summaryrefslogtreecommitdiff
path: root/tools/perf/scripts/python/bin/export-to-sqlite-report
diff options
context:
space:
mode:
authorRaman Varabets <kernel-linux-20260610-80b7ab08@raman.v1.sg>2026-06-10 21:41:19 +0800
committerJiri Kosina <jkosina@suse.com>2026-08-03 19:15:48 +0200
commit2d68d67151eb2a22af7afdb75e1bc5097b082b9d (patch)
tree874391154637aac0ab8518187308e61adb2da04a /tools/perf/scripts/python/bin/export-to-sqlite-report
parentb6eb022890c78285f55381589c1536bd66b8eaeb (diff)
HID: ft260: fix SMBus block read protocol handling
For I2C_SMBUS_BLOCK_DATA reads, ft260_smbus_xfer() passed data->block[0] + 1 as the read length. But on a block read the byte count is supplied by the slave as the first byte of the response; data->block[0] is not initialized by the caller, so the transfer length was taken from stale buffer contents, and the count byte the slave did return was stored without any validation. Implement the SMBus 2.0 block read protocol properly: read the count byte first with a repeated START and no STOP, validate it against I2C_SMBUS_BLOCK_MAX (resetting the bus and returning -EPROTO on a bogus count), then read exactly that many data bytes and finish the transaction with STOP. This keeps the whole sequence within a single I2C transaction: S Addr+Wr A Reg A Sr Addr+Rd A Count A Data... P To support issuing the two reads as one transaction, teach ft260_i2c_read() to honor the caller's flags instead of always forcing a START and unconditionally appending STOP to the last chunk: START is only emitted if requested, and STOP is appended to the final chunk only when the caller asked for it. Signed-off-by: Raman Varabets <kernel-linux-20260610-80b7ab08@raman.v1.sg> Reviewed-by: Michael Zaidman <michaelz@xsightlabs.com> Reviewed-by: Michael Zaidman <michael.zaidman@gmail.com> Signed-off-by: Jiri Kosina <jkosina@suse.com>
Diffstat (limited to 'tools/perf/scripts/python/bin/export-to-sqlite-report')
0 files changed, 0 insertions, 0 deletions