Repository navigation
API System
The System module starts and stops the library, connects modules into a pipeline with bind calls, hands out
timestamps and gives access to a few system facilities. Every program calls IMP_System_Init once before it
creates FrameSource channels, encoder channels, OSD or IVS groups, and IMP_System_Exit at the end.
Pipeline cells: a cell (IMPCell) names one output of one module: device (DEV_ID_FS, DEV_ID_ENC,
DEV_ID_OSD, DEV_ID_IVS, ...), group (or channel) and output index. IMP_System_Bind(src, dst) connects the output
of src to the input of dst, for example FrameSource channel 0 to encoder group 0. OpenIMP records the bind;
the consumer (the encoder, the IVS group) pulls frames from the source when it starts, so bind before the consumer
starts. At most 16 binds exist at a time.
Extensions in this module (OpenIMP only): OpenIMP_* functions that belong to FrameSource, such as the
fallback test frame, are described on the FrameSource page and in
Beyond the Vendor.
Status columns: T10 T20 T21 T23 T30 T31 T40 T41. As of 2026-11-16. Part of the API reference.
- Start and stop (2)
- Binding modules (3)
- Time and identification (4)
- Register access (2)
- Memory pools (16)
- Logging (1)
- Decoder (T10, T20, T21, T23, T30) (8)
- System utilities (SU) (19)
Status symbols are those of the feature matrix (API coverage tables):
-
✅works, device-tested -
✅+works, device-tested, better than the vendor -
✅rtAPI round trip on device, no visible effect expected -
✅rworks: called by the streamer on fleet cameras of this SoC that passed the fleet soak -
✅○vendor no-op, OpenIMP matches that -
✅?counted as done by user decision, not verified on hardware (the lab board lacks the hardware) -
🔧himplemented, host tests only (no camera run yet) -
🔧areally connected per static audit, no device test tracked -
🔧+API round trip in QEMU, no camera -
⚠cvalue only stored and read back, nothing applied -
⚠sreturns 0 without any effect -
❌eexported but fails or has a known defect -
❌not exported by OpenIMP -
·function does not exist on this SoC (not in that SoC's vendor API; not a gap) -
?state not determined -
📋function not provided by the SoC build yet (T32/T33 have a QEMU-only build)
int IMP_System_Init(void);Initialises the library, starts the system clock at zero and prepares the FrameSource, IVS, OSD, encoder and framebuffer parts.
Returns: 0 on success and when already initialised; negative when one of the parts fails (those already started are shut down again).
Call it once, before any other module call.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest t10wrench 2026-11-03 .31
- T20: apitest agg63 2026-10-31 .107
- T21: apitest agg63 2026-10-31 .24
- T23: apitest agg62 2026-10-31 .30
- T30: apitest t30b30 2026-11-05 .231
- T31: apitest agg61 2026-10-31 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Used by: prudynt, raptor, timps.
int IMP_System_Exit(void);Shuts the parts down in the reverse order.
Returns: 0, also when not initialised.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest t10wrench 2026-11-03 .31
- T20: apitest agg63 2026-10-31 .107
- T21: apitest agg63 2026-10-31 .24
- T23: apitest agg62 2026-10-31 .30
- T30: apitest t30b30 2026-11-05 .231
- T31: apitest agg61 2026-10-31 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Used by: prudynt, raptor, timps.
int IMP_System_Bind(IMPCell *srcCell, IMPCell *dstCell);Connects the output cell src to the input cell dst.
Parameters
-
src: Source cell. -
dst: Destination cell.
Returns: 0 on success and when the same bind already exists; -1 for a NULL cell or when all bind slots are used.
Typical binds are FrameSource -> encoder, FrameSource -> IVS, FrameSource -> OSD -> encoder. For an OSD group in the middle use IMP_OSD_AttachToGroup.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest agg63 2026-10-31 .31
- T20: apitest agg63 2026-10-31 .107
- T21: apitest agg63 2026-10-31 .24
- T23: apitest rc3gaps2 2026-10-27 .30
- T30: apitest T30X lab 2026-11-03 .231
- T31: apitest rc3gaps2 2026-10-27 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Used by: prudynt, raptor, timps.
See also: IMP_System_UnBind, IMP_System_GetBindbyDest
int IMP_System_UnBind(IMPCell *srcCell, IMPCell *dstCell);Removes a bind.
Returns: 0 on success; -1 for a NULL cell or when no such bind exists.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest agg63 2026-10-31 .31
- T20: apitest agg63 2026-10-31 .107
- T21: apitest agg63 2026-10-31 .24
- T23: apitest agg47 2026-10-26 .30
- T30: apitest T30X lab 2026-11-03 .231
- T31: apitest rc3gaps2 2026-10-27 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Used by: prudynt, raptor, timps.
int IMP_System_GetBindbyDest(IMPCell *dstCell, IMPCell *srcCell);Finds the source cell that feeds a destination cell.
Parameters
-
dst: Destination cell. -
src: Receives the source cell.
Returns: 0 on success; -1 for a NULL pointer or when nothing is bound to dst.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest agg63 2026-10-31 .31
- T20: apitest agg63 2026-10-31 .107
- T21: apitest agg63 2026-10-31 .24
- T23: apitest agg47 2026-10-26 .30
- T30: apitest T30X lab 2026-11-03 .231
- T31: apitest rc3gaps2 2026-10-27 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Used by: raptor.
int IMP_System_GetVersion(IMPVersion *pstVersion);Writes the version text of the library into a buffer.
Parameters
-
version: Buffer of at least 64 bytes; it is zeroed and the text is copied in.
Returns: -1 for NULL.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest rc3gaps2 2026-10-27 .31
- T20: apitest rc3gaps2 2026-10-27 .107
- T21: apitest rc3gaps2 2026-10-27 .24
- T23: apitest rc3gaps2 2026-10-27 .30
- T30: apitest t30lab 2026-11-03 .231
- T31: apitest rc3gaps2 2026-10-27 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Used by: prudynt, raptor, timps.
Difference from the vendor library: The text is OpenIMP's own, for example libimp.so T31 openimp; the vendor library returns its release number.
const char *IMP_System_GetCPUInfo(void);Returns the name of the SoC variant, for example T31X, T41NQ or T40-XP.
Returns: A pointer to a constant string, never NULL.
The variant is read from the SoC when possible; otherwise a default of the family is returned.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest rc3gaps2 2026-10-27 .31
- T20: apitest rc3gaps2 2026-10-27 .107
- T21: apitest rc3gaps2 2026-10-27 .24
- T23: apitest rc3gaps2 2026-10-27 .30
- T30: apitest t30lab 2026-11-03 .231
- T31: apitest rc3gaps2 2026-10-27 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Used by: prudynt, raptor.
uint64_t IMP_System_GetTimeStamp(void);Returns the system time in microseconds since System_Init (or since the last RebaseTimeStamp), taken from a monotonic clock.
Returns: The time, or -1 when the clock cannot be read.
FrameSource frame timestamps are on this timeline.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest rc3gaps2 2026-10-27 .31
- T20: apitest rc3gaps2 2026-10-27 .107
- T21: apitest rc3gaps2 2026-10-27 .24
- T23: apitest rc3gaps2 2026-10-27 .30
- T30: apitest t30lab 2026-11-03 .231
- T31: apitest rc3gaps2 2026-10-27 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Used by: raptor, timps.
int IMP_System_RebaseTimeStamp(uint64_t basets);Moves the system timeline so that the current moment has the given value.
Parameters
-
timestamp: New "now" in microseconds.
Returns: 0, or -1 when the clock cannot be read.
Use it to align the timeline with an external clock (for example the first RTP timestamp).
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest rc3gaps2 2026-10-27 .31
- T20: apitest rc3gaps2 2026-10-27 .107
- T21: apitest rc3gaps2 2026-10-27 .24
- T23: apitest rc3gaps2 2026-10-27 .30
- T30: apitest t30lab 2026-11-03 .231
- T31: apitest rc3gaps2 2026-10-27 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Used by: prudynt, raptor.
Direct access to hardware registers by physical address through /dev/mem. For debugging and bring-up, not for streamers.
uint32_t IMP_System_ReadReg32(uint32_t addr);Reads a 32-bit hardware register at a physical address.
Parameters
-
addr: Physical address, must be 4-byte aligned.
Returns: The register value; 0 when the address is not aligned or /dev/mem cannot be used (a failure cannot be told from a register that reads 0).
The access maps one page of /dev/mem per call. It was tested by reading and writing a scratch register of the UART.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest rc3gaps2 2026-10-27 .31
- T20: apitest rc3gaps2 2026-10-27 .107
- T21: apitest rc3gaps2 2026-10-27 .24
- T23: apitest rc3gaps2 2026-10-27 .30
- T30: apitest t30lab 2026-11-03 .231
- T31: apitest rc3gaps2 2026-10-27 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Used by: raptor.
void IMP_System_WriteReg32(uint32_t addr, uint32_t val);Writes a 32-bit hardware register at a physical address.
Parameters
-
addr: Physical address, 4-byte aligned. -
value: Value to write.
Returns: no value.
Writing the wrong register can hang the SoC.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest rc3gaps2 2026-10-27 .31
- T20: apitest rc3gaps2 2026-10-27 .107
- T21: apitest rc3gaps2 2026-10-27 .24
- T23: apitest rc3gaps2 2026-10-27 .30
- T30: apitest t30lab 2026-11-03 .231
- T31: apitest rc3gaps2 2026-10-27 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Used by: raptor.
