topic PSECT on a page boundary in Operating System - OpenVMS

PSECT on a page boundary

Kris Clippeleyr — Tue, 18 May 2004 14:29:14 GMT

Hi All,

Currently I'm struggling with the following problem:

I have a little routine that executes in kernel mode at elevated IPL. The routine was originally written in C (generated a psect of 800 or so bytes). In order to prevent pagefaults I have to lock the routine in the working set. So far, so good.
Unfortunately, my "locking" routine only locks one page (8K).
As it happened the other day, the psect started off in one page and ended in the next, leading to the dreaded PGFIPLHI crash.
At first, I searched the C docs, but could not find how to tell the C compiler to align a code psect on a specific boundary (lots of stuff on data though).
Does this "feature" exist? And how is it activated?
Failing to find a solution in C, I rewrote the routine in MACRO-32, and gave the code psect the alignment attribute PAGE. A bit to my surprise, the psect was aligned on a 512-byte boundary. If you think about it, only natural for a VAX (page=512 bytes). But I thought (foolish maybe), on an Alpha PAGE=8K.
Any ideas on how I can put my psects on a boundary larger than 512-bytes?
And euh, rewriting the "locking" routine would break too much of the already existing code.
Furthermore, the code has to run on the 3 architectures.

Any help greatly appreciated.

Greetz,

Kris

Re: PSECT on a page boundary

Hein van den Heuvel — Tue, 18 May 2004 15:01:12 GMT

You mention 3 architectures.
On VAX you need to lock down code & data.
On Alpha code, data and linages... still doable.
On IA-64 (ELF Image) you would need to worry about code, data, short data, Linker generated code (jumps), ... too tricky.
So for iVMS you want to look at using lib$lock_image(va) / lib$unlock_image(va).
This locks the whole image so you want to look at placing high IPL code into a separate shareable image.
This service also exists on Alpha (recent versions).

[I learned all this during the OpenVMS bootkamp this morning!]

In macro PAGE defaults to 512, but... why not force the issue and specuify a number? (13 for 8KB).
http://h71000.www7.hp.com/doc/73final/4515/4515pro_010.html#index_x_401

The Linker manual suggest that specifying 'page' would give you 64KB on Alpha. Maybe you want to all a PSECT line to your LINKER OPTION file to influence matters?

http://h71000.www7.hp.com/doc/73final/4548/4548pro_005.html#psect_att_sec

PAGE = "Alignment on the default target page size, which is 64 KB for Alpha linking and 512 bytes for VAX linking"

Maybe you code psect got collected with other code psects in a way not intended? You makde and studied a link map did you? Maybe attach one to a future reply if you get desperate?

hth,
Hein.

Re: PSECT on a page boundary

Kris Clippeleyr — Wed, 19 May 2004 05:35:22 GMT

Hein,

Thanx for the pointers.
On VAX there was actually no problem; the code there is only 194 bytes in size (fits easily in a 512-byte page).
The code on Alpha is larger (588 bytes); so that presented a problem if aligned on a 512-byte boundary. But, your tip on specifying a number (13) instead of a reserved word (PAGE) for the psect alignment did the trick on Alpha.
I think I was a bit tired last night when reading thru the manuals (in the VAX Macro manual is mentioned a number between 1 and 9). Didn't think a larger number would work, but it does.

Btw, for Itanium I see no problem since I've read in the release notes that, and I quote
"On OpenVMS I64 systems, $LKWSET has been modified to avoid the problems encountered on OpenVMS Alpha systems by locking the entire image referenced by the parameters."

Thanx again, and problem solved.

CU,

Kris

Re: PSECT on a page boundary

Kris Clippeleyr — Fri, 07 Jan 2005 09:49:14 GMT

See remarks above.
Kris (aka Qkcl)