 instead of 0xcc
  Allocated in mtd_kmalloc_up_to+0x98/0xac age=36 cpu=3 pid=6658

Tracing through what was happening I saw that, while we often did DMA
tranfers of 0x1000 bytes, sometimes we'd end up doing ones of 0x41a
bytes. Those 0x41a byte transfers were the problem.

NOTE: a future change will enable the SPI "mem ops" to help avoid this
case, but it still seems good to add the extra check in the transfer.

Fixes: b5762d95607e ("spi: spi-qcom-qspi: Add DMA mode support")
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Vijaya Krishna Nivarthi <quic_vnivarth@quicinc.com>
Reviewed-by: Bjorn Andersson <andersson@kernel.org>
Link: https://lore.kernel.org/r/20230725110226.1.Ia2f980fc7cd0b831e633391f0bb1272914d8f381@changeid
Signed-off-by: Mark Brown <broonie@kernel.org>
drivers/spi/spi-qcom-qspi.c