A pool reserves a block of media memory under an id; channels bound to it with IMP_FrameSource_SetPool or
IMP_Encoder_SetPool take their buffers from it. T41 carves a real contiguous pool out of the rmem arena. On the
other SoCs OpenIMP allocates every channel from its single rmem arena anyway, so a requested pool is validated and
recorded by name but reserves nothing (reserving memory that nothing draws from would only shrink the arena).
int IMP_System_MemPoolRequest(int poolId, size_t size, const char *name);Requests a memory pool under an id.
Parameters
-
poolId: Pool number, 0 to 31 on T31, T40 and T41. -
size: Bytes, above 0. -
name: Label for diagnostics.
Returns: 0 on success; -1 for an invalid id, a size of 0, a size above 2 GiB (T41), when the id is already requested (T31, T40, T41) or when the memory is not available (T41).
SoC: only in the vendor API of T23, T31, T40, T41
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | ✅ | · | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T23: apitest rc3gaps2 2026-10-27 .30
- T31: apitest rc3gaps2 2026-10-27 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Difference from the vendor library: T41 reserves a real block in rmem, as the vendor library does. T23, T31 and T40 only record the request. See Beyond the Vendor: Behaviour that differs from the vendor (better or deviating on purpose).
int IMP_System_MemPoolFree(int poolId);Releases a pool.
Returns: -1 for an invalid or not requested pool. On T41 also -1 while a buffer of the pool is still allocated; the channels bound to the pool are unbound and may be bound again with IMP_FrameSource_SetPool.
SoC: only in the vendor API of T31
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | · | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T31: apitest rc3gaps2 2026-10-27 .26
Difference from the vendor library: Beyond the vendor on T41: the vendor library keeps a pool until the program exits, OpenIMP lets the application release it. See Beyond the Vendor: Behaviour that differs from the vendor (better or deviating on purpose).
int IMP_MemPool_InitPool(int pool_id, uint32_t size, int flags); /* T21, T23, T30, T31 */
int IMP_MemPool_InitPool(int pool_id, uint32_t size, const char *name); /* T40, T41 */Low-level pool creation; not in the public header. On T41 it is the call behind System_MemPoolRequest.
Returns: 0 on success; -1 for an invalid id, a size of 0, a pool that is already in use, or when the memory is not available.
T21, T23, T30 and T31: ids 0 to 15, one DMA block of the given size, flags is ignored. T40 and T41: ids 0 to 31, size up to 2 GiB, the block is carved out of rmem as one contiguous range and registered under the id with its name. On T40 the System call only records the request and does not use this function. Use System_MemPoolRequest.
SoC: only in the vendor API of T23, T31
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | ✅ | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T23: apitest agg62 2026-10-31 .30
- T31: apitest agg61 2026-10-31 .26
int IMP_MemPool_GetById(int pool_id, void *info_out); /* T21, T23, T30, T31 */
void *IMP_MemPool_GetById(int pool_id); /* T40, T41 */Looks up a pool by id. Not in the public header.
Returns: T21, T23, T30, T31: 0 and an IMPDMABufferInfo (name, physical and virtual address, size, id) in info_out; -1 for an invalid id, a NULL pointer or a pool that is not in use. T40, T41: a non-NULL pointer when the pool exists, otherwise NULL.
On T41 FrameSource checks it before it binds a channel to the pool.
SoC: only in the vendor API of T23, T31
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | ✅ | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T23: apitest agg62 2026-10-31 .30
- T31: apitest agg61 2026-10-31 .26
int IMP_MemPool_Release(int pool_id);Low-level release of a pool. Use System_MemPoolFree where it exists.
Returns: 0 on success and for a pool that is not in use; -1 for an invalid id.
Only on T21, T23, T30 and T31 (ids 0 to 15); it frees the DMA block. T40 and T41 release pools through System_MemPoolFree.
SoC: T21, T23, T30 and T31.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | · | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T31: apitest agg61 2026-10-31 .26
uintptr_t IMP_PoolAlloc(int pool_id, void *info, intptr_t size, char *tag);Allocates a buffer from the pool pool_id made with MemPool_InitPool. Not in the public header.
Returns: Descriptor form (info is a descriptor): 0 and the virtual and physical address in the descriptor, -1 on failure. Size form: the virtual address, 0 on failure.
Free with PoolFree. The T31 vendor library takes the same four arguments and allocates through the pool object.
SoC: only in the vendor API of T23, T31
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | ✅ | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T23: apitest agg62 2026-10-31 .30
- T31: apitest agg61 2026-10-31 .26
void IMP_PoolFree(void *ptr);Gives a pool buffer back. Not in the public header.
Returns: 0 on success, non-zero on error (mostly -1; T40 and T41 return their own error codes for some failures).
No return value.
SoC: only in the vendor API of T23, T31
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | ✅ | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T23: apitest agg62 2026-10-31 .30
- T31: apitest agg61 2026-10-31 .26
int IMP_PoolAlloc_Dump(void);Prints the pools and the buffers allocated from them to the log. Debug aid.
Returns: 0.0
SoC: only in the vendor API of T23, T31
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | ✅ | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T23: apitest agg62 2026-10-31 .30
- T31: apitest agg61 2026-10-31 .26
int IMP_PoolAlloc_Set_Attr(uint32_t alignment, uint32_t cache_policy);Stores the alignment and cache policy for later pool allocations.
Returns: 0.0
SoC: only in the vendor API of T23, T31
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | ✅ | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T23: apitest agg62 2026-10-31 .30
- T31: apitest agg61 2026-10-31 .26
Difference from the vendor library: The T31 vendor function is an empty stub that returns 0 and stores nothing. OpenIMP stores the values and PoolAlloc_Get_Attr reads them back; the allocator itself does not use them (extra, documented).
int IMP_PoolAlloc_Get_Attr(uint32_t *alignment, uint32_t *cache_policy);Reads back the values stored by PoolAlloc_Set_Attr (default 4096 and 0).
Returns: 0.0
SoC: only in the vendor API of T23, T31
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | ✅ | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T23: apitest agg62 2026-10-31 .30
- T31: apitest agg61 2026-10-31 .26
Difference from the vendor library: The T31 vendor function returns 0 and leaves both outputs untouched. OpenIMP fills them (extra, documented).
int IMP_PoolFlushCache(void *ptr, uint32_t size);Writes the data cache of a pool buffer back to memory.
Returns: 0; -1 for a NULL pointer or a size of 0.
SoC: only in the vendor API of T23, T31
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | ✅ | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T23: apitest agg62 2026-10-31 .30
- T31: apitest agg61 2026-10-31 .26
Difference from the vendor library: The T31 vendor function is an empty stub that returns 0. OpenIMP flushes the range through the memory driver (extra, documented).
void *IMP_PoolPhys_to_Virt(uint32_t phys);Converts a physical address inside a pool buffer to the virtual address.
Returns: The address, NULL when it is not inside a live buffer.
SoC: only in the vendor API of T23, T31
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | ✅ | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T23: apitest agg62 2026-10-31 .30
- T31: apitest agg61 2026-10-31 .26
Difference from the vendor library: The T31 vendor function always returns 0. OpenIMP converts (extra, documented).
uint32_t IMP_PoolVirt_to_Phys(void *virt);Converts a virtual address inside a pool buffer to the physical address.
Returns: The address, 0 when it is not inside a live buffer.
SoC: only in the vendor API of T23, T31
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | ✅ | · | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T23: apitest agg62 2026-10-31 .30
- T31: apitest agg61 2026-10-31 .26
Difference from the vendor library: The T31 vendor function always returns 0. OpenIMP converts (extra, documented).
void *IMP_Phys_to_Virt(uint32_t phys);Converts a physical address inside an IMP_Alloc buffer to the virtual address (offsets inside the buffer work). Not in the public header.
Returns: The address, NULL when it is not inside a live buffer.
SoC: not in the vendor API of T40, T41
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T10: apitest t10wrench 2026-11-03 .31
- T20: apitest t20pass 2026-11-04 .107
- T21: apitest t21pass 2026-11-04 .24
- T23: apitest agg62 2026-10-31 .30
- T30: apitest T30X b67 2026-11-12 (vendor 1.0.5 and OpenIMP on the same camera, one power cycle, 0 oopses): alloc case: phys+100 maps to the expected virtual address on vendor and open
- T31: apitest agg61 2026-10-31 .26
Difference from the vendor library: T10, T20, T21: the vendor function forwards to the allocator, which converts any address inside its whole memory region linearly (0 outside it). OpenIMP converts inside live buffers only, so an address in the region but outside every buffer gives NULL (stricter).
uint32_t IMP_Virt_to_Phys(void *virt);Converts a virtual address inside an IMP_Alloc buffer to the physical address. Not in the public header.
Returns: The address, 0 when it is not inside a live buffer.
SoC: not in the vendor API of T40, T41
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T10: apitest t10wrench 2026-11-03 .31
- T20: apitest t20pass 2026-11-04 .107
- T21: apitest t21pass 2026-11-04 .24
- T23: apitest agg62 2026-10-31 .30
- T30: apitest T30X b67 2026-11-12 (vendor 1.0.5 and OpenIMP on the same camera, one power cycle, 0 oopses): alloc case: virt+100 maps back to descriptor phys+100 on vendor and open
- T31: apitest agg61 2026-10-31 .26
Difference from the vendor library: T10, T20, T21: the vendor function converts any address inside the allocator region (0 outside); OpenIMP only inside live buffers.
uintptr_t IMP_Sp_Alloc(int size, char *tag); /* T23 form, used on the other SoCs too */
int IMP_Sp_Alloc(IMPDMABufferInfo *info, uintptr_t phys); /* T10, T20, T21 */T23: allocates a physically contiguous buffer, like IMP_Alloc, and returns its virtual address. T10, T20 and T21: takes one more reference on a live buffer and fills the descriptor. Not in the public header.
