Some Eclipse Foundation services are deprecated, or will be soon. Please ensure you've read this important communication.

Bug 331799

Summary:	Memory map assembly needs to comprehend virtual addresses
Product:	[Technology] RTSC	Reporter:	Ramsey Harris <ramsey>
Component:	Platforms	Assignee:	Vikram Adiga <vikram.adiga>
Status:	ASSIGNED ---	QA Contact:
Severity:	normal
Priority:	P3	CC:	akapania, cring, d-russo, dfriedland
Version:	unspecified
Target Milestone:	---
Hardware:	PC
OS:	Windows XP
Whiteboard:

Description Ramsey Harris

2010-12-03 13:21:58 EST

When I define a custom memory map in my platform instance, the configuration phase will validate that my custom memory map fits in the physical memory defined by the catalog and platform. However, when using an MMU, I want to define a virtual memory map. The validation phase would fail because there is no physical memory at the virtual addresses. There needs to be some way to define a virtual memory map and associate the physical addresses as mapped by the MMU. SysLink does this in their configuration by defining both a virtual and physical address for every memory entry. We need something like this in RTSC.

On ti.catalog.t16.Arctic there is no physical memory at address 0x0, but the vector table must be placed at that address. The idea is that the vector table will be placed in external memory and then mapped to address 0x0 using the MMU.

Comment 1 Sasha Slijepcevic

2011-11-08 17:19:55 EST

There is an option to use 'xdc.platform.custom.check' to deal with this. See http://www.sanb.design.ti.com/iliad/webhome/xdoc/xdc/platform/package.html.

Comment 2 David Friedland

2011-12-01 00:57:52 EST

From: Ring, Chris 
Sent: Tuesday, November 22, 2011 8:46 AM
To: Harris, Ramsey; Friedland, David; Slijepcevic, Sasa
Subject: RE: Bug 331799 - Memory map assembly needs to comprehend virtual addresses 

Not sure how to implement virtual memory map 'checking' any differently than disabling the memory map checking.

I think we can create our own 'platform' with the virtual memory map we want, then create instances of that new platform and leave the memory map checking enabled.  But if we want a platform _instance_ to both specify the virtual memory map and validate against its own definition that it's right... I dunno.

Maybe I don't see the vision.

Chris

From: Harris, Ramsey 
Sent: Tuesday, November 22, 2011 8:28 AM
To: Friedland, David; Slijepcevic, Sasa
Cc: Ring, Chris
Subject: RE: Bug 331799 - Memory map assembly needs to comprehend virtual addresses 

David,

No, in my opinion, the work around is just that and not a fix/solution to the problem. Please plan to address and fix this bug.

~ Ramsey

From: Friedland, David 
Sent: Monday, November 21, 2011 4:45 PM
To: Harris, Ramsey; Slijepcevic, Sasa
Cc: Ring, Chris
Subject: Bug 331799 - Memory map assembly needs to comprehend virtual addresses 

Ramsey,

Bug 331799 - Memory map assembly needs to comprehend virtual addresses 

Given the work-around listed by Sasha in the comments, can he go ahead and mark this bug as Resolved/Won’t Fix?

Comment 3 David Friedland

2012-07-03 14:28:40 EDT

This bug has been languishing in the New state.  I suggest that, if we want to get this resolved, we will need a real meeting that includes Ramsey, Chris, Sasha and Dave to decide what to do.

Comment 4 Ramsey Harris

2012-12-19 13:05:37 EST

Memory map checking is very important. I continue to benefit from this each time I make a mistake in the memory map. I don't want to lose this just because I'm using a virtual memory map, nor do I want to make a new platform each time I invent a new virtual map.

I would like to see an extension to the memory map which allows me to specify the virtual to physical mapping. Then my virtual map can be checked to make sure I don't have any overlap in either my virtual address space or my mapping to physical address space.