Skip to contentSkip to conversation

mem= handling mess.

DISCUSSION Dave Jones 2 participants 3 messages May 27, 2004

lore
Share & tools
lore

Dave Jones

author
#0 lore raw

At some point in time during 2.4, parse_cmdline_early() changed so that it handled such boot command lines as..

mem=exactmap mem=640k@0 mem=511m@1m

And all was good. This change propagated forward into 2.5, where it sat for a while, until hpa freaked out and Randy Dunlap sent in cset 1.889.364.25

ChangeSet 1.889.364.25 2003/03/16 23:22:16 [email protected]
[PATCH] Fix mem= options

Patch from "Randy.Dunlap" <[email protected]>

Reverts the recent alteration of the format of the mem=' option. This is because mem=' is interpreted by bootloaders and may not be freely changed.

Instead, the new functionality to set specific memory region usages is provided via the new "memmap=" option.

The documentation for memmap= is added, and the documentation for mem= is updated.

This is all well and good, but 2.4 never got the same treatment. Result ? Now users are upgrading their 2.4 systems to 2.6, and finding that they don't boot any more.
(See https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=124312
for example).

The "`mem=' is interpreted by bootloaders and may not be freely changed." obviously hasn't broken the however many users of this we have in 2.4 so I don't buy that it'll break in 2.6 either. As its now in 2.4 (and has been there for some time), this is something that bootloaders will just have to live with.

Dave

#1 lore raw
Quoted 29 lines · Dave Jones

At some point in time during 2.4, parse_cmdline_early() changed so that it handled such boot command lines as..

mem=exactmap mem=640k@0 mem=511m@1m

And all was good. This change propagated forward into 2.5, where it sat for a while, until hpa freaked out and Randy Dunlap sent in cset 1.889.364.25

ChangeSet 1.889.364.25 2003/03/16 23:22:16 [email protected]
[PATCH] Fix mem= options

Patch from "Randy.Dunlap" <[email protected]>

Reverts the recent alteration of the format of the mem=' option. This is because mem=' is interpreted by bootloaders and may not be freely changed.

Instead, the new functionality to set specific memory region usages is provided via the new "memmap=" option.

The documentation for memmap= is added, and the documentation for mem= is updated.

This is all well and good, but 2.4 never got the same treatment. Result ? Now users are upgrading their 2.4 systems to 2.6, and finding that they don't boot any more.
(See https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=124312
for example).

Quoted 6 lines

The "`mem=' is interpreted by bootloaders and may not be freely changed." obviously hasn't broken the however many users of this we have in 2.4 so I don't buy that it'll break in 2.6 either. As its now in 2.4 (and has been there for some time), this is something that bootloaders will just have to live with.

It was changed to memmap= I thought. The command line suggested above, for example, WILL NOT boot with any correctly operating boot loader. Unfortunately, given its history I suspect GRUB is nowhere in that category.

I think telling people to use memmap= instead of mem= is the only sane way to deal with this.

What is even more puzzling is that the command line used by the user in that question is bogus, and should be *exactly* identical to "mem=512M", but the bug report indicates that doesn't work on that machine. Thus, there is something more seriously wrong.

-hpa

#2 lore raw
Followup to:  <[email protected]>
By author:    "H. Peter Anvin" <[email protected]>
In newsgroup: linux.dev.kernel
Quoted 8 lines
Quoted 1 line · Dave Jones

mem=exactmap mem=640k@0 mem=511m@1m

It was changed to memmap= I thought. The command line suggested above, for example, WILL NOT boot with any correctly operating boot loader. Unfortunately, given its history I suspect GRUB is nowhere in that category.

I take that back. It works by pure accident: since "mem=511m@1m" comes last most bootloaders will read it as "mem=511m" which is safe given this particular memory map.

In other words, this and stuff like this work by pure accident if they work at all, which is to some degree even worse -- partially because it let people be lazy about it and thus the wrong thing was allowed to simmer, as evidenced above. The "right thing", if there is such a thing, is probably to detect entries of this form and report an error to the user ("use memmap= instead.")

The only other sane alternative is worse in terms of incompatibility:
completely throw away the old initrd protocol and design a better one. The initrd loading protocol is horrible, but it's been around for a long time and it's not likely to go away.

-hpa