Returns: T23: the virtual address, 0 on failure. T10, T20, T21: 0 and the descriptor filled; -1 for a NULL descriptor, a physical address of 0 or an address that is not a live buffer.
T10, T20, T21: phys is the physical address of a buffer from IMP_Alloc (exact start address, as the vendor allocator compares it). Every Sp_Alloc needs one more IMP_Free; the buffer is released by the last one. Disassembly of the T20/T10 3.12.0 and T21 1.0.33 libraries: the vendor function is the reference-count increment of the allocator (counter at +16 of the allocation record, initial value 1, free decrements and releases at 0).
SoC: not in the vendor API of T40, T41
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅+ | ✅ | · | · |
Tested (tool, date, camera, what was checked):
- T10: apitest t10wrench 2026-11-03 .31
- T20: apitest t20pass 2026-11-04 .107
- T21: apitest t21pass 2026-11-04 .24
- T23: apitest agg62 2026-10-31 .30
- T30: apitest T30X b67 2026-11-12: open PASS (4096 bytes written and freed); vendor 1.0.5 FAILs (IMP_Sp_Alloc -1, Alloc_Set_Attr -99): vendor crash/fail, no vendor line ...
- T31: apitest agg61 2026-10-31 .26
Difference from the vendor library: The T31 vendor function also takes a descriptor plus a second non-NULL argument and returns 0 or -1, so it very likely is the same reference call; the OpenIMP call form on T31 still follows the T23 library and is not tested on a T31 camera. Better than the vendor on T30.
int IMP_Log_Get_Option(void);
void IMP_Log_Set_Option(int option);Reads or sets which fields (file, line, function, ...) the library prints in front of a log message.
Parameters
-
option: Bit mask of theIMP_LOG_OP_*flags;IMP_LOG_OP_NONEgives the terse vendor line.
Returns: Get returns the mask, Set has no return value.
OpenIMP prints warnings and errors to stderr always, informational and debug messages only when OPENIMP_DEBUG_TRACE is set. Syslog output is enabled with OPENIMP_LOG_SYSLOG=1. The default mask is the full one.
SoC: not in the vendor API of T30
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | · | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: apitest agg47-cases 2026-10-26 .31
- T20: apitest agg47-cases 2026-10-26 .107
- T21: apitest agg47-cases 2026-10-26 .24
- T23: apitest agg47-cases 2026-10-26 .30
- T31: apitest next+ko/libimp swap 2026-10-27 .26
- T40: apitest t40pass 2026-11-04 .184
- T41: apitest t41pass 2026-11-04 .34
Difference from the vendor library: The vendor library starts with no fields and prints everything it has; OpenIMP is quiet by default so a camera is not flooded.
A JPEG decoder on channel 0, implemented by OpenIMP on T10, T20, T21, T23 and T30 (the T40 and T41 vendor libraries do not export it; OpenIMP answers -1/ENOTSUP there). The vendor libimp of T10, T20 and T21 decodes in the VPU (ijpegd); OpenIMP builds of these SoCs use the software decoder only. It decodes JPEG pictures into NV12, NV21 or YUV420P. The vendor library decodes on the VPU; OpenIMP decodes in its own software decoder (baseline JPEG, gray or YCbCr, 4:4:4, 4:2:2, 4:4:0 or 4:2:0, restart intervals; progressive and arithmetic-coded files are refused) in the calling thread. A build with the OEM helper process (OPENIMP_T23_HELIX_BRIDGE=1) uses the helper's VPU decoder instead and falls back to the software decoder when the helper cannot be started. The picture is always 4:2:0 with the width and height of the file (no alignment padding), the colour planes taken straight from the file's Cb and Cr.
/* IMP_Decoder_CreateChn: exported by libimp; declared in the vendor SDK header, not in the OpenIMP include directory */Creates the decoder channel 0 (software decoder, or the OEM helper when built in).
Parameters
-
chn: Channel, only 0. -
attr: Decoder attribute; the type must be JPEG, the maximum size non-zero, the pixel format NV12, NV21 or YUV420P.
Returns: -1 for another channel, another payload type or pixel format, a zero size, an existing channel or when the worker cannot be started.
SoC: Implemented by OpenIMP on T10, T20, T21, T23 and T30 (software decoder; on T30 since 2026-11-12, the vendor 1.0.5 library exports it too). T40 and T41 export it as -1 with errno ENOTSUP (no vendor decoder there).
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅+ | · | · | · |
Tested (tool, date, camera, what was checked):
- T10: apitest dec-t1x 2026-11-01 .31 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL in the case run; QEMU T10 first: 8/8, frames flow) ...
- T20: apitest dec-t1x 2026-11-01 .107 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T20 first: 8/8, frames flow) ...
- T21: apitest dec-t1x 2026-11-01 .24 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T21 first: 8/8, frames flow) ...
- T23: apitest t23pass 2026-11-04 .30 (case decoder_t23, 8/8 PASS; software decode, small known test picture, no host-decode comparison on this SoC yet)
- T30: apitest decoder_t2x 2026-11-12 QEMU T30 (OpenIMP claude/t30-gaps, open tx-isp-t30 of claude/t30-283b, sc4236 model): 8/8 PASS: a 32x24 baseline JPEG decoded into NV12 on channel 0 (Y 208/70, Cb/Cr 33/143 as the known picture), H.264 ...
- T40: vendor libimp T40 1.3.1 (uclibc and glibc libimp.so, libimp.a): no IMP_Decoder_* symbol exported (nm -D, strings); the names only exist in the SDK header imp_decoder.h, so the stock library answers a program with an unresolved symbol (checked 2026-10-31, #66)
- T41: apitest agg5y 2026-11-01 .34 (open-tx-isp next a84b6be2 + OpenIMP next 246bf938, full-order cases run): 8/8 PASS, every call -1/ENOTSUP, arguments untouched, frames still flow
Difference from the vendor library: Better than the vendor on T30.
/* IMP_Decoder_DestroyChn: exported by libimp; declared in the vendor SDK header, not in the OpenIMP include directory */Destroys the channel (stops the helper when one is used) and releases all frames.
Returns: -1 for another channel or a channel that does not exist.
SoC: Implemented by OpenIMP on T10, T20, T21, T23 and T30 (software decoder; on T30 since 2026-11-12, the vendor 1.0.5 library exports it too). T40 and T41 export it as -1 with errno ENOTSUP (no vendor decoder there).
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅+ | · | · | · |
Tested (tool, date, camera, what was checked):
- T10: apitest dec-t1x 2026-11-01 .31 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL in the case run; QEMU T10 first: 8/8, frames flow) ...
- T20: apitest dec-t1x 2026-11-01 .107 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T20 first: 8/8, frames flow) ...
- T21: apitest dec-t1x 2026-11-01 .24 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T21 first: 8/8, frames flow) ...
- T23: apitest t23pass 2026-11-04 .30 (case decoder_t23, 8/8 PASS; software decode, small known test picture, no host-decode comparison on this SoC yet)
- T30: apitest decoder_t2x 2026-11-12 QEMU T30 (OpenIMP claude/t30-gaps, open tx-isp-t30 of claude/t30-283b, sc4236 model): 8/8 PASS: a 32x24 baseline JPEG decoded into NV12 on channel 0 (Y 208/70, Cb/Cr 33/143 as the known picture), H.264 ...
- T40: vendor libimp T40 1.3.1 (uclibc and glibc libimp.so, libimp.a): no IMP_Decoder_* symbol exported (nm -D, strings); the names only exist in the SDK header imp_decoder.h, so the stock library answers a program with an unresolved symbol (checked 2026-10-31, #66)
- T41: apitest agg5y 2026-11-01 .34 (open-tx-isp next a84b6be2 + OpenIMP next 246bf938, full-order cases run): 8/8 PASS, every call -1/ENOTSUP, arguments untouched, frames still flow
Difference from the vendor library: Better than the vendor on T30.
/* IMP_Decoder_StartRecvPic: exported by libimp; declared in the vendor SDK header, not in the OpenIMP include directory */Marks the channel as receiving.
Returns: -1 for another channel or a channel that does not exist.
SoC: Implemented by OpenIMP on T10, T20, T21, T23 and T30 (software decoder; on T30 since 2026-11-12, the vendor 1.0.5 library exports it too). T40 and T41 export it as -1 with errno ENOTSUP (no vendor decoder there).
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅+ | · | · | · |
Tested (tool, date, camera, what was checked):
- T10: apitest dec-t1x 2026-11-01 .31 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL in the case run; QEMU T10 first: 8/8, frames flow) ...
- T20: apitest dec-t1x 2026-11-01 .107 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T20 first: 8/8, frames flow) ...
- T21: apitest dec-t1x 2026-11-01 .24 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T21 first: 8/8, frames flow) ...
- T23: apitest t23pass 2026-11-04 .30 (case decoder_t23, 8/8 PASS; software decode, small known test picture, no host-decode comparison on this SoC yet)
- T30: apitest decoder_t2x 2026-11-12 QEMU T30 (OpenIMP claude/t30-gaps, open tx-isp-t30 of claude/t30-283b, sc4236 model): 8/8 PASS: a 32x24 baseline JPEG decoded into NV12 on channel 0 (Y 208/70, Cb/Cr 33/143 as the known picture), H.264 ...
- T40: vendor libimp T40 1.3.1 (uclibc and glibc libimp.so, libimp.a): no IMP_Decoder_* symbol exported (nm -D, strings); the names only exist in the SDK header imp_decoder.h, so the stock library answers a program with an unresolved symbol (checked 2026-10-31, #66)
- T41: apitest agg5y 2026-11-01 .34 (open-tx-isp next a84b6be2 + OpenIMP next 246bf938, full-order cases run): 8/8 PASS, every call -1/ENOTSUP, arguments untouched, frames still flow
Difference from the vendor library: Better than the vendor on T30.
