Documentation
¶
Overview ¶
Chips are made by vendors, and an individual vendor is defined by a 1 to 8 byte vendor id stored in the chip. An instance of a type of chip, with, e.g., a particular size, is defined by a 1 to 3 or so byte device id stored in the chip. The vendor, device id pair can be used to find both unique properties of a chip (e.g. size) and the common properties of a chip (e.g. erase value, voltages, erase blocks. etc.) It is not at all unusual to know a vendor but not a device id. Hence we need to be able to get a vendor given a vendor id, and then a chip given a chip id. It's ok, however, to know the vendor and fail to know the chip.
Sadly, device ids are not unique; they are reused per vendor. And, as mentioned, both vendor and device id are variable length. In a not uncommon failure of vision, they started out as 1 byte each, grew to 2, then 3 in some cases, 7 in other. Good times. Life would be easier if everybody just made these things strings in the beginning.
An ID identifies a vendor, but not the same vendor over time. Vendor names for a given ID change over time, due to buyouts, bankruptcies, and the occasional near depression. For example, what was AMD is now Spansion. This name changing complicates the picture a bit, so we maintain a list of vendor names for a given part, with the first name in the list being the current name. This will allow us to accomodate scripts that might have the wrong vendor name. As time goes by, and bankruptcies accumulate, this first name can change.
Hence, it is useful to have 3 bits of knowledge o list of vendor names given a vendor id o list of chips and their unique properties given a device id o list of common properties which can be referenced from a chip
We wish to embed this code in FLASH so if needed we can burn a chip from FLASH-embedded u-root.
This code uses strings, not integers, since device and vendor IDs are now variable length, depending on year of manufacture. Further, it is just nicer to work with strings.
In most cases, we will walk these tables once, so we design for exhaustive search. The tables are short and are traversed in microseconds, you only do it once, and it's important to keep data as compact as possible.
A note on flashing. Writing is not zero cost: each erase/write cycle reduces chip lifetime. Data in the chip need not be erased to be written: 0xee can be changed to 0xcc without an erase cycle in many parts. Code can make a guess a guess at an optimal erase/write pattern based on the size of the regions to be written, the content of regions, and the size of the blocks available. Getting this calculation right has proven to be tricky, as it has to balance time costs of writing, expected costs of too many erase cycles, and several other factors I can not recall just now. Watch this space.
TODO: figure out some minimum set of config options for Linux, with the proviso that this will be very kernel version dependent.
Index ¶
Constants ¶
This section is empty.
Variables ¶
var DevName = "/dev/mtd0"
DevName is the default name for the MTD device.
Functions ¶
Types ¶
type Chip ¶
type Chip interface {
// Name returns the chip name
Name() ChipName
// ID returns the chip ID
ID() ChipID
// Size returns the chip size.
Size() ChipSize
// Synonyms returns all the alternate names for a chip
Synonyms() []ChipName
// String returns as much information as one can stand about a chip.
String() string
}
Chip defines operations on Chips.
type ChipDevice ¶
type ChipDevice struct {
// contains filtered or unexported fields
}
ChipDevice has information about a chip, include Vendor, Device, sizes, and so on; and a reference to common properties. As in Vendors, there are several names for a chip.
func (*ChipDevice) Name ¶
func (c *ChipDevice) Name() ChipName
Name returns the canonical chip name.
func (*ChipDevice) String ¶
func (c *ChipDevice) String() string
String is a stringer for a ChipDevice.
func (*ChipDevice) Synonyms ¶
func (c *ChipDevice) Synonyms() []ChipName
Synonyms returns all synonyms for a chip.
type ChipID ¶
type ChipID uint64
ChipID is the integer associated with a ChipName It began as 8 bits, and never stopped growing.
type Dev ¶
Dev contains information about ongoing MTD status and operation.
func (*Dev) QueueWriteAt ¶ added in v0.9.0
QueueWrite adds a []byte to the pending write queue.
type Flasher ¶
type Flasher interface {
// ReadAt implements io.ReadAt for a flash device.
ReadAt([]byte, int64) (int, error)
// QueueWrite queues a sequence of writes into a flash device.
QueueWriteAt([]byte, int64) (int, error)
// SyncWrite assembles the queued writes and figures out a reasonable
// plan for actually writing the part.
SyncWrite() error
// Close implements io.Close for a flash device.
Close() error
// Name returns the name of the flash device
Name() string
}
Flasher defines the interface to flash drivers.
Many devices must have lazy writes; SyncWrite should always be called after a sequence of QueueWrite commands. Close should return an error if there are queued write commands. To erase a device, one calls chip Blank(), QueueWrite(), and SyncWrite(). The operators are deined for the Flasher, not the Chipper, since flashing can involve driver-level operations such as unlocking protection bits on a bridge that are more than just a chip operation.
type Vendor ¶
type Vendor interface {
// Chip returns a Chip, given a DeviceID
Chip(ChipID) (Chip, error)
// ID Returns the VendorID
ID() VendorID
// Name() returns the canonical name
Name() VendorName
// Synonyms returns all the names
Synonyms() []VendorName
}
Vendor defines operations on vendor data.
func VendorFromID ¶
VendorFromID returns a Vendor or error given a VendorID.
func VendorFromName ¶
func VendorFromName(v VendorName) (Vendor, error)
VendorFromName returns a Vendor or error given a VendorName.
Source Files