LWN: Comments on "Execute-in-place"

Execute-in-place

tohoyn — Wed, 26 Apr 2006 11:40:39 +0000

I found all the patches with "Search".

Execute-in-place

tohoyn — Wed, 26 Apr 2006 11:13:19 +0000

Where is this filemap_xip_nopage? I can't find it in the patch.

Execute-in-place

jg — Thu, 01 Sep 2005 14:46:26 +0000

I see assertions, with no data, I think often by people with no experience in the area.

Certainly on the iPAQ we've lived fine without XIP, and flash is more precious that RAM. And the paging system does very well at throwing away unused pages of text that have been pulled into RAM.

So I think we need data here, rather than proof by loud assertion...
- Jim

Execute-in-place

dmag — Sat, 21 May 2005 20:17:20 +0000

There is also a latency issue. XIP will boot faster. It will also run faster in the case where you have more flash than RAM.

Execute-in-place

obobo — Sun, 15 May 2005 19:57:54 +0000

It is typical for microcontrollers with on-die flash and RAM to have significantly more flash than RAM. Granted that today, most of these microcontrollers don't have enough memory to run Linux with or without XIP, but bigger chips will come through at some point.

Interesting point, though.

Execute-in-place

phiggins — Fri, 13 May 2005 06:26:41 +0000

Depends on the type of RAM. XIP would save a lot of power if you use DRAM, but not so much with SRAM and the Magnetic RAM. I think those cost more than flash right now, though.

Execute-in-place

oak — Thu, 12 May 2005 19:19:45 +0000

But doesn't RAM need refresh whereas Flash doesn't?
I.e. XIP with less RAM might make your battery last longer...

Execute-in-place

karim — Thu, 12 May 2005 13:58:43 +0000

It would be important to highlight that there isn't concensus on the availability/use of XIP within the embedded Linux community. David Woodhouse who maintains the MTD layer, for example, has said a number of times that the only ones who profit from XIP are flash manufacturers. The truth of the matter is that flash costs more than RAM and having XIP means having the binaries uncompressed in flash. The more XIP you have, the larger the flash you need. From that point of view, it's clear that having the classic compressed FS in flash and the apps running in RAM is usually the best way to go. ... not that this fact is going to stop developers from asking for this feature ...

There may be an advantage for XIP in the case of mass-produced masked-ROMs, maybe, but that probably benefits to a handfull of companies on this planet only ...

Execute-in-place

cotte — Thu, 12 May 2005 11:06:14 +0000

...maybe I should read the article to the end before complaining ;)

Execute-in-place

cotte — Thu, 12 May 2005 10:33:49 +0000

>So far, the XIP patches enable fast, memory-to-memory device access,
>but they do not yet implement true execute-in-place operation.
That is not true. In fact the filemap_xip_nopage function for xip in mm/filemap.c directly maps on-disk content into userspace - that's exactly what execute in place is.