SELF

SELF stands for Simple Executable Loader Format. It is a very simple take on the standard [ELF] executable format. It is proposed that SELF be the the standard format for payload code stored in a LAR file and loaded and executed by coreboot-v3. The bayou chooser will also load and run payloads in the SELF format.

Description

Each SELF file is defined as a group of different segments. Each segment either loads data into memory, zeroes a section of memory, or provides information to coreboot or the payload. The segments can be one of the following types:

ID	Segment	Description
'CODE' (0x45444F43)	CODE	This segment has executable code to be copied from the data section to memory. Each CODE segment has a corresponding block of data, which may be compressed in the LAR. Multiple segments of this type are allowed.
'DATA' (0x41543144)	DATA	This segment has non-executable data to be copied from the data section into memory. Each DATA segment has a corresponding block of data, which may be compressed in the LAR. Multiple segments of this type are allowed.
'BSS ' (0x20535342)	BSS	This segment defines a section of memory to be zeroed. BSS segments do not have any data associated with them. Multiple segments of this type are allowed.
'NAME' (0x434D4145)	NAME	This segment contains a descriptive string in the data section that is used by the payload loader to identify the current payload. NAME segments have corresponding blocks of data, but they are not loaded into system memory. The segment data must not be compressed. Multiple segments of this type are allowed, but likely won't be used by the payload loader.
'NOTE' (0x45544F4E)	NOTES	This segment contains the contents of the .notes section in the data segment. NOTES segments have corresponding blocks of data, but they are not loaded into system memory. The segment data must not be compressed. Multiple segments of this type are allowed.
'ENTR' (0x52544E45)	ENTRY	This segment defines the entry point for execution. This type signifies the end of the list of segments. It is mandatory and can only be used once.

Format

A SELF file is comprised of two parts: the segment table and the data section. Each is described below.

Segment Table

The segment table is located at the start of the ELF file, with one entry per segement. Each segment is identified by its type (as listed above). The structure of the header entry is as follows:

struct self_header {
    unsigned long type;
    unsigned long offset;
    unsigned long long load_addr;
    unsigned long len;
    unsigned long mem_len;
};

All fields are present in each segment entry, though not every field will be valid for every segment type. The value of each field will be stored in little endian format. Big endian architectures will need to byte swap the value before using it. The following table lists which members are used and what they mean:

Type	offset	load_addr	len	mem_len
CODE	Offset of the segment in the SELF file	address where the code should be copied in memory	length of the data in the segment data section	length of the data in memory
DATA	Offset of the segment in the SELF file	address where the code should be copied in memory	length of the data in the segment data section	length of the data in memory
BSS	N/A	Start address of the block in memory to be zeroed	N/A	Length of the block in memory to be zeroed
NAME	Offset of the segment in the SELF file	N/A	length of the data in the segment data section	N/A
NOTES	Offset of the segment in the SELF file	N/A	length of the data in the segment data section	N/A
ENTRY	N/A	Address where execution should start	N/A	N/A

Segment Section

The data section immediately follows the final entry in the segment table. Each block of segment data is written sequentially in this section, with the start of each block aligned to a 32 bit boundary. CODE and DATA sections may be compressed according to the LAR compression scheme, NAME and NOTES sections must be uncompressed.

Typical Use

There are two usage models for the SELF. The first is the payload loader that wishes to determine the name of the payload for a graphical chooser. The following is the psuedo code for accomplishing this:

get_name(char *name) {
  ptr = SELF_start;
  struct self_header *header;

  do {
    header = (struct self_header *) ptr;

    if (header->type == SELF_NAME) {
           memcpy(name, SELF_start + header->offset, header->size);
           return 0;
    }
    
    ptr += sizeof(*header);
  } while(header->type != SELF_ENTRY);

  return -1;
}
   

The second usage model is actually loading and running the code - this is accomplished in a stream fashion like so:

decompress_and_run(void) {
  ptr = SELF_start;

  while(1) {
    header = (struct self_header *) ptr;
    switch(header->type) {
       case TYPE_CODE:
       case TYPE_DATA:
           dlen = decompress(SELF_start + header->offset, header->load_addr, header->len);
           memset(header->load_addr + dlen, 0, header->mem_len - dlen);
           break;
       case BSS:
            memset(header->load_addr, 0, header->len);
            break;
       case LOAD:
            return jump_to(header->load_addr);
    }
    
    ptr += sizeof(*header);
  }
}

Changelog

• Chage ID of each segment to 4 text characters
• Each block of data needs to be 32 bit aligned
• Specified that all the segment header members are stored in little endian
• Changed the load address in the segment header structure to 64 bits.
• Clarified the compression of the code / data segments