In close cooperation with Stefan, I refactored and rewrote the MIDAS event
buffering code (bm_send_event, bm_flush_cache, bm_receive_event and bm_push_event).

The main goal of this update is to make sure the event buffering code does not
have any infinite loops: in the past, we have seen mlogger and some frontends
loop forever consuming 100% CPU in the event buffering code. This should now be
completely fixed.

As additional bonuses, the refactored code is easier to read, has less code
duplication and should be more robust. A few potential logical problems have
been corrected and one case of reproducible infinite looping has been fixed.

The new code has passed the low-level consumer-producer tests, but has not yet
been used in anger in any real experiment. One hopes any new bugs introduced
would cause outright failures and core dumps (rather than silent data corruption).

All are welcome to try the new code. If it explodes, please send me the error
messages, stack traces and core dumps.

K.O.

> In close cooperation with Stefan, I refactored and rewrote the MIDAS event
> buffering code (bm_send_event, bm_flush_cache, bm_receive_event and bm_push_event).
>
> All are welcome to try the new code. If it explodes, please send me the error
> messages, stack traces and core dumps.

Stefan quickly found one new error (a typoe in a check against infinite looping) and
then I found one old error present in the old code that caused event loss when the
buffer became exactly 100% full (0 bytes free).

Both errors are now fixed in svn commit 3294.

K.O.

Some time ago, I accidentally introduced a bug in mfe.c- if there is data
congestion in the system, mfe.c can exit with the error "bm_flush_cache(ASYNC)
error 209" because it did not expect the valid return value BM_ASYNC_RETURN
(209) from bm_flush_cache(ASYNC). This error has now been fixed. K.O.

• Remove following line:
  

  #define DRIVER(_i) ...

  
  
• Find all lines containing
  

  DRIVER(i)(CMD_xxx, info->dd_info[i], ...)

  
  and replace them with
  

  device_driver(info->driver[i], CMD_xxx, ...)

  
  note that info->dd_info[i] is not passed any more. Instead, you pass info->driver[i]. Pleae note that the arguments passed after CMD_xxx are not checked by the compiler, since they are a variable argument list. Any error there will not produce a compiler warning, but will just crash the front-end.
  
  
• Find the line with
  

  status = pequipment->driver[i].dd(CMD_INIT, hKey, &pequipment->driver[i].dd_info,
                                          pequipment->driver[i].channels,
                                          pequipment->driver[i].flags,
                                          pequipment->driver[i].bd);

  
  and replace it with
  

  status = device_driver(&pequipment->driver[i], CMD_INIT, hKey);

  
  
• Find the line with
  

  pequipment->driver[i].dd(CMD_EXIT, pequipment->driver[i].dd_info);

  
  and replace it with
  

  device_driver(&pequipment->driver[i], CMD_EXIT);

  
  
• Find following lines
  

  hv_info->driver[i] = pequipment->driver[index].dd;
  hv_info->dd_info[i] = pequipment->driver[index].dd_info;
  hv_info->channel_offset[i] = offset;
  hv_info->flags[i] = pequipment->driver[index].flags;

  
  and replace them with
  

  hv_info->driver[i] = &pequipment->driver[index];
  hv_info->channel_offset[i] = offset;

INT cd_xxx(INT cmd, EQUIPMENT * pequipment)
{
   INT i, status;

   switch (cmd) {
   case CMD_INIT:
      ...

   case CMD_STOP:
      for (i = 0; pequipment->driver[i].dd != NULL &&
                  pequipment->driver[i].flags & DF_MULTITHREAD ; i++)
         status = device_driver(&pequipment->driver[i], CMD_STOP);
      break;

   case CMD_IDLE:
      ...

   return status;
}

DEVICE_DRIVER multi_driver[] = {
   {"Input", nulldev, 2, null, DF_INPUT | DF_MULTITHREAD},
   {"Output", nulldev, 2, null, DF_OUTPUT | DF_MULTITHREAD},
   {""}
};

By default, MIDAS creates shared memory event data buffers of default size
EVENT_BUFFER_SIZE defined in midas.h and until now making of large data buffers
for high data rate or large event size experiments was complicated.

Now, bm_open_buffer() will try to read the event buffer size from ODB. If
"/Experiment/Buffer Sizes/BUFFER_NAME" of type DWORD exists, it's value is used
as the buffer size, overriding the default value.

For example, to increase the size of the default MIDAS event buffer ("SYSTEM")
to 2000000 bytes, shutdown all MIDAS programs, delete the old .SYSTEM.SHM file
(and the shared memory segment, using ipcrm). Then run odbedit, cd /Experiment,
mkdir "Buffer Sizes", cd "Buffer Sizes", create DWORD SYSTEM, set SYSTEM
2000000. Then start the rest of the MIDAS programs. Check that the buffer has
the correct size by looking at the size of .SYSTEM.SHM and of the shared memory
segment (ipcs).