/* IMP_Decoder_StopRecvPic: exported by libimp; declared in the vendor SDK header, not in the OpenIMP include directory */Marks the channel as not receiving and wakes any waiting caller.
Returns: -1 for another channel or a channel that does not exist.
SoC: Implemented by OpenIMP on T10, T20, T21, T23 and T30 (software decoder; on T30 since 2026-11-12, the vendor 1.0.5 library exports it too). T40 and T41 export it as -1 with errno ENOTSUP (no vendor decoder there).
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅+ | · | · | · |
Tested (tool, date, camera, what was checked):
- T10: apitest dec-t1x 2026-11-01 .31 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL in the case run; QEMU T10 first: 8/8, frames flow) ...
- T20: apitest dec-t1x 2026-11-01 .107 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T20 first: 8/8, frames flow) ...
- T21: apitest dec-t1x 2026-11-01 .24 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T21 first: 8/8, frames flow) ...
- T23: apitest t23pass 2026-11-04 .30 (case decoder_t23, 8/8 PASS; software decode, small known test picture, no host-decode comparison on this SoC yet)
- T30: apitest decoder_t2x 2026-11-12 QEMU T30 (OpenIMP claude/t30-gaps, open tx-isp-t30 of claude/t30-283b, sc4236 model): 8/8 PASS: a 32x24 baseline JPEG decoded into NV12 on channel 0 (Y 208/70, Cb/Cr 33/143 as the known picture), H.264 ...
- T40: vendor libimp T40 1.3.1 (uclibc and glibc libimp.so, libimp.a): no IMP_Decoder_* symbol exported (nm -D, strings); the names only exist in the SDK header imp_decoder.h, so the stock library answers a program with an unresolved symbol (checked 2026-10-31, #66)
- T41: apitest agg5y 2026-11-01 .34 (open-tx-isp next a84b6be2 + OpenIMP next 246bf938, full-order cases run): 8/8 PASS, every call -1/ENOTSUP, arguments untouched, frames still flow
Difference from the vendor library: Better than the vendor on T30.
/* IMP_Decoder_SendStreamTimeout: exported by libimp; declared in the vendor SDK header, not in the OpenIMP include directory */Decodes one JPEG picture. It waits up to the timeout for a free picture buffer, decodes and queues the picture.
Parameters
-
stream: JPEG data (p_payload,i_payload) and a timestamp that is passed on to the picture. -
timeout_ms: Maximum wait in milliseconds for a buffer and for the decode.
Returns: -1 for another channel, an empty or NULL stream, no free buffer within the timeout, a file that is not a baseline JPEG, damaged or truncated, or a picture larger than the channel's maximum size.
Number of pictures that can be held at once is the nrKeepStream of the channel attribute (at least 1). Give every picture back with ReleaseFrame.
SoC: Implemented by OpenIMP on T10, T20, T21, T23 and T30 (software decoder; on T30 since 2026-11-12, the vendor 1.0.5 library exports it too). T40 and T41 export it as -1 with errno ENOTSUP (no vendor decoder there).
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅+ | · | · | · |
Tested (tool, date, camera, what was checked):
- T10: apitest dec-t1x 2026-11-01 .31 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL in the case run; QEMU T10 first: 8/8, frames flow) ...
- T20: apitest dec-t1x 2026-11-01 .107 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T20 first: 8/8, frames flow) ...
- T21: apitest dec-t1x 2026-11-01 .24 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T21 first: 8/8, frames flow) ...
- T23: apitest t23pass 2026-11-04 .30 (case decoder_t23, 8/8 PASS; software decode, small known test picture, no host-decode comparison on this SoC yet)
- T30: apitest decoder_t2x 2026-11-12 QEMU T30 (OpenIMP claude/t30-gaps, open tx-isp-t30 of claude/t30-283b, sc4236 model): 8/8 PASS: a 32x24 baseline JPEG decoded into NV12 on channel 0 (Y 208/70, Cb/Cr 33/143 as the known picture), H.264 ...
- T40: vendor libimp T40 1.3.1 (uclibc and glibc libimp.so, libimp.a): no IMP_Decoder_* symbol exported (nm -D, strings); the names only exist in the SDK header imp_decoder.h, so the stock library answers a program with an unresolved symbol (checked 2026-10-31, #66)
- T41: apitest agg5y 2026-11-01 .34 (open-tx-isp next a84b6be2 + OpenIMP next 246bf938, full-order cases run): 8/8 PASS, every call -1/ENOTSUP, arguments untouched, frames still flow
Difference from the vendor library: Better than the vendor on T30.
/* IMP_Decoder_PollingFrame: exported by libimp; declared in the vendor SDK header, not in the OpenIMP include directory */Waits until a decoded picture is ready.
Returns: 0 when ready; -1 on timeout, for another channel or when the channel does not exist.
SoC: Implemented by OpenIMP on T10, T20, T21, T23 and T30 (software decoder; on T30 since 2026-11-12, the vendor 1.0.5 library exports it too). T40 and T41 export it as -1 with errno ENOTSUP (no vendor decoder there).
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅+ | · | · | · |
Tested (tool, date, camera, what was checked):
- T10: apitest dec-t1x 2026-11-01 .31 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL in the case run; QEMU T10 first: 8/8, frames flow) ...
- T20: apitest dec-t1x 2026-11-01 .107 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T20 first: 8/8, frames flow) ...
- T21: apitest dec-t1x 2026-11-01 .24 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T21 first: 8/8, frames flow) ...
- T23: apitest t23pass 2026-11-04 .30 (case decoder_t23, 8/8 PASS; software decode, small known test picture, no host-decode comparison on this SoC yet)
- T30: apitest decoder_t2x 2026-11-12 QEMU T30 (OpenIMP claude/t30-gaps, open tx-isp-t30 of claude/t30-283b, sc4236 model): 8/8 PASS: a 32x24 baseline JPEG decoded into NV12 on channel 0 (Y 208/70, Cb/Cr 33/143 as the known picture), H.264 ...
- T40: vendor libimp T40 1.3.1 (uclibc and glibc libimp.so, libimp.a): no IMP_Decoder_* symbol exported (nm -D, strings); the names only exist in the SDK header imp_decoder.h, so the stock library answers a program with an unresolved symbol (checked 2026-10-31, #66)
- T41: apitest agg5y 2026-11-01 .34 (open-tx-isp next a84b6be2 + OpenIMP next 246bf938, full-order cases run): 8/8 PASS, every call -1/ENOTSUP, arguments untouched, frames still flow
Difference from the vendor library: Better than the vendor on T30.
/* IMP_Decoder_GetFrame: exported by libimp; declared in the vendor SDK header, not in the OpenIMP include directory */Hands out the oldest decoded picture (a pointer to its frame info with address and size).
Returns: -1 when none is ready, for another channel, NULL or when the channel does not exist.
SoC: Implemented by OpenIMP on T10, T20, T21, T23 and T30 (software decoder; on T30 since 2026-11-12, the vendor 1.0.5 library exports it too). T40 and T41 export it as -1 with errno ENOTSUP (no vendor decoder there).
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅+ | · | · | · |
Tested (tool, date, camera, what was checked):
- T10: apitest dec-t1x 2026-11-01 .31 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL in the case run; QEMU T10 first: 8/8, frames flow) ...
- T20: apitest dec-t1x 2026-11-01 .107 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T20 first: 8/8, frames flow) ...
- T21: apitest dec-t1x 2026-11-01 .24 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T21 first: 8/8, frames flow) ...
- T23: apitest t23pass 2026-11-04 .30 (case decoder_t23, 8/8 PASS; software decode, small known test picture, no host-decode comparison on this SoC yet)
- T30: apitest decoder_t2x 2026-11-12 QEMU T30 (OpenIMP claude/t30-gaps, open tx-isp-t30 of claude/t30-283b, sc4236 model): 8/8 PASS: a 32x24 baseline JPEG decoded into NV12 on channel 0 (Y 208/70, Cb/Cr 33/143 as the known picture), H.264 ...
- T40: vendor libimp T40 1.3.1 (uclibc and glibc libimp.so, libimp.a): no IMP_Decoder_* symbol exported (nm -D, strings); the names only exist in the SDK header imp_decoder.h, so the stock library answers a program with an unresolved symbol (checked 2026-10-31, #66)
- T41: apitest agg5y 2026-11-01 .34 (open-tx-isp next a84b6be2 + OpenIMP next 246bf938, full-order cases run): 8/8 PASS, every call -1/ENOTSUP, arguments untouched, frames still flow
Difference from the vendor library: Better than the vendor on T30.
/* IMP_Decoder_ReleaseFrame: exported by libimp; declared in the vendor SDK header, not in the OpenIMP include directory */Gives a picture back so the buffer can be reused.
Returns: -1 for another channel, NULL, or a frame that was not handed out.
SoC: Implemented by OpenIMP on T10, T20, T21, T23 and T30 (software decoder; on T30 since 2026-11-12, the vendor 1.0.5 library exports it too). T40 and T41 export it as -1 with errno ENOTSUP (no vendor decoder there).
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅+ | · | · | · |
Tested (tool, date, camera, what was checked):
- T10: apitest dec-t1x 2026-11-01 .31 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL in the case run; QEMU T10 first: 8/8, frames flow) ...
- T20: apitest dec-t1x 2026-11-01 .107 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T20 first: 8/8, frames flow) ...
