From a09060ffe516a0e55f29c89b7da2da760c9487d7 Mon Sep 17 00:00:00 2001 From: Jeff Garzik Date: Mon, 28 May 2007 08:17:06 -0400 Subject: [libata] sata_sx4, sata_via: minor documentation updates sata_sx4: - describe overall driver theory of operation - add a few constants that will be used in the future sata_via: - remove mention of an old-EH function that is going away Signed-off-by: Jeff Garzik --- drivers/ata/sata_sx4.c | 51 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 51 insertions(+) (limited to 'drivers/ata/sata_sx4.c') diff --git a/drivers/ata/sata_sx4.c b/drivers/ata/sata_sx4.c index ff0a78dc8b8..3d83ee27338 100644 --- a/drivers/ata/sata_sx4.c +++ b/drivers/ata/sata_sx4.c @@ -30,6 +30,54 @@ * */ +/* + Theory of operation + ------------------- + + The SX4 (PDC20621) chip features a single Host DMA (HDMA) copy + engine, DIMM memory, and four ATA engines (one per SATA port). + Data is copied to/from DIMM memory by the HDMA engine, before + handing off to one (or more) of the ATA engines. The ATA + engines operate solely on DIMM memory. + + The SX4 behaves like a PATA chip, with no SATA controls or + knowledge whatsoever, leading to the presumption that + PATA<->SATA bridges exist on SX4 boards, external to the + PDC20621 chip itself. + + The chip is quite capable, supporting an XOR engine and linked + hardware commands (permits a string to transactions to be + submitted and waited-on as a single unit), and an optional + microprocessor. + + The limiting factor is largely software. This Linux driver was + written to multiplex the single HDMA engine to copy disk + transactions into a fixed DIMM memory space, from where an ATA + engine takes over. As a result, each WRITE looks like this: + + submit HDMA packet to hardware + hardware copies data from system memory to DIMM + hardware raises interrupt + + submit ATA packet to hardware + hardware executes ATA WRITE command, w/ data in DIMM + hardware raises interrupt + + and each READ looks like this: + + submit ATA packet to hardware + hardware executes ATA READ command, w/ data in DIMM + hardware raises interrupt + + submit HDMA packet to hardware + hardware copies data from DIMM to system memory + hardware raises interrupt + + This is a very slow, lock-step way of doing things that can + certainly be improved by motivated kernel hackers. + + */ + #include #include #include @@ -58,6 +106,8 @@ enum { PDC_INT_SEQMASK = 0x40, /* Mask of asserted SEQ INTs */ PDC_HDMA_CTLSTAT = 0x12C, /* Host DMA control / status */ + PDC_CTLSTAT = 0x60, /* IDEn control / status */ + PDC_20621_SEQCTL = 0x400, PDC_20621_SEQMASK = 0x480, PDC_20621_GENERAL_CTL = 0x484, @@ -89,6 +139,7 @@ enum { PDC_MASK_INT = (1 << 10), /* HDMA/ATA mask int */ PDC_RESET = (1 << 11), /* HDMA/ATA reset */ + PDC_DMA_ENABLE = (1 << 7), /* DMA start/stop */ PDC_MAX_HDMA = 32, PDC_HDMA_Q_MASK = (PDC_MAX_HDMA - 1), -- cgit v1.2.3