NVMEM overview

This article introduces how NVMEM Linux® framework manages BSEC OTP data and how to read/write from/to it.

1 Framework purpose[edit]

The NVMEM Linux® framework provides a generic interface for the device non-volatile memory data such as:

  • OTP (one-time programmable) fuses

It offers kernel space and user space interfaces to read and/or write data such as analog calibration data or MAC address.

2 System overview[edit]

NVMEM sysfs interfaceNVMEM consumers interfaceBSEC internal peripheral
NVMEM system overview

2.1 Component description[edit]

  • NVMEM user (user space)

The user can use the NVMEM sysfs interface, from a user terminal or a custom application, to read/write data from/to NVMEM device(s) from user space.

  • NVMEM user (kernel space)

User drivers can use the NVMEM API to read/write data from/to NVMEM device(s) from kernel space (such as the analog calibration data used by an ADC driver).

  • NVMEM framework (kernel space)

The NVMEM core provides sysfs interface and NVMEM API. They can be used to implement NVMEM user and NVMEM controller drivers.

  • NVMEM drivers (kernel space)

Provider drivers such as BSEC Linux® driver that exposes OTP data to the core.

  • NVMEM hardware

NVMEM controller(s) such as the BSEC internal peripheral[1]

2.2 API description[edit]

The NVMEM kernel documentation[2] describes:

  • Kernel space API for NVMEM providers and NVMEM consumers.
  • Userspace binary interface (sysfs).

See also sysfs-bus-nvmem[3] ABI documentation.

3 Configuration[edit]

3.1 Kernel configuration[edit]

Activate NVMEM framework in the kernel configuration through the Linux® menuconfig tool, Menuconfig or how to configure kernel (CONFIG_NVMEM=y):

Device Drivers  --->
   [*] NVMEM Support  --->
      <*>   STMicroelectronics STM32 factory-programmed memory support

3.2 Device tree configuration[edit]

The NVMEM data device tree bindings describe:

  • The location of non-volatile memory data
  • The NVMEM data providers[4]
  • The NVMEM data consumers[5]

The BSEC internal peripheral[1] device tree bindings are explained in BSEC device tree configuration article.

4 How to use the framework[edit]

4.1 How to use NVMEM with sysfs interface[edit]

4.1.1 How to list NVMEM devices[edit]

The available NVMEM devices can be listed in sysfs:

# Example to list nvmem devices
 ls /sys/bus/nvmem/devices/

The data content of an NVMEM device can be dumped to a binary file, and then displayed.

4.1.2 How to read BSEC lower OTPs using NVMEM[edit]

The 32 lower OTPs can be read from non-secure when using either:

  • the trusted boot chain (using TF-A)
  • the basic boot chain (using U-Boot SPL)
# Example to read lower nvmem data content
 dd if=/sys/bus/nvmem/devices/stm32-romem0/nvmem of=/tmp/file bs=4 count=32
# Example to display nvmem data content
 hexdump -C -v /tmp/file

4.1.3 How to read BSEC upper OTPs using NVMEM[edit]

Info white.png Information
Only the 32 lower OTPs can be accessed when using the basic boot chain, as it doesn't implement secure services (CONFIG_HAVE_ARM_SMCCC). So this section concerns only the trusted boot chain (using TF-A) as SMC feature is available.

Default behavior for upper OTPS is normally restricted to security. If user needs more than the 32 lower OTPs, there is an exception management explained in BSEC device tree configuration.

It is then possible to access to some upper NVMEM information.

# Example to read the MAC address from upper OTP area, using secure services:
 dd if=/sys/bus/nvmem/devices/stm32-romem0/nvmem of=/tmp/file skip=57 bs=4 count=2 status=none
 hexdump -C -v /tmp/file
Info white.png Information
A dedicated chapter of the reference manual describes the OTP mapping.

4.1.4 How to write BSEC OTPs using NVMEM[edit]

Warning white.png Warning
The below examples show how to write data to an NVMEM device. This may cause unrecoverable damage to the STM32 device (for example when writing to an OTP area)
Info white.png Information
Note that lower OTPs are using 2:1 redundancy, so they can be written bit per bit, whereas upper OTPs only support one time 32-bit programming.

Whatever the boot chain, the full lower NVMEM data content can be written as follows (if we suppose it has been previously read as described above, and updated directly in /tmp/file):

# Example to write lower nvmem data content
 dd if=/tmp/file of=/sys/bus/nvmem/devices/stm32-romem0/nvmem bs=4 count=32

Only on Trusted boot chain, and under the condition the device tree authorizes it, an upper NVMEM data can be written.
Example of 32-bit data word writing (filling it with ones) in OTP n°95:

# Create a 4 bytes length file filled with ones, e.g. 0xffffffff)
# Then, write it (32-bits, e.g. 4bytes) to OTP data 95
 dd if=/dev/zero count=1 bs=4 | tr '\000' '\377' > file
 dd if=file bs=4 seek=95 of=/sys/bus/nvmem/devices/stm32-romem0/nvmem
Info white.png Information
When a new OTP value has been written using this SYSFS interface, it may be necessary to reboot the board before reading it back. The OTP value can't be read directly after a write because the OTP value is read in a shadow area not directly in the OTP area.

5 How to trace and debug the framework[edit]

5.1 How to trace[edit]

Ftrace can be used to trace the NVMEM framework:

 cd /sys/kernel/debug/tracing
 cat available_filter_functions | grep nvmem             # Show available filter functions

Enable the kernel function tracer, then start using nvmem and display the result:

 echo function > current_tracer
 echo "*nvmem*" > set_ftrace_filter                      # Trace all nvmem filter functions
 echo 1 > tracing_on                                     # start ftrace
 hexdump -C -v /sys/bus/nvmem/devices/stm32-romem0/nvmem # dump nvmem
00000000  17 00 00 00 01 80 00 00  00 00 00 00 00 00 00 00  |................|
 echo 0 > tracing_on                                     # stop ftrace
 cat trace
# tracer: function
#                              _-----=> irqs-off
#                             / _----=> need-resched
#                            | / _---=> hardirq/softirq
#                            || / _--=> preempt-depth
#                            ||| /     delay
#           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION
#              | |       |   ||||       |         |
         hexdump-478   [000] ....   423.502278: bin_attr_nvmem_read <-sysfs_kf_bin_read
         hexdump-478   [000] ....   423.502290: nvmem_reg_read <-bin_attr_nvmem_read
         hexdump-478   [000] ....   423.515804: bin_attr_nvmem_read <-sysfs_kf_bin_read

6 References[edit]