I really find this interesting, since presumably, the execution of a small
number of critical RT tasks is a common use case in RTLinux.
Tomek, et al,
In a use case where there is, say a couple of periodic RT tasks that work
with a very large buffer (say 32 meg), how does one allocate this memory,
and ensure that it is not swapped, and does not interfere with other kernel
operations (on, say, a 64 meg machine). Do you need to do anything special?
thom
> From: Tomasz Motylewski <[EMAIL PROTECTED]>
> Date: Sat, 18 Mar 2000 17:26:52 +0100 (CET)
> To: Alain Rollé <[EMAIL PROTECTED]>
> Cc: [EMAIL PROTECTED]
> Subject: Re: [rtl] virtual memory in RTL
>
> On Sat, 18 Mar 2000, Alain [iso-8859-1] Rollé wrote:
>
>> fault. The time required to bring the referenced data into physical memory
>> from
>> disk can cause a real-time application to lose its determinism. The Solaris
>> Operating Environment addresses this problem by allowing the locking of a
>> process' pages into memory using mlock(3C) or mlockall(3C).
>
> These functions are also available for user space processes in Linux. In
> fact, they are recommended if you want to use soft real time featues of
> standard Linux.
>
>> If I understand well, this problem does not occur in RTL because virtual
>> memory
>> is not a feature which is supported by the Real Time kernel. This makes me
>> wonder, however, how the RT kernel obtains memory for the real time tasks. Is
>> it a fixed partition of memory that never gets swapped out ? How does it work
>> ?
>
> The memory in RT and non-RT Linux kernel is the same. For small allocations,
> kmalloc is used, physical memory is allocated (but is accessed through
> virtual adress, different from the physical address). For bigger allocations
> vmalloc() is used, this gets a list of free blocks, and remaps them so they
> look continuous. All kinds of memory allocated for kernel use are not
> swappable of course. But it is not a fixed partition, everything is
> dynamically resized, including disk cache etc. For tasks executing in
> hard real time, this means when the kernel proper may have interrupts
> disabled situation is more complicated - they can not allocate memory using
> standard Linux kernel calls (hard real time tasks can preempt critical
> regions of Linux kernel code). Memory used in ISR and periodic tasks should
> be preallocated.
>
>> To avoid jitter, the Solaris Operating Environment provides for early binding
>> of dynamic libraries. By setting the LD_BIND_NOW environment variable to "1",
>> libraries are bound at application startup time.
>> I presume the same technique is used in RTLinux ? If not, how is it done ?
>
> No need for it. The library consists of other (standard) RT modules you load
> before your module. It is not possible to defer the binding. This is all
> kernel space.
>
> Best regards,
> --
> Tomek
>
> -- [rtl] ---
> To unsubscribe:
> echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
> echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
> ---
> For more information on Real-Time Linux see:
> http://www.rtlinux.org/~rtlinux/
>
-- [rtl] ---
To unsubscribe:
echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
---
For more information on Real-Time Linux see:
http://www.rtlinux.org/rtlinux/
