Re: [9fans] sdkw - SATA drive detection broken (was ARM and u-boot)

[erik quanstrom <quanstro@quanstro.net>]

this is fun.  i wrote this driver a long time ago.  but all the little
quirks (even of the documentation) come back quickly.  i put
the kernel i'm running up on 9atom  you can download them
@
	http://ftp.9atom.org/other/9plug
	http://ftp.9atom.org/other/s9plug
or
	9fs atom; fcp /n/atom/ftp/*9plug /some/where

please try that kernel and let me know.  since the driver
does hotplugging, it might be after you've connected to the file
server that you get a connection.

also try unplugging the usb devices.  the host bridge error is
omnious.  often a host bridge error will screw up multiple devices.

here's the results i get

	kw# cat /dev/kmesg
	127 holes free
	009a9000 03be5000 52674560
	52674560 bytes free
	l1 D: 16384 bytes, 4 ways 128 sets 32 bytes/line; write-through only
	l1 I: 16384 bytes, 4 ways 128 sets 32 bytes/line; write-back type `reg 7 ops, format C' (016) possible
	l2 cache: 256K or 512K: 4 ways, 32-byte lines, write-back, sdram only
	cpu0: 1200MHz ARM Marvell 88F6281 A0; arm926ej-s arch v5te rev 2.1 part 131
	#S/sd0: sata ii 2 ports
	#F0: kwnand: Hy27UF084G2M 536,870,912 bytes pagesize 2048 erasesize 131,072 spares per page 64
	#l0: 88e1116: 1000Mbps port 0xf1072000 irq 11 tu 1514: 00504301db37
	#l1: 88e1116: 1000Mbps port 0xf1076000 irq 15 tu 1514: 00504301db38
	#u/usb/ep1.0: ehci: port 0xf1050100 irq 19
	preallocate 16384 x 4096 KB 0x03be5000-0x07be5000
	504M memory: 52M kernel data, 452M user, 0M swap
	usb/hub... version...time...
	
	init: starting /bin/rc
	kw# sd01: status: 000 -> 123: new
	sd01: llba 78,165,360 sectors
	  INTEL SSDSA2M040G2GC 2CV102HD CVGB03800093040NGN [newdrive]


	kw# cat /dev/sd00/ctl
	inquiry 
	state	null
	sig	00000000
	link	down
	sstatus	00000000
	serror	00000000
	sctl	00000300
	isr	00804000
	icfg	009b7095
	ifccr	00000000
	geometry 0 0


	kw# cat /dev/sd01/ctl
	inquiry INTEL SSDSA2M040G2GC
	state	ready
	sig	01010101
	model	INTEL SSDSA2M040G2GC
	serial	CVGB03800093040NGN
	firm	2CV102HD
	wwn	50015179593f82f0
	tler	5000
	link	up
	sstatus	00000123
	serror	14010000
	sctl	00000300
	isr	00404034
	icfg	009b7095
	ifccr	00000000
	geometry 78165360 512
	part data 0 78165360

you mention that the first thing you do is this

	d->reg[Sctl] = 3*Aipm | 0*Aspd | Adet; (I’m omitting | 3*Aspm as it done in NetBSD, though the results are bad even with it).

this doesn't jive with the sata spec (the ahci spec, which for the S registers
mirrors it, is more accessible).  it may be beneficial to read the spec.
the actions that need to be done are
	1.  clear the Serror register (sdkw:1228)
	2.  set the Icfg register (sdkw:1229)
		(the setting of emphasis was a guess and could be wrong for
		your device.  it may also vary by device.)
	3.  reset the phy.  this is a 3 step process in linkrst()
		a) turn on device detection
		b) wait 1ms  (unsure of this timing,  it could be longer)
		c) turn off device detection

3*Aspm requests transition to active state.  if you get bad results
by turning the device on, then something else is wrong.  :-)

in fact you mention that Sctl = 300, which means that the device
is asleep. (Iactive|Isleepy if you follow pc/ahci.h)

> Not changed at all… WTF?

you must clear the error register first.

> Maybe NetBSD SoC driver code does some init, which Plan 9 doesn’t?

i don't think that's it.

- erik