- T21: apitest dec-t1x 2026-11-01 .24 (library from /tmp, streamer restarted, stock module md5 unchanged): 8/8 PASS (32 PASS / 0 FAIL; QEMU T21 first: 8/8, frames flow) ...
- T23: apitest t23pass 2026-11-04 .30 (case decoder_t23, 8/8 PASS; software decode, small known test picture, no host-decode comparison on this SoC yet)
- T30: apitest decoder_t2x 2026-11-12 QEMU T30 (OpenIMP claude/t30-gaps, open tx-isp-t30 of claude/t30-283b, sc4236 model): 8/8 PASS: a 32x24 baseline JPEG decoded into NV12 on channel 0 (Y 208/70, Cb/Cr 33/143 as the known picture), H.264 ...
- T40: vendor libimp T40 1.3.1 (uclibc and glibc libimp.so, libimp.a): no IMP_Decoder_* symbol exported (nm -D, strings); the names only exist in the SDK header imp_decoder.h, so the stock library answers a program with an unresolved symbol (checked 2026-10-31, #66)
- T41: apitest agg5y 2026-11-01 .34 (open-tx-isp next a84b6be2 + OpenIMP next 246bf938, full-order cases run): 8/8 PASS, every call -1/ENOTSUP, arguments untouched, frames still flow
Difference from the vendor library: Better than the vendor on T30.
int SU_Base_SetWkupMode(int mode);Selects how the RTC wake-up behaves (mode 1 or 2). T41 only.
Returns: 0 when the RTC accepted the request, -1 (with errno) for a mode other than 1 or 2 (EINVAL, nothing is touched), a missing /dev/rtc0, or an RTC driver that does not know the request.
The request is the one of the vendor libsysutils: a three-word block {0, 0x2c, mode code} sent to /dev/rtc0. The call changes the wake-up configuration, so test it only on purpose.
SoC: only in the vendor API of T41
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | · | · | · | · | ✅? |
Tested (tool, date, camera, what was checked):
- T41: mx1112-r2 2026-11-12 T41 test camera (Vanhua, agg68a, no /dev/rtc0): OpenIMP libimp.so (0ce3e85a) and the vendor 1.2.6 libsysutils answer -1 for modes 0/1/2 (vendor errno ENOENT from the /dev/rtc0 open; OpenIMP refuses mode 0 before the open, EINVAL ...
Difference from the vendor library: The board's libsysutils writes the mode number into /sys/power/state, which the kernel rejects; OpenIMP sends the vendor request instead. The symbol is in libimp.so, so it wins only where libimp is resolved first.
int SU_Base_GetTime(SUTime *time);
int SU_Base_SetTime(SUTime *time);Reads or sets the battery-backed RTC (/dev/rtc0). Month is 1 to 12 and the year is the full year.
Returns: 0, -1 when the RTC node cannot be opened or the ioctl fails (a board without an RTC driver).
SoC: T40 (ported from the vendor libsysutils into libimp; other SoCs use the vendor libsysutils.so).
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? |
Tested (tool, date, camera, what was checked):
- T10: QEMU 0.7.14 ingenic-t10 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s, past alarm refused (vendor arms it; vendor Raw2SUTime year and GetAlarm date differ, open lib correct) ...
- T20: Wyze Cam v2 (T20X) 2026-11-11: no RTC/ADC/AES node on the board; board and vendor 3.12.0 libsysutils return the same error (#306) | QEMU 0.7.14 ingenic-t20 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s ...
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: no /dev/rtc0 (kernel without RTC_CLASS): GetTime/SetTime/GetAlarm/DisableAlarm/PollingAlarm -1 errno 2 with all three, same result ...
- T23: apitest su 2026-11-11 T23 Tapo C500 (no /dev/rtc0): board and vendor T23 1.3.0 libsysutils both answer -1, same result; nohw | QEMU 0.7.14 ingenic-t23 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s ...
- T30: T30X b61 2026-11-12 (T30X lab camera, one power cycle, vendor stack then open stack by runtime swap, open-tx-isp 0fdfacb4 + OpenIMP 4aa7e40a, apitest su): no RTC/ADC node on the board ...
- T31: QEMU 0.7.14 ingenic-t31 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s, past alarm refused (vendor arms it; vendor Raw2SUTime year and GetAlarm date differ, open lib correct) ...
- T40: apitest su 2026-11-11 (Eufy T40XP, t40-mx, #310): OpenIMP libimp vs vendor 1.3.1 libsysutils
- T41: SU test 2026-11-11 on the Vanhua T41 test camera (gc5603, open-stack image) (#318): image libsysutils (neo-sysutils) and vendor T41 1.2.6 libsysutils (from /tmp, vendor libalog preloaded) give the same answer: -1 with errno ENOENT for GetTime, SetTime ...
int SU_Base_GetAlarm(SUTime *time);
int SU_Base_SetAlarm(SUTime *time);Reads or sets the RTC alarm time.
Returns: 0, -1 on a missing RTC or a failing ioctl.
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? |
Tested (tool, date, camera, what was checked):
- T10: QEMU 0.7.14 ingenic-t10 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s, past alarm refused (vendor arms it; vendor Raw2SUTime year and GetAlarm date differ, open lib correct) ...
- T20: Wyze Cam v2 (T20X) 2026-11-11: no RTC/ADC/AES node on the board; board and vendor 3.12.0 libsysutils return the same error (#306) | QEMU 0.7.14 ingenic-t20 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s ...
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: no /dev/rtc0 (kernel without RTC_CLASS): GetTime/SetTime/GetAlarm/DisableAlarm/PollingAlarm -1 errno 2 with all three, same result ...
- T23: apitest su 2026-11-11 T23 Tapo C500 (no /dev/rtc0): board and vendor T23 1.3.0 libsysutils both answer -1, same result; nohw | QEMU 0.7.14 ingenic-t23 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s ...
- T30: T30X b61 2026-11-12 (T30X lab camera, one power cycle, vendor stack then open stack by runtime swap, open-tx-isp 0fdfacb4 + OpenIMP 4aa7e40a, apitest su): no RTC/ADC node on the board ...
- T31: QEMU 0.7.14 ingenic-t31 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s, past alarm refused (vendor arms it; vendor Raw2SUTime year and GetAlarm date differ, open lib correct) ...
- T40: apitest su 2026-11-11 (Eufy T40XP, t40-mx, #310): OpenIMP libimp vs vendor 1.3.1 libsysutils
- T41: SU test 2026-11-11 on the Vanhua T41 test camera (gc5603, open-stack image) (#318): image libsysutils (neo-sysutils) and vendor T41 1.2.6 libsysutils (from /tmp, vendor libalog preloaded) give the same answer: -1 with errno ENOENT for GetTime, SetTime ...
int SU_Base_EnableAlarm(void);Switches the RTC alarm interrupt on. Refused with -1 unless the alarm time lies after the current time.
Returns: 0 on success, non-zero on error (mostly -1; T40 and T41 return their own error codes for some failures).
An armed alarm can wake or power the camera on boards that wire it up.
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? |
Tested (tool, date, camera, what was checked):
- T10: QEMU 0.7.14 ingenic-t10 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s, past alarm refused (vendor arms it; vendor Raw2SUTime year and GetAlarm date differ, open lib correct) ...
- T20: Wyze Cam v2 (T20X) 2026-11-11: no RTC/ADC/AES node on the board; board and vendor 3.12.0 libsysutils return the same error (#306) | QEMU 0.7.14 ingenic-t20 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s ...
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: no /dev/rtc0 (kernel without RTC_CLASS): GetTime/SetTime/GetAlarm/DisableAlarm/PollingAlarm -1 errno 2 with all three, same result ...
- T23: apitest su 2026-11-11 T23 Tapo C500 (no /dev/rtc0): board and vendor T23 1.3.0 libsysutils both answer -1, same result; nohw | QEMU 0.7.14 ingenic-t23 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s ...
- T30: T30X b61 2026-11-12 (T30X lab camera, one power cycle, vendor stack then open stack by runtime swap, open-tx-isp 0fdfacb4 + OpenIMP 4aa7e40a, apitest su): no RTC/ADC node on the board ...
- T31: QEMU 0.7.14 ingenic-t31 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s, past alarm refused (vendor arms it; vendor Raw2SUTime year and GetAlarm date differ, open lib correct) ...
- T40: apitest su 2026-11-11 (Eufy T40XP, t40-mx, #310): OpenIMP libimp vs vendor 1.3.1 libsysutils
- T41: SU test 2026-11-11 on the Vanhua T41 test camera (gc5603, open-stack image) (#318): image libsysutils (neo-sysutils) and vendor T41 1.2.6 libsysutils (from /tmp, vendor libalog preloaded) give the same answer: -1 with errno ENOENT for GetTime, SetTime ...
int SU_Base_DisableAlarm(void);Switches the RTC alarm interrupt off.
Returns: 0 on success, non-zero on error (mostly -1; T40 and T41 return their own error codes for some failures).
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? |
Tested (tool, date, camera, what was checked):
- T10: QEMU 0.7.14 ingenic-t10 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s, past alarm refused (vendor arms it; vendor Raw2SUTime year and GetAlarm date differ, open lib correct) ...
- T20: Wyze Cam v2 (T20X) 2026-11-11: no RTC/ADC/AES node on the board; board and vendor 3.12.0 libsysutils return the same error (#306) | QEMU 0.7.14 ingenic-t20 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s ...
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: no /dev/rtc0 (kernel without RTC_CLASS): GetTime/SetTime/GetAlarm/DisableAlarm/PollingAlarm -1 errno 2 with all three, same result ...
