[ECOS] net: splsoftnet - problems

[ECOS] net: splsoftnet - problems

[Joerg Troeger]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 740 bytes --]

I have some concerns related to the "splsoftnet()" function of the TCP/IP stack.

1. splsoftnet may block due to cyg_mutex_lock() - but is called in the
   context of the real-time-clock DSR (tcp_fasttimo, tcp_slowtimo).

2. splsoftnet may try to lock a mutex twice!!
   Consider a timer interrupt right after calling "cyg_mutex_loc(&softnet_mutex)"
   and before returning from "cyg_scheduler_lock()" - it est between line #246 and
   #247 of support.c rev. 1.5. 
   The timer interrupt may call tcp_fasttimo, which calls splsoftnet - which tries to
   lock the same mutex once more (this has been seen happen!).

I am quiet not sure how to solve this problems and would be glad if this can be commented!

Jörg Tröger

Re: [ECOS] net: splsoftnet - problems

[Bart Veer]

>>>>> "Joerg" == Joerg Troeger <jtroeger@nortelnetworks.com> writes:

    Joerg> SSBoYXZlIHNvbWUgY29uY2VybnMgcmVsYXRlZCB0byB0aGUgInNwbHNvZnRuZ
    Joerg> b24gb2YgdGhlIFRDUC9JUCBzdGFjay4NCg0KMS4gc3Bsc29mdG5ldCBtYXkgY
    ...

Please set up your mailer to send in plain text, not in MIME. Not
everybody uses mime-aware mail readers, so such messages may well get
ignored by the people who would be able to answer. In this case I'll
make sure that the relevant engineer sees both your postings,
decrypted. 

   Joerg> I have some concerns related to the "splsoftnet()" function
   Joerg> of the TCP/IP stack.

   Joerg> 1. splsoftnet may block due to cyg_mutex_lock() - but is
   Joerg> called in the context of the real-time-clock DSR
   Joerg> (tcp_fasttimo, tcp_slowtimo).

   Joerg> 2. splsoftnet may try to lock a mutex twice!! Consider a
   Joerg> timer interrupt right after calling
   Joerg> "cyg_mutex_loc(&softnet_mutex)" and before returning from
   Joerg> "cyg_scheduler_lock()" - it est between line #246 and #247
   Joerg> of support.c rev. 1.5. The timer interrupt may call
   Joerg> tcp_fasttimo, which calls splsoftnet - which tries to lock
   Joerg> the same mutex once more (this has been seen happen!).

   Joerg> I am quiet not sure how to solve this problems and would be
   Joerg> glad if this can be commented!

Officially the TCP/IP stack is still in beta, there are still some
problems which are being resolved. I know that there were some changes
to the timeout.c and support.c files just yesterday, which may or may
not relate to the issues you have raised. The next anoncvs snapshot
should contain those and other changes.

Of course any help we get in sorting out the remaining problems will
be greatly appreciated.

There is a slight complication in that anoncvs updates have been
suspended for a little while, to restabilize the tree after some major
changes. They should start again real soon now (as in days, not
weeks).

Bart Veer // eCos net maintainer