This method work for all MIDAS buffers, except for ODB, where the size has to be
specified at creation time using the odbedit command "-s" argument.

K.O.

We have been seeing consistent user memory corruption while setting up a new
experiment. This has been traced to a user memory overwrite in ODB db_set_data()
function and this problem is now fixed. This error was triggered by our frontend
code constantly changing the size of a MIDAS data bank that was also written
into ODB via the RO_ODB option. K.O.

While investigating the mhttpd password error with the MacOS Safari browser, I
found that it was caused by an strcpy() buffer overflow. With Stefan's blessing,
I now converted most uses of strcpy() and strcat() to strlcpy() and strlcat().

This fixes the Safari password problem (it was memory corruption in mhttpd).

While validating these changes, I also found an incorrect use of sizeof() in the
mhttpd history code for plotting run markers. I fixed that as well.

P.S. The remaining strcpy() calls look safe wrt buffer overflows. There are no
strcat() calls left. But there is still a large number of unsafe-looking
sprintf() uses.

K.O.

I observe that the mhttpd passwords do not work correctly for the Safari web browser on MacOS 10.4.7: 
Safari 2.0.4 (419.3). For example, I cannot submit elog messages- the system gets stuck on the 
"Password" page. The Safari browser in MacOS 10.3 works fine. Mozilla/Firefox works fine. (Also would be 
useful if "remember password" worked with MIDAS, in any browser). K.O.

I fixed the compilation of musbstd.h (it required -DHAVE_LIBUSB on Linux, but
nothing knew about defining it) and put musbstd.o back into libmidas (USB
support should be part of the standard base midas library). K.O.

> I fixed the compilation of musbstd.h (it required -DHAVE_LIBUSB on Linux, but
> nothing knew about defining it) and put musbstd.o back into libmidas (USB
> support should be part of the standard base midas library). K.O.

I'm not so sure about that. One could consider musbstd.o as a driver, and the
philosophy used for midas programs is that drivers get added explicitly when
compiling a frontend. We do not put mvmestd.c and mcstd.c into libmidas since for
different interfaces a different driver might be required. If we at some point use
an usb library different than libusb.a, we would have to compile different
libmidas for these different drivers.

I know it's convenient to have things in libmidas and not having to specify it
expliceitely for each frontend, but it is then somehow inconsistent with drivers
for vme and camac. So please reconsider this again.

- Stefan

Fresh svn checkout of MIDAS does not build on Fedora 5, I get this error:

cc -c -g -O2 -Wall -Wuninitialized -Iinclude -Idrivers -I../mxml -Llinux/lib
-DINCLUDE_FTPLIB   -D_LARGEFILE64_SOURCE -DHAVE_ROOT -pthread
-I/triumfcs/trshare/olchansk/root/root_v5.10.00_SL40/include -m32 -DOS_LINUX
-fPIC -Wno-unused-function -o linux/lib/odb.o src/odb.c
src/odb.c: In function 'db_open_database':
src/odb.c:805: warning: dereferencing type-punned pointer will break
strict-aliasing rules
src/odb.c: In function 'db_lock_database':
src/odb.c:1350: warning: dereferencing type-punned pointer will break
strict-aliasing rules
cc: Internal error: Segmentation fault (program cc1)
Please submit a full bug report.
See <URL:http://bugzilla.redhat.com/bugzilla> for instructions.
make: *** [linux/lib/odb.o] Error 1

If I compile odb.c without "-O2", the rest of MIDAS builds without any more errors.

The observed warnings are (I do not know what they mean):
warning: dereferencing type-punned pointer will break strict-aliasing rules
warning: missing sentinel in function call (Cannot do without sentinels, eh?)
warning: pointer targets in passing argument 3 of 'getsockname' differ in signedness
warning: non-local variable '<anonymous struct> out_info' uses anonymous type

The "invalid lvalue" errors seem to have been successfully vanquished.

K.O.

I commited the latest VMIC VME driver we use at TRIUMF. It has working support
for D32 and D64 DMA and can move data from the SIS3820 multiscaler through the
MIDAS frontend at > 30 Mbytes/sec on our VMICVME-7805 machines. The actual DMA
speed on the VME bus is around 50 Mbytes/sec, effective data rate is lower
because of a memcpy() from the kernel DMA buffer into user memory (required by
the MIDAS mvmestd.h interface, quite inefficient for DMA operations). K.O.

When using mhist interactively, I could not look at array data:
1) if the array is the only variable, the question "what array index to use?"
was not asked, zero was assumed,
2) even if the question was asked, the answer was ignored, zero was used.
Fixes commited to utils/mhist.c
K.O.

It appears that the latest version of MIDAS malfunctions on PowerPC/VxWorks
machines, below are two problem reports. As reported, previous versions of MIDAS
work fine, I guess that reduces the probability of it being buggy user code. At
least one of the problems feels like a missing endian conversion somewhere, but
I am not aware of any recent changes in the MIDAS RPC code... We will be trying
to debug both problems, but any insight would be greatly appreciated.