- T23: apitest su 2026-11-11 T23 Tapo C500 (no /dev/rtc0): board and vendor T23 1.3.0 libsysutils both answer -1, same result; nohw | QEMU 0.7.14 ingenic-t23 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s ...
- T30: T30X b61 2026-11-12 (T30X lab camera, one power cycle, vendor stack then open stack by runtime swap, open-tx-isp 0fdfacb4 + OpenIMP 4aa7e40a, apitest su): no RTC/ADC node on the board ...
- T31: QEMU 0.7.14 ingenic-t31 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s, past alarm refused (vendor arms it; vendor Raw2SUTime year and GetAlarm date differ, open lib correct) ...
- T40: apitest su 2026-11-11 (Eufy T40XP, t40-mx, #310): OpenIMP libimp vs vendor 1.3.1 libsysutils
- T41: SU test 2026-11-11 on the Vanhua T41 test camera (gc5603, open-stack image) (#318): image libsysutils (neo-sysutils) and vendor T41 1.2.6 libsysutils (from /tmp, vendor libalog preloaded) give the same answer: -1 with errno ENOENT for GetTime, SetTime ...
int SU_Base_PollingAlarm(uint32_t timeoutMsec);Waits up to timeoutMsec for the RTC alarm.
Returns: 0, also after the timeout (the vendor only logs it); -1 when the RTC node cannot be opened.
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? |
Tested (tool, date, camera, what was checked):
- T10: QEMU 0.7.14 ingenic-t10 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s, past alarm refused (vendor arms it; vendor Raw2SUTime year and GetAlarm date differ, open lib correct) ...
- T20: Wyze Cam v2 (T20X) 2026-11-11: no RTC/ADC/AES node on the board; board and vendor 3.12.0 libsysutils return the same error (#306) | QEMU 0.7.14 ingenic-t20 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s ...
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: no /dev/rtc0 (kernel without RTC_CLASS): GetTime/SetTime/GetAlarm/DisableAlarm/PollingAlarm -1 errno 2 with all three, same result ...
- T23: apitest su 2026-11-11 T23 Tapo C500 (no /dev/rtc0): board and vendor T23 1.3.0 libsysutils both answer -1, same result; nohw | QEMU 0.7.14 ingenic-t23 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s ...
- T30: T30X b61 2026-11-12 (T30X lab camera, one power cycle, vendor stack then open stack by runtime swap, open-tx-isp 0fdfacb4 + OpenIMP 4aa7e40a, apitest su): no RTC/ADC node on the board ...
- T31: QEMU 0.7.14 ingenic-t31 (#327, 2026-11-11): clock set/read back, alarm 3 s ahead fires after 3.0 s, past alarm refused (vendor arms it; vendor Raw2SUTime year and GetAlarm date differ, open lib correct) ...
- T40: apitest su 2026-11-11 (Eufy T40XP, t40-mx, #310): OpenIMP libimp vs vendor 1.3.1 libsysutils
- T41: SU test 2026-11-11 on the Vanhua T41 test camera (gc5603, open-stack image) (#318): image libsysutils (neo-sysutils) and vendor T41 1.2.6 libsysutils (from /tmp, vendor libalog preloaded) give the same answer: -1 with errno ENOENT for GetTime, SetTime ...
int SU_Base_Reboot(void);Syncs and sends SIGTERM to init (reboot on busybox init). Returns 0.
Returns: 0 on success, non-zero on error (mostly -1; T40 and T41 return their own error codes for some failures).
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
Tested (tool, date, camera, what was checked):
- T10: sutool + apitest su t10-mx 2026-11-11 .31 (SU_Base_Reboot with board and vendor libsysutils: ret 0, camera rebooted to the flashed stack)
- T20: sutool + apitest su t20-mx 2026-11-11 .107 (SU_Base_Reboot with board and vendor libsysutils: ret 0, camera rebooted to the flashed stack)
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: SU_Base_Reboot called for real with the vendor 5.4.0 library and with the image library: both return 0 and send kill(1 ...
- T23: 2026-11-11 T23 Jooan A6M (sc1a4t, open stack, kernel 3.10.14, no /sys/power, no RTC/ADC/AES nodes), image libsysutils (neo-sysutils-1.0) vs vendor T23 ISVP 1.3.0 libsysutils (with its libalog) loaded from /tmp: SU_Base_Reboot called for real with the image lib ...
- T30: apitest/extras T30X b68 2026-11-13 (audit/t30/b68, one power cycle, vendor 1.0.5 stack then OpenIMP next c3319ab1 libimp + b67 ko, no flash, 0 oopses in su/encx): SU_Base_Reboot via vendor libsysutils after the open stack: ret 0, no fallback ...
- T31: reboot-1112 2026-11-12 T31L camera (sc2336, open stack): SU_Base_Reboot with the image libsysutils and the vendor T31 libsysutils (with its libalog) loaded from /tmp: both return 0 ...
- T40: reboot-1112 2026-11-12 Eufy T40XP (sc830ai, open stack): SU_Base_Reboot with the image libsysutils and the vendor libsysutils from the OEM image loaded from /tmp: both return 0, the camera went down after 10 s and was back after 34-35 s with a new boot id ...
- T41: SU test 2026-11-11 on the Vanhua T41 test camera (gc5603, open-stack image) (#318): SU_Base_Reboot called for real with the image libsysutils (neo-sysutils) and with the vendor T41 1.2.6 libsysutils loaded from /tmp: both return 0 (sync(), kill(1 ...
int SU_Base_Suspend(void);Writes mem to /sys/power/state.
Returns: 0, -1 when the node cannot be opened or written.
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| 🔧a | 🔧a | 🔧a | 🔧a | 🔧a | 🔧a | 🔧h | 🔧a |
Tested (tool, date, camera, what was checked):
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: SU_Base_Suspend -1 errno 2 with all three (no /sys/power on this kernel); stays aud: a real suspend cannot be proven on this board
- T23: 2026-11-11 T23 Jooan A6M (sc1a4t, open stack, kernel 3.10.14, no /sys/power, no RTC/ADC/AES nodes), image libsysutils (neo-sysutils-1.0) vs vendor T23 ISVP 1.3.0 libsysutils (with its libalog) loaded from /tmp: stays aud: the kernel has no /sys/power (no suspe ...
- T30: static audit 2026-10-31 (OpenIMP claude/t30-audit) || T30X b88 2026-11-14 (audit/t30/b88, one power cycle, 3 min, no flash, 0 oopses): neo libsysutils (Lu-Fi/ingenic-system-libs-neo aperto 2d689bd, built for T30) from /tmp vs vendor 1.0.5 ...
- T40: host test tests/core/su_sysutils_test.c (fake ioctl and temp files, 2026-10-31); ported from the vendor libsysutils 1.3.1 disassembly; no camera run yet (the Eufy .184 may have no RTC / ADC / gpio_keys / LED, apitest reports N/A there)
int SU_ADC_Init(void);Opens the six ADC channel nodes /dev/ingenic_adc_aux_0 to _5.
Returns: 0, -1 when one node is missing (the board has no ADC driver).
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| · | · | · | · | · | · | ✅? | ✅ |
Tested (tool, date, camera, what was checked):
- T40: apitest su 2026-11-11 (Eufy T40XP, t40-mx, #310): OpenIMP libimp vs vendor 1.3.1 libsysutils
- T41: apitest agg5w 2026-11-01 .34 on a TEST image (kernel with CONFIG_MFD_INGENIC_SADC_V13/_AUX, not in any firmware PR): SU_ADC_Init/EnableChn/GetChnValue/DisableChn/Exit PASS, 6 reload cycles clean
int SU_ADC_Exit(void);Closes the channel nodes.
Returns: 0 on success, non-zero on error (mostly -1; T40 and T41 return their own error codes for some failures).
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? |
Tested (tool, date, camera, what was checked):
- T10: QEMU 0.7.14 ingenic-t10 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS, image libsysutils vs vendor T10 libsysutils in the same guest: vendor-equal or better; the test boards have no such node (board lacks the hardware, apitest N/A)
- T20: Wyze Cam v2 (T20X) 2026-11-11: no RTC/ADC/AES node on the board; board and vendor 3.12.0 libsysutils return the same error (#306) | QEMU 0.7.14 ingenic-t20 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS ...
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: no ADC node (kernel without SADC aux): SU_ADC_Init -1 errno 2 with all three, same result; nohw | QEMU 0.7.14 ingenic-t21 (#327 ...
- T23: apitest su 2026-11-11 T23 Tapo C500 (no ADC node): SU_ADC_Init -1 with board and vendor T23 1.3.0 libsysutils, same result; nohw | QEMU 0.7.14 ingenic-t23 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS ...
- T30: T30X b61 2026-11-12 (T30X lab camera, one power cycle, vendor stack then open stack by runtime swap, open-tx-isp 0fdfacb4 + OpenIMP 4aa7e40a, apitest su): no RTC/ADC node on the board ...
- T31: QEMU 0.7.14 ingenic-t31 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS, image libsysutils vs vendor T31 libsysutils in the same guest: vendor-equal or better; the test boards have no such node (board lacks the hardware, apitest N/A)
- T40: apitest su 2026-11-11 (Eufy T40XP, t40-mx, #310): OpenIMP libimp vs vendor 1.3.1 libsysutils
- T41: SU test 2026-11-11 on the Vanhua T41 test camera (gc5603, open-stack image) (#318): image libsysutils (neo-sysutils) and vendor T41 1.2.6 libsysutils (from /tmp, vendor libalog preloaded) give the same answer: Init -1 ENOENT ...
int SU_ADC_EnableChn(uint32_t chn_num);Switches an ADC channel on (0 to 5).
