User talk:MrNuke

Ideas for generic handling of devices

IMPORTANT! Feel free to edit this page to share your thoughts and ideas. This is the official MMC open brainstorming page.

Proposal 1: Unified API

Chan is an IO channel.

This struct is used in Inferno and has been for a long time; so it works. It's also in the opcodes somewhat like what we did for EMMC on ARM.

struct Dev
{
	char*	name;
	void	(*reset)(void);
	void	(*init)(void);
	void	(*shutdown)(void);
	Chan*	(*attach)(char*); /* tell the device you want to use it */
	Walkqid*	(*walk)(Chan*, Chan*, char**, int); /* walk to a name in the device's managed name space; return a handle */
	int	(*stat)(Chan*, uchar*, int); // status info
	Chan*	(*open)(Chan*, int); /* get access to a resource in the device name space */
	void	(*close)(Chan*); /* tell it you are done with whatever it is. */
	long	(*read)(Chan*, void*, long, vlong);
	long	(*write)(Chan*, void*, long, vlong);
	void	(*power)(int);	/* power mgt: power(1) ? on, power (0) ? off */
};

Questions

1. Do we want to expose or hide the block nature of some devces (i.e. force reading multiples of blocksize, or allow reading any number of bytes, with no alignment requirement) ?
2. How do we connect the dots, such that most of the details can be handled transparently ?
3. How simple or complex do we want the API to be such that it can work in any stage (including bootblock, assuming some SRAM and a stack are available) ?
4. Can we integrate this into libpayload such that the same sources can be used for both coreboot and libpayload ?
   • think "Device specfic storage drivers -> [*] Allwinner A10 MMC driver" in libpayload config
5. More questions coming soon (TM)

Extending CBFS to support partitioning

Reasoning

• To separate write-protected (RO) and writeable (RW) regions of flash chip
• To incorporate relevant FMAP features directly into CBFS
• To be able to better handle the requirements of ChromeOS

Tentative terminology

• Partition
  • A region indexed by CBFS
• File
  • Same meaning as before. A CBFS file that resides entirely within the boundaries of one partition.

Design constraints

• Backwards compatibility
  • New CBFS structure should be backwards compatible with old structure
  • Old payloads should still be able to access old features of CBFS

Tentative API

• cbfs_get_file(const char *name) <- Get file in main partition, also backwards compatible
• cbfs_get_partition(const char *name) <- Get location of partition other than main (or even main)
• cbfs_get_partition_file(cbfs_part_t partition, const char *name) <- Get file in given partition

Hypothetical examples

Partition in read-write portion of flash

Here, we assume that "rw1" is not in a write-protected area of flash.

void *rw1 = cbfs_get_partition("rw1");
file *romstage cbfs_get_partition_file(rw1, "fallback/romstage);
load_and_run_stage(romstage);

MRC cache

void *mrc_cache = cbfs_get_partition("mrc.cache");
/* Access raw partition, there is no file in this partition */
mrc_cache_check(mrc_cache);
mrc_cache_write(mrc_cache, data);

Technical stuff

Current CBFS master header

All values are big endian

Offset	0	1	2	3	4	5	6	7	8	9	A	B	C	D	E	F
0x00	"ORBC"				version				romsize				bootblocksize
0x10	align				offset				arch				padding

The meaning of each member is as follows:

• magic is a 32 bit number that identifies the ROM as a CBFS type. The magic number is 0x4F524243, which is 'ORBC' in ASCII.
• version is a version number for CBFS header. cbfs_header structure may be different if version is not matched.
• romsize is the size of the ROM in bytes. Coreboot will subtract 'size' from 0xFFFFFFFF to locate the beginning of the ROM in memory.
• bootblocksize is the size of bootblock reserved in firmware image.
• align is the number of bytes that each component is aligned to within the ROM. This is used to make sure that each component is aligned correctly with regards to the erase block sizes on the ROM - allowing one to replace a component at runtime without disturbing the others.
• offset is the offset of the the first CBFS component (from the start of the ROM). This is to allow for arbitrary space to be left at the beginning of the ROM for things like embedded controller firmware.
• architecture describes which architecture (x86, arm, ...) this CBFS is created for.

