[PATCH V2] workaround for mpc52xx erratum #364 (serial may not be reset in break state)

Grant Likely schrieb:

On Mon, Nov 3, 2008 at 12:32 PM, René Bürgel [off-list ref] wrote:
  

quoted

Hi

This patch is a workaround for bug #364 found in the MPC52xx processor.
The errata document can be found under
http://www.freescale.com/files/32bit/doc/errata/MPC5200E.pdf?fpsp=1&WT_TYPE=Errata&WT_VENDOR=FREESCALE&WT_FILE_FORMAT=pdf&WT_ASSET=Documentation

    

This is an MPC5200 errata.  It doesn't exist on the MPC5200B.  The
bugfix code should be enabled at runtime only if running on the
original MPC5200.  You can use the value of the compatible property to
determine whether or not to enable it.  Optionally you can further
reduce impact by checking if CONFIG_PPC_MPC5200_BUGFIX is defined.

  

This bug is definetly present on the MPC5200B, although it is not listed in the errata sheet. I've had the ability to test it. We have custom boards using CPU modules from TQ with both processor versions, MPC5200 and MPC5200B. Both versions show exactly the same behaviour and the patch fixes it.

But as the serial driver is also used for the MPC5121, we may have to distinguish anyway. Does anyone have the possibility to test if the bug in still present  on MPC5121?
I disabled the bugfix on MPC5121-processors.

quoted

diff --git a/drivers/serial/mpc52xx_uart.c b/drivers/serial/mpc52xx_uart.c
index 6117d3d..929524b 100644
--- a/drivers/serial/mpc52xx_uart.c
+++ b/drivers/serial/mpc52xx_uart.c

@@ -496,6 +496,27 @@ mpc52xx_uart_break_ctl(struct uart_port *port, int ctl)
       spin_unlock_irqrestore(&port->lock, flags);
 }

+/* macro with helper macros to safely reset rx which mustn't be done in

break state.
+ * This is a workaround for processor bug #364 described in MPC5200 (L25R)
Errata.
+ *
+ * The workaround resets the baudrate to 4800, waits for a stable state and
resets break state repeatedly if necessary
+ * optionally it can release the lock while waiting.
+ * 1 character at 4800 baud takes 2ms, 3ms should be enough for 1 character
at higher speed and 1 char at lowest
+ * works only with longer delays
+ */
+#define LOCK(code) code
+#define DONT_LOCK(code)
+#define mpc52xx_uart_reset_rx(LOCK)
                   \
+       out_8(&psc->ctur,0x01);
                    \
+       out_8(&psc->ctlr,0xae);
                    \
+       do {
                   \
+               out_8(&psc->command,MPC52xx_PSC_RST_ERR_STAT);
                   \
+               LOCK(disable_irq(port->irq);
spin_unlock_irqrestore(&port->lock, flags));       \
+               mdelay(10);
                    \
+               LOCK(spin_lock_irqsave(&port->lock, flags);
enable_irq(port->irq));             \
+       } while ((in_be16(&psc->mpc52xx_psc_status)) & MPC52xx_PSC_SR_RB);
                   \
+       out_8(&psc->command,MPC52xx_PSC_RST_RX);
+
    

ick.  ugly.  Don't mess about with a macro here, just write a
function.  Let the compiler decide if it should be inlined.

g.

  

The purpose of the macro wasn't to enforce inlining, but to enhance the readability of the call and to ensure, that the was no runtime-check needed to decide whether the lock has to be hold or may be released. I'd like to avoid the runtime-checking here Anyway, i transformed the macro into a function like suggested. Any  proposals on this are welcome :)

-- 
René Bürgel
Software Engineer
Unicontrol Systemtechnik GmbH
OT Dittersbach
Sachsenburger Weg 34
09669 Frankenberg

Tel.: 03 72 06/ 88 73 - 19
Fax: 03 72 06/ 88 73 - 60
E-Mail: r.buergel@unicontrol.de
Internet: www.unicontrol.de

Unicontrol Systemtechnik GmbH
Geschäftsführer: Dipl.-Ing. Siegfried Heinze
Sitz der Gesellschaft: Frankenberg
Registergericht: Amtsgericht Chemnitz, HRB 15 475