Skip to contentSkip to conversation

Linux/Pro -- clusters

DISCUSSION [email protected] 2 participants 2 messages Dec 10, 2001

lore
Share & tools
lore
#0 lore raw

From [email protected] Mon Dec 10 17:50:02 2001

Basically you propose to take the current system, replace it with something without clear memory management ("let it leak") and then try to fix the resulting mess.

Al - you are using debating tricks instead of logic, using negative words ("unclear", "leak", "mess") instead of arguments. Maybe you are unable to refute the soundness of the system I propose?

It is quite possible that I overlook some detail. On the other hand, I have been running these systems. You are not able to convince me that something is wrong just by handwaving. Real arguments are required.

What I do is go from the present situation, in a series of steps, to a new situation where the source looks different but the system behaves provably the same. Consequently, no "fixing" is required. "Mess" is a matter of taste, I'll not discuss that except by saying that I vastly prefer the situation without arrays. "Leak" is false. "Dangling pointers" is false.

Andries

[About "leak": What happens today is that a driver like sd.c allocates arrays and fills them. In my version this driver allocates structures and fills them. When the module is removed, today the arrays are freed. In my version the structures are freed at that point. So, no leakage occurs.
About "dangling pointers": The correctness condition for this scheme is that no struct that contains kdev_t fields survives removal of the module.
It seems to me that that is true already, and in any case will be easy to ensure. If you have other opinions, please come with explicit examples where fundamental problems would occur.]

[and, Linus, the name of the beast makes no difference; kdev_t or kbdev_t or struct block_device *; it is the same amount of work]

#1 lore raw
Quoted 9 lines · On Mon, 10 Dec 2001 [email protected] wrote:

From [email protected] Mon Dec 10 17:50:02 2001

Basically you propose to take the current system, replace it with something without clear memory management ("let it leak") and then try to fix the resulting mess.

Al - you are using debating tricks instead of logic, using negative words ("unclear", "leak", "mess") instead of arguments. Maybe you are unable to refute the soundness of the system I propose?

<blink>

Quoted 11 lines

It is quite possible that I overlook some detail. On the other hand, I have been running these systems. You are not able to convince me that something is wrong just by handwaving. Real arguments are required.

What I do is go from the present situation, in a series of steps, to a new situation where the source looks different but the system behaves provably the same. Consequently, no "fixing" is required. "Mess" is a matter of taste, I'll not discuss that except by saying that I vastly prefer the situation without arrays. "Leak" is false. "Dangling pointers" is false.

What??? You've just said that on the first stage you are not going to free these objects and then add freeing them and audit the whole thing at that point.

The first is commonly known as leak (objects are allocated but not freed). Dangling pointers is what you will have to fight during that audit - places where something retains kdev_t after your object had been freed.

Let me rephrase it: with your plan we will have much more complex audit needed at the moment when you introduce freeing your objects. Reason:
it will have to involve all subsystems using kdev_t at once. That's my problem with your plan. Sigh...