2008-11-08 11:03:55

by Stefan Richter

Subject: [PATCH] firewire: detect and log TSB81BA3 erratum

While testing a larger 1394b bus, I saw the following Texas Instruments
TSB81BA3 hardware bug at work: On a 1394b-only bus with three or more
PHYs and differing cable lengths or differing hop speeds, the PHYs
repeatedly cause bus resets during traffic on the bus, as documented in
TI's "TSB81BA3 Errata (Rev. C)" SLLZ015C. These resets cause massive
I/O failures.

Self-ID complete events after such resets consistently show more than
one self-ID zero packet with i bit set ( = initiated reset). If this
happens, emit a warning in dmesg and suggest hardware workarounds to the
user. (There is no software workaround.)

The hardware bug is fixed in TSB81BA3 revision D but there are probably
many FireWire 800 products out there which have a buggy version of this
chip.

Signed-off-by: Stefan Richter <[email protected]>
---
drivers/firewire/fw-topology.c | 13 ++++++++++++-
1 file changed, 12 insertions(+), 1 deletion(-)

Index: linux/drivers/firewire/fw-topology.c
===================================================================
--- linux.orig/drivers/firewire/fw-topology.c
+++ linux/drivers/firewire/fw-topology.c
@@ -179,7 +179,7 @@ static struct fw_node *build_tree(struct
struct list_head stack, *h;
u32 *next_sid, *end, q;
int i, port_count, child_port_count, phy_id, parent_count, stack_depth;
- int gap_count;
+ int gap_count, reset_initiators;
bool beta_repeaters_present;

local_node = NULL;
@@ -190,6 +190,7 @@ static struct fw_node *build_tree(struct
phy_id = 0;
irm_node = NULL;
gap_count = SELF_ID_GAP_COUNT(*sid);
+ reset_initiators = 0;
beta_repeaters_present = false;

while (sid < end) {
@@ -285,6 +286,8 @@ static struct fw_node *build_tree(struct
list_add_tail(&node->link, &stack);
stack_depth += 1 - child_port_count;

+ reset_initiators += SELF_ID_PHY_INITIATOR(q);
+
if (node->phy_speed == SCODE_BETA &&
parent_count + child_port_count > 1)
beta_repeaters_present = true;
@@ -307,6 +310,14 @@ static struct fw_node *build_tree(struct
card->gap_count = gap_count;
card->beta_repeaters_present = beta_repeaters_present;

+ if (reset_initiators != 1) {
+ fw_error("%d bus reset initiators, hardware bug?\n",
+ reset_initiators);
+ if (beta_repeaters_present)
+ fw_error("try cables with matched length "
+ "or add a 1394a node\n");
+ }
+
return local_node;
}


--
Stefan Richter
-=====-==--- =-== -=---
http://arcgraph.de/sr/

2008-11-08 19:46:33

by Stefan Richter

Subject: Re: [PATCH] firewire: detect and log TSB81BA3 erratum

I wrote:
[...]
> TI's "TSB81BA3 Errata (Rev. C)" SLLZ015C. These resets cause massive
> I/O failures.
>
> Self-ID complete events after such resets consistently show more than
> one self-ID zero packet with i bit set ( = initiated reset). If this
> happens, emit a warning in dmesg and suggest hardware workarounds to the
> user.

Apparently two initiators are OK when a node joins a bus. My alarm goes
off too quickly.
--
Stefan Richter
-=====-==--- =-== -=---
http://arcgraph.de/sr/