This is something I'd tripped over earlier, and wanted to follow up on
to get an acceptable fix in for everyone's benefit before it falls through
the cracks again.
There seems to be an issue with recent 83xx/85xx SOC UARTs, in which a break
triggers a short lived IRQ storm (hence killing any hope of using SysRQ).
The only fix I found to work was to just ignore the bogus events that
had the associated signature bit set.
This fix is what I was using against earlier kernels, but I hate to add more
board/arch specific ifdefs to files like 8250.c, so I'm wondering if
anyone has any other suggestions before I simply end up cleaning up the
boardlist (now ppc is dead) and respinning the patch much as it is now
and resending.
Thanks,
Paul.
-----------------
Variation/update of version1 of the patch -- orig. discussion at:
http://patchwork.ozlabs.org/linuxppc/patch?id=15986
Update is: Don't use MPC8540 as a selector for the 8548 boards,
since testing on a genuine (older) MPC8540ADS board shows that
it doesn't have the issue. This will get coverage on the
sbc8349, fsl_mpc8349_mitx, fsl_mpc832xe (and the similar 8323RDB)
and the sbc8548 and the fsl_8548cds. (i.e all the boards where
the problem has been demonstrated to exist so far.)
Signed-off-by: Paul Gortmaker <[email protected]>
---
drivers/serial/8250.c | 17 +++++++++++++++++
include/linux/serial_reg.h | 1 +
2 files changed, 18 insertions(+)
--- a/drivers/serial/8250.c
+++ b/drivers/serial/8250.c
@@ -1347,6 +1347,25 @@ serial8250_handle_port(struct uart_8250_
status = serial_inp(up, UART_LSR);
+#if defined(CONFIG_MPC834x) || defined(CONFIG_MPC85xx_CDS) || \
+ defined(CONFIG_SBC8548) || defined (CONFIG_PPC_MPC832x)
+ /*
+ * There appears to be a quirk in the implementation on some 8xxx
+ * where after a break is rec'd (UART_LSR_BI), the UART generates
+ * a short duration burst of bogus IRQ events with the signature
+ * of RFE set (along with "normal" bits set) in the LSR.
+ */
+
+#define RFE_8xxx_ERR_BITS ( UART_LSR_RFE | UART_LSR_TEMT | \
+ UART_LSR_THRE | UART_LSR_BI | \
+ UART_LSR_DR )
+
+ if (status == RFE_8xxx_ERR_BITS) {
+ spin_unlock(&up->port.lock);
+ return;
+ }
+#endif
+
DEBUG_INTR("status = %x...", status);
if (status & UART_LSR_DR) {
--- a/include/linux/serial_reg.h
+++ b/include/linux/serial_reg.h
@@ -109,6 +109,7 @@
#define UART_MCR_DTR 0x01 /* DTR complement */
#define UART_LSR 5 /* In: Line Status Register */
+#define UART_LSR_RFE 0x80 /* Rx FIFO Error (BE, FE, or PE) */
#define UART_LSR_TEMT 0x40 /* Transmitter empty */
#define UART_LSR_THRE 0x20 /* Transmit-hold-register empty */
#define UART_LSR_BI 0x10 /* Break interrupt indicator */
On Jul 31, 2008, at 10:26 AM, Paul Gortmaker wrote:
>
> This is something I'd tripped over earlier, and wanted to follow up on
> to get an acceptable fix in for everyone's benefit before it falls
> through
> the cracks again.
>
> There seems to be an issue with recent 83xx/85xx SOC UARTs, in which
> a break
> triggers a short lived IRQ storm (hence killing any hope of using
> SysRQ).
> The only fix I found to work was to just ignore the bogus events that
> had the associated signature bit set.
>
> This fix is what I was using against earlier kernels, but I hate to
> add more
> board/arch specific ifdefs to files like 8250.c, so I'm wondering if
> anyone has any other suggestions before I simply end up cleaning up
> the
> boardlist (now ppc is dead) and respinning the patch much as it is now
> and resending.
How did you test this or generate it? I was thinking about this the
other day and figured we need to try and track this down with the HW
guys. If we can generate a simple test I can try and run it through
the various boards/parts we have and see which ones show the issue and
which dont.
- k
On Thu, Jul 31, 2008 at 4:31 PM, Kumar Gala <[email protected]> wrote:
>
> On Jul 31, 2008, at 10:26 AM, Paul Gortmaker wrote:
>
>>
>> This is something I'd tripped over earlier, and wanted to follow up on
>> to get an acceptable fix in for everyone's benefit before it falls through
>> the cracks again.
>>
>> There seems to be an issue with recent 83xx/85xx SOC UARTs, in which a
>> break
>> triggers a short lived IRQ storm (hence killing any hope of using SysRQ).
>> The only fix I found to work was to just ignore the bogus events that
>> had the associated signature bit set.
>>
>> This fix is what I was using against earlier kernels, but I hate to add
>> more
>> board/arch specific ifdefs to files like 8250.c, so I'm wondering if
>> anyone has any other suggestions before I simply end up cleaning up the
>> boardlist (now ppc is dead) and respinning the patch much as it is now
>> and resending.
>
> How did you test this or generate it? I was thinking about this the other
> day and figured we need to try and track this down with the HW guys. If we
> can generate a simple test I can try and run it through the various
> boards/parts we have and see which ones show the issue and which dont.
Actually, it is dead easy to reproduce. The obvious symptom is that you'll
get a random complaint that SysRQ doesn't work. You send the break via
whatever comm program you use, and 9 out of 10 times, you get the SysRQ
help menu pop up immediately, before you've even entered another character.
Once you dig a bit deeper, then you see that simply sending the break via
the comm program (and nothing else) will result in roughly 300 to 2000 events
as reported via "cat /proc/interrupts" on the UART in question.
I've tested with at least 5 different boards, with minicom, ckermit and even
with a commercial digiport in the path, and in all cases it came up the same.
Paul.
>
> - k
> --
> To unsubscribe from this list: send the line "unsubscribe linux-serial" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
>