[ECOS] SMC2 - no interrupts?

[ECOS] SMC2 - no interrupts?

[Jamie Guinan]

Hi all,

[ Note - I'll also send a formal bug report related to this under
  my client's support contract, so some Cygnus folks may see this twice. ]

I'm using eCos on an MBX 821 board, and I'm trying to coerce SMC2
into a functioning state.  A few weeks ago I got as far as figuring out
I needed a transceiver (duh), and I verified that the port was operational
by sending EPPCBug's console out SMC2.

I've been studying quicc_smc_serial.c, and I'm a bit stumped as to
why nothing is coming out the SMC2 port in my test application.

My simple test app is attached, but in a nutshell, the code goes,

  const char *device = "/dev/ser2";
  char *msg;
  cyg_io_handle_t handle;
  int len;

  cyg_io_lookup(device,  &handle);
  msg = "Hello from SMC2\n";
  len = strlen(msg);
  cyg_io_write(handle, msg, &len);

and nothing comes out the port.  Debugging the driver, I found that
after it fills the transmit buffer, it calls quicc_smc_serial_flush(),
which sets the R and I bits (Ready/Interrupt) in the tx buffer descriptor,
so that an interrupt will be generated once the data is sent.  It *should*
work, but since nothing was happening I decided to add some counter 
variables (see attached patch file) and guess what- neither
quicc_smc_serial_ISR() nor quicc_smc_serial_DSR() seem to get called.  :/

I'm working out of CVS at the moment (updated late night, Nov 8th).
My configuration goes,

  tclsh .../packages/pkgconf.tcl --target powerpc --platform mbx --startup ram --disable-uitron

Then I manually tweak some pkgconf settings,

hal.h:
#define CYGIMP_HAL_COMMON_INTERRUPTS_USE_INTERRUPT_STACK
#define CYGNUM_HAL_COMMON_INTERRUPTS_STACK_SIZE 4096
#undef  CYGSEM_HAL_COMMON_INTERRUPTS_ALLOW_NESTING
#undef  CYGDBG_HAL_COMMON_INTERRUPTS_SAVE_MINIMUM_CONTEXT
#undef  CYGIMP_HAL_COMMON_INTERRUPTS_CHAIN
#undef  CYGIMP_HAL_COMMON_INTERRUPTS_IGNORE_SPURIOUS
#undef  CYGDBG_HAL_COMMON_CONTEXT_SAVE_MINIMUM
#define  CYGDBG_HAL_DEBUG_GDB_INCLUDE_STUBS
#define  CYGDBG_HAL_DEBUG_GDB_BREAK_SUPPORT
#undef   CYGDBG_HAL_DEBUG_GDB_CTRLC_SUPPORT
#define  CYGDBG_HAL_DEBUG_GDB_THREAD_SUPPORT

io_serial.h:
#define CYGPKG_IO_SERIAL_POWERPC_QUICC_SMC
#define CYGPKG_IO_SERIAL_POWERPC_QUICC_SMC_SMC1
(defaults, 38400, etc.)
#define CYGPKG_IO_SERIAL_POWERPC_QUICC_SMC_SMC2
(defaults, 38400, etc.)

hal_powerpc_mbx.h:
#define CYGHWR_HAL_POWERPC_BOARD_SPEED       40

infra.h:
#define CYGPKG_INFRA_DEBUG
#define CYGDBG_INFRA_DIAG_USE_DEVICE

Is there anything else I need to do to get the interrupts into
operation?  Maybe its a cache issue?

For what its worth, here is a dump of the smc_chan structure passed
to quicc_smc_serial_flush().  It looks Ok to me (pram, txbd, etc. seem 
reasonable at least),

(gdb) p *smc_chan
$1 = {channel = 176, int_num = 55, brg = 0xff0009f4, pram = 0xff003f80,
ctl = 0xff000a90, 
  txbd = 0xff002840, rxbd = 0xff002860, tbase = 0xff002840, rbase =
0xff002860, txsize = 16, 
  rxsize = 16, serial_interrupt = {vector = 55, priority = 0, isr =
0x130dc <quicc_smc_serial_ISR>, 
    dsr = 0x12ce0 <quicc_smc_serial_DSR>, data = 322024, dsr_count = 0,
next_dsr = 0x0}, 
  serial_interrupt_handle = 322208, tx_enabled = 0}


Thanks in advance...
-Jamie

================================================================
Jamie Guinan                         Blue Button Solutions, Inc.
guinan@bluebutton.com                  http://www.bluebutton.com
================================================================

Re: [ECOS] SMC2 - no interrupts?

[Jesper Skov]

>>>>> "Jamie" == Jamie Guinan <guinan@bluebutton.com> writes:

Jamie> Is there anything else I need to do to get the interrupts into
Jamie> operation?  Maybe its a cache issue?

It's a long shot, but you may have to prime the transmitter. See the
cogent serial driver:

// Enable the transmitter on the device
static void
cogent_serial_start_xmit(serial_channel *chan)
{
    cogent_serial_info *cogent_chan = (cogent_serial_info *)chan->dev_priv;
    cyg_addrword_t port = cogent_chan->base;
    cyg_uint8 _ier;

    HAL_READ_UINT8(port+SER_16550_IER, _ier);
    _ier |= IER_XMT;                    // Enable xmit interrupt
    HAL_WRITE_UINT8(port+SER_16550_IER, _ier);

    (chan->callbacks->xmt_char)(chan);
}


The last line loads the first character into the transmit
buffer. Reason being that the TX interrupt doesn't happen until the TX
buffer goes from full->empty.

This is also done in the quicc_smc(1?) driver so if your are using the
same sources it's probably something else.

Jesper