Returns: 0, -1 for a channel above 5, before SU_ADC_Init or on an ioctl error.
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? |
Tested (tool, date, camera, what was checked):
- T10: QEMU 0.7.14 ingenic-t10 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS, image libsysutils vs vendor T10 libsysutils in the same guest: vendor-equal or better; the test boards have no such node (board lacks the hardware, apitest N/A)
- T20: Wyze Cam v2 (T20X) 2026-11-11: no RTC/ADC/AES node on the board; board and vendor 3.12.0 libsysutils return the same error (#306) | QEMU 0.7.14 ingenic-t20 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS ...
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: no ADC node (kernel without SADC aux): SU_ADC_Init -1 errno 2 with all three, same result; nohw | QEMU 0.7.14 ingenic-t21 (#327 ...
- T23: apitest su 2026-11-11 T23 Tapo C500 (no ADC node): SU_ADC_Init -1 with board and vendor T23 1.3.0 libsysutils, same result; nohw | QEMU 0.7.14 ingenic-t23 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS ...
- T30: T30X b61 2026-11-12 (T30X lab camera, one power cycle, vendor stack then open stack by runtime swap, open-tx-isp 0fdfacb4 + OpenIMP 4aa7e40a, apitest su): no RTC/ADC node on the board ...
- T31: QEMU 0.7.14 ingenic-t31 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS, image libsysutils vs vendor T31 libsysutils in the same guest: vendor-equal or better; the test boards have no such node (board lacks the hardware, apitest N/A)
- T40: apitest su 2026-11-11 (Eufy T40XP, t40-mx, #310): OpenIMP libimp vs vendor 1.3.1 libsysutils
- T41: SU test 2026-11-11 on the Vanhua T41 test camera (gc5603, open-stack image) (#318): image libsysutils (neo-sysutils) and vendor T41 1.2.6 libsysutils (from /tmp, vendor libalog preloaded) give the same answer: Init -1 ENOENT ...
int SU_ADC_DisableChn(uint32_t chn_num);Switches an ADC channel off.
Returns: 0 on success, non-zero on error (mostly -1; T40 and T41 return their own error codes for some failures).
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? |
Tested (tool, date, camera, what was checked):
- T10: QEMU 0.7.14 ingenic-t10 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS, image libsysutils vs vendor T10 libsysutils in the same guest: vendor-equal or better; the test boards have no such node (board lacks the hardware, apitest N/A)
- T20: Wyze Cam v2 (T20X) 2026-11-11: no RTC/ADC/AES node on the board; board and vendor 3.12.0 libsysutils return the same error (#306) | QEMU 0.7.14 ingenic-t20 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS ...
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: no ADC node (kernel without SADC aux): SU_ADC_Init -1 errno 2 with all three, same result; nohw | QEMU 0.7.14 ingenic-t21 (#327 ...
- T23: apitest su 2026-11-11 T23 Tapo C500 (no ADC node): SU_ADC_Init -1 with board and vendor T23 1.3.0 libsysutils, same result; nohw | QEMU 0.7.14 ingenic-t23 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS ...
- T30: T30X b61 2026-11-12 (T30X lab camera, one power cycle, vendor stack then open stack by runtime swap, open-tx-isp 0fdfacb4 + OpenIMP 4aa7e40a, apitest su): no RTC/ADC node on the board ...
- T31: QEMU 0.7.14 ingenic-t31 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS, image libsysutils vs vendor T31 libsysutils in the same guest: vendor-equal or better; the test boards have no such node (board lacks the hardware, apitest N/A)
- T40: apitest su 2026-11-11 (Eufy T40XP, t40-mx, #310): OpenIMP libimp vs vendor 1.3.1 libsysutils
- T41: SU test 2026-11-11 on the Vanhua T41 test camera (gc5603, open-stack image) (#318): image libsysutils (neo-sysutils) and vendor T41 1.2.6 libsysutils (from /tmp, vendor libalog preloaded) give the same answer: Init -1 ENOENT ...
int SU_ADC_GetChnValue(uint32_t chn_num, int *value);Reads the current value of an ADC channel.
Returns: 0, -1 for a bad channel, NULL pointer or a failing read.
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? |
Tested (tool, date, camera, what was checked):
- T10: QEMU 0.7.14 ingenic-t10 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS, image libsysutils vs vendor T10 libsysutils in the same guest: vendor-equal or better; the test boards have no such node (board lacks the hardware, apitest N/A)
- T20: Wyze Cam v2 (T20X) 2026-11-11: no RTC/ADC/AES node on the board; board and vendor 3.12.0 libsysutils return the same error (#306) | QEMU 0.7.14 ingenic-t20 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS ...
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: no ADC node (kernel without SADC aux): SU_ADC_Init -1 errno 2 with all three, same result; nohw | QEMU 0.7.14 ingenic-t21 (#327 ...
- T23: apitest su 2026-11-11 T23 Tapo C500 (no ADC node): SU_ADC_Init -1 with board and vendor T23 1.3.0 libsysutils, same result; nohw | QEMU 0.7.14 ingenic-t23 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS ...
- T30: T30X b61 2026-11-12 (T30X lab camera, one power cycle, vendor stack then open stack by runtime swap, open-tx-isp 0fdfacb4 + OpenIMP 4aa7e40a, apitest su): no RTC/ADC node on the board ...
- T31: QEMU 0.7.14 ingenic-t31 (#327, 2026-11-11): Init/Enable/GetChnValue (3559)/Disable/Exit PASS, image libsysutils vs vendor T31 libsysutils in the same guest: vendor-equal or better; the test boards have no such node (board lacks the hardware, apitest N/A)
- T40: apitest su 2026-11-11 (Eufy T40XP, t40-mx, #310): OpenIMP libimp vs vendor 1.3.1 libsysutils
- T41: SU test 2026-11-11 on the Vanhua T41 test camera (gc5603, open-stack image) (#318): image libsysutils (neo-sysutils) and vendor T41 1.2.6 libsysutils (from /tmp, vendor libalog preloaded) give the same answer: Init -1 ENOENT ...
int SU_Key_ReadEvent(int evfd, int *keyCode, SUKeyEvent *event);Blocks until a key event arrives on the handle of SU_Key_OpenEvent; returns the key code (low 8 bits of the input code) and pressed / released.
Returns: 0, -1 on a read error.
Key press events of the gpio_keys input device only.
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅? | ✅ |
Tested (tool, date, camera, what was checked):
- T10: SU test 2026-11-12 T10 (Secuplug SP1 T10L, jxh42, open stack), image libsysutils (neo-sysutils) vs vendor T10 3.12.0 libsysutils loaded from /tmp: key press injected through /dev/input/event0 (gpio-keys, key 28, 20 ms ...
- T20: SU test 2026-11-12 T20 (Wyze Cam v2 T20X, jxf23, open stack), image libsysutils (neo-sysutils) vs vendor T20 3.12.0 libsysutils loaded from /tmp: key press injected through /dev/input/event0 (gpio-keys, key 102, 20 ms ...
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: key press injected through /dev/input/event0 (button daemon stopped): SU_Key_OpenEvent + ReadEvent give key 28 press (1) then releas ...
- T23: 2026-11-11 T23 Jooan A6M (sc1a4t, open stack, kernel 3.10.14, no /sys/power, no RTC/ADC/AES nodes), image libsysutils (neo-sysutils-1.0) vs vendor T23 ISVP 1.3.0 libsysutils (with its libalog) loaded from /tmp: key press injected through /dev/input/event0 (gpi ...
- T30: T30X b70 2026-11-13 (#439, audit/t30/b70, vendor libsysutils 1.0.5 on both stacks, node event0): inject KEY 102 -> SU_Key_ReadEvent ret 0, key 102 event 1; negative control KEY 28 times out (no event) ...
- T31: SU test 2026-11-12 T31 (Cinnado D1 T31L, sc2336, open stack), image libsysutils (neo-sysutils) vs vendor T31 1.1.6 libsysutils loaded from /tmp: key press injected through /dev/input/event0 (gpio-keys, key 28, 20 ms ...
- T40: keys 2026-11-12 .184: vendor 1.3.1 and OpenIMP both OpenEvent -1 (name gpio_keys vs gpio-keys); audit/matrix1112/keys/out-184*.txt | user decision 2026-11-12 (C4): vendor and OpenIMP fail identically (gpio_keys vs gpio-keys), rated nohw, no name fallback
int SU_Key_EnableEvent(int keyCode);Takes a hardware key out of the disabled_keys list of the gpio_keys driver again.
Returns: 0, -1 when the gpio_keys sysfs nodes are missing.
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| 🔧a | 🔧a | 🔧a | 🔧a | ✅ | ✅ | ✅? | ✅ |
Tested (tool, date, camera, what was checked):
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: image library EnableEvent -1/EACCES vs vendor 0 (issue #313), DisableEvent -1/EINVAL on all three; stays aud
- T23: 2026-11-11 T23 Jooan A6M (sc1a4t, open stack, kernel 3.10.14, no /sys/power, no RTC/ADC/AES nodes), image libsysutils (neo-sysutils-1.0) vs vendor T23 ISVP 1.3.0 libsysutils (with its libalog) loaded from /tmp: stays aud (finding #313): Disable(28) -> -1 EINVA ...
- T30: apitest/extras T30X b68 2026-11-13 (audit/t30/b68, one power cycle, vendor 1.0.5 stack then OpenIMP next c3319ab1 libimp + b67 ko, no flash, 0 oopses in su/encx): endis 28 with vendor libsysutils 1.0.5 (serves both stacks, no image library): Enable(28) ret 0 ...
