Documentation/superio: add generic PNP device documentation
Signed-off-by: Felix Held <firstname.lastname@example.org>
Tested-by: build bot (Jenkins) <email@example.com>
Reviewed-by: Paul Menzel <firstname.lastname@example.org>
Reviewed-by: Patrick Rudolph <email@example.com>
diff --git a/Documentation/superio/common/pnp.md b/Documentation/superio/common/pnp.md
new file mode 100644
@@ -0,0 +1,114 @@
+# PNP devices
+Typical PNP devices are Super I/Os, LPC-connected TPMs and board
+PNP devices are usually connected to the LPC or eSPI bus of a system
+and shouldn't be confused with PCI(e) devices that use a completely
+different plug and play mechanism. PNP originates in the ISA plug and
+play specification. Since the original ISA bus is more or less extinct,
+the auto-detection part of ISA PNP is mostly irrelevant nowadays. For
+the register offsets for different functionality, appendix A of that
+specification is still the main reference though.
+## Configuration access and config mode
+Super I/O chips connected via LPC to the southbridge usually have their
+I/O-mapped configuration interface with a size of two bytes at the base
+address 0x2e or 0x4e. Other PNP devices have their configuration
+interface at other addresses.
+The two byte registers allow access to an indirect 256 bytes big
+register space that contains the configuration. By writing the index
+to the lower byte (e.g. 0x2e), you can access the register contents at
+that index by reading/writing the higher byte (e.g. 0x2f).
+To prevent accidental changes of the Super I/O (SIO) configuration,
+the SIOs need a configuration mode unlock sequence. After changing the
+configuration, the configuration mode should be left again, by sending
+the configuration mode lock sequence.
+## Logical device numbers (LDN)
+Each PNP device can contain multiple logical devices. The bytes from
+0x00 to 0x2f in the indirect configuration register space are common
+for all LDNs, but some SIO chips require a certain LDN to be selected
+in order to write certain registers in there. An LDN gets selected by
+writing the LDN number to the LDN select register 0x07. Registers 0x30
+to 0xFF are specific to each LDN number.
+coreboot encodes the physical LDN number in the lower byte of the LDN
+### Virtual logical device numbers
+Register 0x30 is the LDN enable register and since it is an 8 bit
+register, it can contain up to 8 enable bits for different parts of
+the functionality of that logical device. To set a certain enable bit
+in one physical LDN, the concept of virtual LDNs was introduced.
+Virtual LDNs share the registers of their base LDN, but allow to
+specify which part of a LDN should be enabled.
+coreboot encodes the enable bit number and by that the virtual LDN
+part in the lower 3 bits of the higher byte of the LDN number.
+## I/O resources
+Starting at register address 0x60, each LDN has 2 byte wide I/O base
+address registers. The size of an I/O resource is always a power of
+### I/O resource masks
+The I/O resource masks encode both the size and the maximum base
+address of the corresponding IO resource. The number of zeros counted
+from the least significant bit encode the resource size. If N is the
+number of LSBs being zero, which can also be zero if the LSB is a one,
+the resource has N address bits and a size of 2\*\*N bytes. The mask
+address is also the highest possible address to map the I/O region.
+A typical example for an I/O resource mask is 0x07f8 which is
+0b0000011111111000 in binary notation. The three LSBs are zeros here,
+so it's an eight byte I/O resource with three address offset bits
+inside the resource. The highest base address it can be mapped to is
+0x07f8, so the region will end at 0x07ff.
+The Super I/O datasheets typically contain the information about the
+I/O resource masks. On most Super I/O chips the mask can also be found
+out by writing 0xffff to the corresponding I/O base address register
+and reading back the value; since the lowest and highest bits are
+hard-wired to zero according to the I/O resource size and maximal
+possible I/O address, this gives the mask.
+## IRQ resources
+Each physical LDN has up to two configurable interrupt request
+register pairs 0x70, 0x71 and 0x72, 0x73. Each pair can be configured
+to use a certain IRQ number. Writing 1 to 15 into the first register
+selects the IRQ number generated by the corresponding IRQ source and
+enables IRQ generation; writing 0 to it disables the generation of
+IRQs for the source. The second register selects the IRQ type (level
+or edge) and IRQ level (high or low). For LPC SIOs the IRQ type is
+hard-wired to edge.
+On the LPC bus a shared SERIRQ line is used to signal IRQs to the
+host; the IRQ number gets encoded by the number of LPC clock cycles
+after the start frame before the device pulls the open drain
+SERIRQ can be used in two different modes: In the continuous SERIRQ
+mode the host continuously sends IRQ frame starts and the devices
+signal their IRQ request by pulling low the SERIRQ line at the right
+time. In quiet SERIRQ mode the host doesn't send IRQ frame starts, so
+the devices have to send both the IRQ frame start and the encoded IRQ
+number. The quiet mode is often broken.
+## DRQ resources
+Each physical LDN has two legacy ISA-style DMA request channel
+registers at 0x74 and 0x75. Those are only used for legacy devices
+like parallel printer ports or floppy disk controllers.
+Each device using LPC legacy DMA needs its own LDMA line to the host.
+Some newer chipsets have dropped the LDMA line and with that the
+legacy DMA capability on LPC.
diff --git a/Documentation/superio/index.md b/Documentation/superio/index.md
index 39965fd..053663b 100644
@@ -7,4 +7,5 @@
+- [PNP devices](common/pnp.md)
- [SSDT generator for generic SuperIOs](common/ssdt.md)