K.O.


From suz@triumf.ca  Tue May 30 16:58:16 2006
Date: Tue, 30 May 2006 16:58:16 -0700 (PDT)
From: Suzannah Daviel <suz@triumf.ca>
To: konstantin olchanski <olchansk@triumf.ca>
Subject: rpc problems

Hi Konstantin,

Herewith a description of the problems,

Suzannah

Problem on system A:
--------------------

After upgrading the Linux operating system from RH9 to SL4, and installing
latest Midas software, the first time a manual trigger is issued, the VxWorks
frontend (running
on a PPC) crashes:


Output on PPC consol:

trigger histo event from status page

rpc_client_accept: starting with sock:11

program
Exception current instruction address: 0x01ac7388
Machine Status Register: 0x0008b030
Condition Register: 0x24000082
Task: 0x1b47908 "mfe"



The histo event is usually large so is fragmented. It is sent out by a
manual trigger and at end of run. When the run is ended (before an event
request using a manual trigger so program has not yet crashed) the histo
event is sent successfully.

After returning to the previous version of Midas but still running SL4,
this problem disappeared.




Problem on system B:
--------------------

Again, SL9 was installed, and the Midas software updated to the latest.
When sending a periodic (non-fragmented) event, after a while, one of the
parameters appears to become corrupted, and a lot of rpc_call error
messages appear. These continue while data is still successfully sent out
until the run is ended.



Tue May  9 05:20:29 2006 [Mdarc] *** data saved in file
/is01_data/bnmr/dlog/2006/040377.msr_v5 at Tue May  9 05:20:29
2006 (SN=5) ***
Tue May  9 05:21:30 2006 [Mdarc] *** data saved in file
/is01_data/bnmr/dlog/2006/040377.msr_v6 at Tue May  9 05:21:30
2006 (SN=6) ***
Tue May  9 05:22:31 2006 [Mdarc] *** data saved in file
/is01_data/bnmr/dlog/2006/040377.msr_v7 at Tue May  9 05:22:31
2006 (SN=7) ***

Tue May  9 05:23:12 2006 [feBNMR] [midas.c:9325:rpc_call] parameters
(1099059848) too large for network buffer
(524344); param_size=1099059808
Tue May  9 05:23:12 2006 [feBNMR] [midas.c:9325:rpc_call] parameters
(1099059848) too large for network buffer
(524344); param_size=1099059808
........................................
Tue May  9 05:23:31 2006 [feBNMR] [midas.c:9325:rpc_call] parameters
(1099059848) too large for network buffer
(524344); param_size=1099059808
Tue May  9 05:23:32 2006 [feBNMR] [midas.c:9325:rpc_call] parameters
(1099059848) too large for network buffer
(524344); param_size=1099059808

Tue May  9 05:23:32 2006 [Mdarc] *** data saved in file
/is01_data/bnmr/dlog/2006/040377.msr_v8 at Tue May  9 05:23:32
2006 (SN=8) ***

Tue May  9 05:23:32 2006 [feBNMR] [midas.c:9325:rpc_call] parameters
(1099059848) too large for network buffer
(524344); param_size=1099059808
Tue May  9 05:23:33 2006 [feBNMR] [midas.c:9325:rpc_call] parameters
(1099059848) too large for network buffer
(524344); param_size=1099059808
etc.

Another example showing that the corrupted parameter varies in size:

Thu Apr 13 19:00:00 2006 [mhttpd] Run #30005 started
Thu Apr 13 19:00:08 2006 [Mdarc] *** Saved data file
/is01_data/bnmr/dlog/2006/030005.msr_v1 at Thu Apr 13 19:00:08 2006 ***
Thu Apr 13 19:01:10 2006 [Mdarc] *** Saved data file
/is01_data/bnmr/dlog/2006/030005.msr_v2 at Thu Apr 13 19:01:10 2006 ***
Thu Apr 13 19:02:14 2006 [Mdarc] *** Saved data file
/is01_data/bnmr/dlog/2006/030005.msr_v3 at Thu Apr 13 19:02:14 2006 ***
Thu Apr 13 19:03:20 2006 [Mdarc] *** Saved data file
/is01_data/bnmr/dlog/2006/030005.msr_v4 at Thu Apr 13 19:03:20 2006 ***
Thu Apr 13 19:04:22 2006 [Mdarc] *** Saved data file
/is01_data/bnmr/dlog/2006/030005.msr_v5 at Thu Apr 13 19:04:22 2006 ***
Thu Apr 13 19:05:12 2006 [feBNMR] [midas.c:9323:rpc_call] parameters
(1077739560) too large for network buffer
(524344)
Thu Apr 13 19:05:13 2006 [feBNMR] [midas.c:9323:rpc_call] parameters
(1077739560) too large for network buffer
(524344)
etc.