- T40: keys 2026-11-12 .184: vendor 1.3.1 and OpenIMP both OpenEvent -1 (name gpio_keys vs gpio-keys); audit/matrix1112/keys/out-184*.txt | user decision 2026-11-12 (C4): vendor and OpenIMP fail identically (gpio_keys vs gpio-keys), rated nohw, no name fallback
int SU_Key_DisableEvent(int keyCode);Masks a hardware key in the gpio_keys driver (disabled_keys).
Returns: 0, -1 when the gpio_keys sysfs nodes are missing.
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅ | ✅? | ✅? | ✅ |
Tested (tool, date, camera, what was checked):
- T10: t10-mx 2026-11-11 (T10 test camera): DisableEvent(28) -1 EINVAL with the image and the vendor 3.12.0 libsysutils (board key not maskable): vendor-equal, counted nohw 2026-11-12 (mx1112-r2 ...
- T20: t20-mx 2026-11-11 (T20 test camera): DisableEvent(28 and 102) -1 EINVAL with the image and the vendor libsysutils (the kernel refuses the mask): vendor-equal, counted nohw 2026-11-12 (mx1112-r2). EnableEvent differs (#314)
- T21: SU test 2026-11-11 T21 PC420: DisableEvent -1/EINVAL with the image and the vendor T21 libsysutils 4.7.2 and 5.4.0: vendor-equal, counted nohw 2026-11-12 (mx1112-r2). EnableEvent differs (#313)
- T23: mx1112-r2 2026-11-12 T23 test camera (agg68a): image libsysutils (091d6c6a) vs vendor T23 1.3.0 libsysutils, keys 28 and 102 with disabled_keys read back: Disable(28) -1 EINVAL with both (key not maskable) ...
- T30: apitest/extras T30X b68 2026-11-13 (audit/t30/b68, one power cycle, vendor 1.0.5 stack then OpenIMP next c3319ab1 libimp + b67 ko, no flash, 0 oopses in su/encx): endis 28: Disable(28) ret 0 with the vendor libsysutils 1.0.5 (same library on both stacks) ...
- T31: mx1112-r2 2026-11-12 T31 test camera (agg68a): image libsysutils (091d6c6a) vs vendor T31 1.1.6 libsysutils, keys 28 and 102 with disabled_keys read back: Disable(28) -1 EINVAL with both (key not maskable) ...
- T40: keys 2026-11-12 .184: vendor 1.3.1 and OpenIMP both OpenEvent -1 (name gpio_keys vs gpio-keys); audit/matrix1112/keys/out-184*.txt | user decision 2026-11-12 (C4): vendor and OpenIMP fail identically (gpio_keys vs gpio-keys), rated nohw, no name fallback
int SU_LED_Command(int ledNum, SULedCmd cmd);Drives the LED on GPIO ledNum through sysfs (export, active_low 0, direction out, value). LED_GPIO_UNEXPORT writes 0 and releases the GPIO.
Returns: 0, -1 when one of the sysfs nodes cannot be opened.
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅ | ✅? |
Tested (tool, date, camera, what was checked):
- T10: both the vendor libsysutils 3.12.0 and the image libsysutils (neo-sysutils-1.0) open /proc/board/power/led; called on the T10 Secuplug SP1 with the LED GPIOs of the thingino profile (led_r GPIO 20, confirmed by the camera's gpio owner labels) and N=0/1 ...
- T20: both the vendor libsysutils 3.12.0 and the image libsysutils (neo-sysutils-1.0) open /proc/board/power/led; called on the T20 Wyze Cam v2 with the LED GPIOs of the thingino profile (led_b 39, led_y 38, active low ...
- T21: both the vendor libsysutils 1.0.33 and the image libsysutils (neo-sysutils-1.0) open /proc/board/power/led; called on the T21 PC420 with the LED GPIOs of the thingino profile (led_g 75, led_r 78, confirmed by the camera's gpio owner labels) and N=0/1 ...
- T23: both the vendor libsysutils 1.3.0 and the image libsysutils (neo-sysutils-1.0) open /proc/board/power/led; called on the T23 Jooan A6M with the LED GPIOs of the thingino profile (led_g 57, led_r 58, confirmed by the camera's gpio owner labels) and N=0/1 ...
- T30: apitest/extras T30X b68 2026-11-13 (audit/t30/b68, one power cycle, vendor 1.0.5 stack then OpenIMP next c3319ab1 libimp + b67 ko, no flash, 0 oopses in su/encx): led 0 1/0 with vendor libsysutils 1.0.5: ret -1 errno 2 ...
- T31: vendor 1.1.6 (and 1.1.2-1.1.5.2) exports no SU_LED_Command; 1.1.1 and 1.1.5-zrt open /proc/board/power/led like the image libsysutils (neo-sysutils-1.0). Called on the T31 Cinnado D1 with the profile LED GPIOs (led_b 9, led_r 8 ...
- T40: led1112 2026-11-12 Eufy T40XP: vendor libsysutils 1.3.1 (sysfs GPIO: export N, active_low 0, direction out, value; cmd 2 value 0 + unexport) vs OpenIMP libimp (src/core/su_sysutils.c) from the image, LED GPIOs from the profile (defconfig LED_B_GPIO=46 ...
- T41: both the vendor libsysutils 1.2.6 (1.0.1 to 1.2.6 alike) and the image libsysutils (neo-sysutils-1.0) open /proc/board/power/led; called on the T41 Vanhua T55A and Cinnado D1 T41NQ with the LED GPIOs of the thingino profile (led_r 79; led_b 82 ...
int SU_CIPHER_Init(void);Hardware AES engine of the vendor library. Not provided: the open stack has no AES/DES kernel driver.
Returns: -1 with errno ENOTSUP; the same for SU_CIPHER_Exit, CreateHandle, DestroyHandle, ConfigHandle, Encrypt, Decrypt, DES_Init, DES_Exit and DES_Test.
The vendor library answers -1 as well on a kernel without /dev/aes and /dev/jz-des. Software AES/DES would be an extra beyond the vendor and is not offered.
SoC: T40.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? | ✅? |
Tested (tool, date, camera, what was checked):
- T10: QEMU 0.7.14 ingenic-t10 (#327, 2026-11-11): AES device opened, handle created/destroyed, closed, image libsysutils vs vendor T10 libsysutils in the same guest: vendor-equal or better; the test boards have no /dev/aes
- T20: Wyze Cam v2 (T20X) 2026-11-11: no RTC/ADC/AES node on the board; board and vendor 3.12.0 libsysutils return the same error (#306)
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: no /dev/aes, /dev/jz-des (kernel without JZ_AES, jz-aes.ko shipped but not usable): SU_CIPHER_Init -11 errno 2 with all three ...
- T23: apitest su 2026-11-11 T23 Tapo C500 (no /dev/aes, no jz-des node): board and vendor lib lack the engine, same result; nohw Same on the Jooan A6M (2026-11-11): no /dev/rtc0, no ADC node, no /dev/aes or /dev/jz-des (#307).
- T30: apitest/extras T30X b68 2026-11-13 (audit/t30/b68, one power cycle, vendor 1.0.5 stack then OpenIMP next c3319ab1 libimp + b67 ko, no flash, 0 oopses in su/encx): su: no /dev/aes, vendor -11 and open -11 (same code) ...
- T31: QEMU 0.7.14 ingenic-t31 (#327, 2026-11-11): AES device opened, handle created/destroyed, closed, image libsysutils vs vendor T31 libsysutils in the same guest: vendor-equal or better; the test boards have no /dev/aes
- T40: apitest su 2026-11-11 (Eufy T40XP, t40-mx, #310): OpenIMP libimp vs vendor 1.3.1 libsysutils
- T41: SU test 2026-11-11 on the Vanhua T41 test camera (gc5603, open-stack image) (#318): image libsysutils (neo-sysutils) and vendor T41 1.2.6 libsysutils (from /tmp ...
int SU_Base_Shutdown(void);Syncs the file systems and tells the init process to power off.
Returns: 0 when the signal reached init, -1 (with errno) when it could not be sent.
The call returns before the shutdown has finished. Like the vendor library it sends SIGUSR2 (signal 17 on MIPS) to init after the sync, which powers off on busybox init. The environment variable OPENIMP_SU_SHUTDOWN=reboot sends SIGTERM instead (reboot); unset, poweroff and vendor mean the vendor behaviour.
| T10 | T20 | T21 | T23 | T30 | T31 | T40 | T41 |
|---|---|---|---|---|---|---|---|
| 🔧h | 🔧h | 🔧h | 🔧h | 🔧h | 🔧h | 🔧h | 🔧h |
Tested (tool, date, camera, what was checked):
- T21: SU test 2026-11-11 T21 PC420 (JXF23), open stack, image libsysutils vs vendor T21 libsysutils 4.7.2 and 5.4.0 loaded from /tmp: kill() shim on the camera: OpenIMP libimp SU_Base_Shutdown (default, poweroff and vendor modes) sends kill(1, 17) ...
Difference from the vendor library: Same as the vendor library (sync, kill(1, 17)); OpenIMP returns -1 with errno when the signal could not be sent, the vendor always returns 0.
Open Ingenic camera stack
- Home
- Supported SoCs and Sensors
- Per-SoC Status
- Feature Matrix
- Architecture
- Building
- Installing and Switching
- Module Parameters
- Module Parameters: complete list
- Environment Variables
- Beyond the Vendor
- API reference
- Memory
- Troubleshooting
- Reset Diagnostics
- Testing
- Test Pattern
- Performance
- Changelog
- Release Scheme
- Release Gates
- Website: status and light-box protocols
- Live feature matrix
- Project board
Current: RC3 (v2026.11.07-rc3), released 2026-11-07