| Message ID | 20241023-imx-ele-ocotp-fixes-v1-0-4adc00ce288f@pengutronix.de (mailing list archive) |
|---|---|
| Headers | show |
| Series | nvmem: imx-ocotp-ele: fix reading from ELE OCOTP | expand |
Hi Sascha, On Wed, Oct 23, 2024 at 5:12 AM Sascha Hauer <s.hauer@pengutronix.de> wrote: > > Commits b2ab0edaf484 ("arm64: dts: imx93: add nvmem property for fec1") > and 0d4fbaffbdca ("arm64: dts: imx93: add nvmem property for eqos") > introduced NVMEM cell bindings for reading MAC addresses from the ELE > OCOTP. This doesn't work as expected due to bugs in the driver: > > - imx_ocotp_reg_read() interprets the incoming offset as 32bit word > offset, but it really is in bytes which means the driver reads bogus > values whenever the offset is non zero > - imx_ocotp_reg_read() reads wrong results when the offset is not 32bit > word aligned > - MAC addresses are stored in reverse byte order in the ELE OCOTP, we > have to swap the order before passing them to the upper layer > > This likely went through unnoticed because the bootloader normally adds > the MAC addresses to the ethernet nodes and in this case they are > preferred over the NVMEM addresses. > > This series fixes these issues. Patches 2, 3, and 4 should have a Fixes tag and Cc: stable. Thanks
Commits b2ab0edaf484 ("arm64: dts: imx93: add nvmem property for fec1") and 0d4fbaffbdca ("arm64: dts: imx93: add nvmem property for eqos") introduced NVMEM cell bindings for reading MAC addresses from the ELE OCOTP. This doesn't work as expected due to bugs in the driver: - imx_ocotp_reg_read() interprets the incoming offset as 32bit word offset, but it really is in bytes which means the driver reads bogus values whenever the offset is non zero - imx_ocotp_reg_read() reads wrong results when the offset is not 32bit word aligned - MAC addresses are stored in reverse byte order in the ELE OCOTP, we have to swap the order before passing them to the upper layer This likely went through unnoticed because the bootloader normally adds the MAC addresses to the ethernet nodes and in this case they are preferred over the NVMEM addresses. This series fixes these issues. Sascha Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de> --- Sascha Hauer (4): nvmem: imx-ocotp-ele: simplify read beyond device check nvmem: imx-ocotp-ele: fix reading from non zero offset nvmem: imx-ocotp-ele: fix MAC address byte order nvmem: imx-ocotp-ele: set word length to 1 drivers/nvmem/imx-ocotp-ele.c | 38 +++++++++++++++++++++++++++++++------- 1 file changed, 31 insertions(+), 7 deletions(-) --- base-commit: c2ee9f594da826bea183ed14f2cc029c719bf4da change-id: 20241023-imx-ele-ocotp-fixes-f7b216171e1e Best regards,