• PROPOSED: Change padding to a signed offset pointing to partition table; all '1's means no partition table is available
  • The offset shall be from the start of "ORBC" signature.
  • alternatively: limit number of partitions to something reasonable, move names into partitions (so partition entries are fixed size) and append the partition table to the master header. Existence proven by master_header's version field (cbfs_core ignores it right now, which is good).

Proposal 1: Variable name length partition table

All values are big endian

Entry format

Offset	0	1	2	3	4	5	6	7	8	9	A	B	C	D	E	F
0x00	"CBPT"				offset				size				region_name
0x10	region_name ...

• region_name shall be null-terminated.
• Each partition entry shall be 16 byte aligned.
• Subsequent entries shall be placed at the closest 16-byte aligned offset

Proposed partition table

Offset	0	1	2	3	4	5	6	7	8	9	A	B	C	D	E	F
0x10	CBPT1
0x_0	CBPT2
0x_0	CBPT3
0x_0	terminator all '1's

• Generally, CBPT1 will point to the same region as the master header, but this is not required. The idea is that old payloads will still be able to read CBFS by using the region pointed to by the master header. Whether this region is the same as CBPT1 or not is not guaranteed. This is to not impose a specific order on the partition entries.

Example partition table

Offset	0	1	2	3	4	5	6	7	8	9	A	B	C	D	E	F
0x00	"CBPT"				0x00070000				0x00010000				c	o	r	e
0x10	b	o	o	t	\0	0xff
0x20	"CBPT"				0x00040000				0x00020000				t	h	i	s
0x30	_	p	a	r	t	i	t	i	o	n	_	h	a	s	_	a
0x40	_	l	u	d	i	c	r	o	u	s	l	y	_	l	o	n
0x50	g	.	n	a	m	e	\0	0xff
0x60	"CBPT"				0x00060000				0x00010000				m	r	c	.
0x70	c	a	c	h	e	\0	0xff
0x80	0xffffffff

• "coreboot": 0x00070000 + 0x00010000
• "this_partition_has_a_ludicrously_long.name" 0x00040000 + 0x00020000
• "mrc.cache": 0x00070000 + 0x00010000

Questions

• Should region_name be fixed to 32-bytes, like in FMAP?
• Should we include a field for partition flags?
  • What are the use-cases
  • Would this really be useful?
  • Would this actually get used in the wild?
• Should we just use the FMAP format?

Proposal 2: Fixed name length partition table

Entry format

Offset	0	1	2	3	4	5	6	7	8	9	A	B	C	D	E	F
0x00	"CBPT"				offset				size				flags
0x10	region_name
0x20	region_name

• region_name shall be null-terminated.
  • It is always 32-bytes, but can be null-terminated earlier
  • Unused bytes will be all '1's
• This structure is marginally easier to parse as opposed to variable length

Table format

Offset	0	1	2	3	4	5	6	7	8	9	A	B	C	D	E	F
0x00	CBPT1
0x30	CBPT2
0x60	CBPT3
0x90	terminator all '1's

• Generally, CBPT1 will point to the same region as the master header, but this is not required. The idea is that old payloads will still be able to read CBFS by using the region pointed to by the master header. Whether this region is the same as CBPT1 or not is not guaranteed. This is to not impose a specific order on the partition entries.

Example table

Offset	0	1	2	3	4	5	6	7	8	9	A	B	C	D	E	F
0x00	"CBPT"				0x00070000				0x00010000				0x00000000
0x10	c	o	r	e	b	o	o	t	\0	0xff
0x20	0xff
0x30	"CBPT"				0x00040000				0x00020000				0x00000000
0x40	t	h	i	s	_	p	a	r	t	i	t	i	o	n	_	h
0x50	a	s	_	a	_	l	u	d	i	c	r	o	u	s	l	\0
0x60	"CBPT"				0x00060000				0x00010000				0x00000000
0x70	m	r	c	.	c	a	c	h	e	\0	0xff
0x80	0xffffffff
0x90	0xffffffff

• "coreboot": 0x00070000 + 0x00010000
• "this_partition_has_a_ludicrousl" 0x00040000 + 0x00020000
• "mrc.cache": 0x00070000 + 0x00010000