A CXL driver fix making its way upstream [1] rejects a repeat write of an endpoint decoder to the region targetN attribute that already holds it. Without the fix, that write leaves the region half detached from the endpoint's ports, and a later write can crash the kernel.
Add a test case to backstop that fix: write target0 twice on a partially populated region and check that the second write fails and leaves the region's decoders attached. [1] https://lore.kernel.org/all/94e9157aac2e99df1251d733769d539b914fdefc.1790829690.git.alison.schofi...@intel.com/ Assisted-by: LLM Signed-off-by: Alison Schofield <[email protected]> --- test/cxl-region-sysfs.sh | 16 +++++++++++++--- 1 file changed, 13 insertions(+), 3 deletions(-) diff --git a/test/cxl-region-sysfs.sh b/test/cxl-region-sysfs.sh index db1a163efb30..957a1792fbd7 100644 --- a/test/cxl-region-sysfs.sh +++ b/test/cxl-region-sysfs.sh @@ -87,9 +87,19 @@ do select(.mode == \"pmem\") | .decoder") done -# attach all endpoint decoders to the region -pos=0 -for i in ${endpoint[@]} +# a repeat write of an attached endpoint decoder must fail and leave +# every decoder between it and the root attached +echo "${endpoint[0]}" > "/sys/bus/cxl/devices/$region/target0" +nr_attached=$(cat /sys/bus/cxl/devices/decoder*/region | grep -cx "$region") +if echo "${endpoint[0]}" > "/sys/bus/cxl/devices/$region/target0" 2>/dev/null; then + err "$LINENO: repeat write of ${endpoint[0]} to target0 succeeded" +fi +nr=$(cat /sys/bus/cxl/devices/decoder*/region | grep -cx "$region") +[ "$nr" -ne "$nr_attached" ] && err "$LINENO: $nr of $nr_attached decoders still attached" + +# attach the remaining endpoint decoders to the region +pos=1 +for i in ${endpoint[@]:1} do echo $i > /sys/bus/cxl/devices/$region/target$pos pos=$((pos+1)) -- 2.37.3
