Re: [PATCH] zram: rework reset and destroy path

[Sergey Senozhatsky]

Hello Minchan,

On (02/04/15 08:42), Minchan Kim wrote:
> > +	kfree(zram_devices);
> > +	unregister_blkdev(zram_major, "zram");
> > +	pr_debug("Destroyed %u device(s)\n", nr);
> 
> Create_device just shows the number of created device so I think
> no worth to emit per-device information in destroy_devices.
> Let's just emit clean up done like old in zram_exit but
> use pr_info instead of pr_debug.

not critical let's keep it as is (it just mirrors the message from init()),
and I wouldn't say it's totally useless now. we allocate space for devices,
disk, queue, etc. and we destroy it here.

please see below.

> Another concern is I'd like to keep per-device interface(e,g.
> create_device, destroy_device) because there was requirement
> to add new zram device dynamically. I guess you could remember
> that. Although I didn't have a enough time to response,
> Alex finally convinced me so I hope a contributor who have time
> will do it if he has an interest about that.

yes, I was going to tell you that perhaps I'll do that. I had several
discussions on google+ and it seems that people want to see this
feature. so I was going to ask your opinion.

the solution I'm thinking about now is to replace zram devices fixed
size array with a list, protected by mutex or spin lock (doesn't
matter at this point). this will change destroy_devices() from array
iteration to destroy each list entry.

so:
a)	pr_debug("Destroyed %u device(s)\n", nr)
it will show the actual number of active devices by that time.

b) I'll refactor destroy_devices().

this rework will not make it into the upcoming merge window, so we'll
have enough time.

I haven't decided yet, if I wan't to keep released zram devices in the
list (idle zram devices list, picking up the first available device
when user requests new zram device) or I will destroy abondoned devices
and , thus, the list will represent only active devices.


I tend to select the latter one -- destroy unused zram devices. I don't
want to give additional sysfs knob to limit the number of idle devices,
waste memory for idle devices, etc. We already  have a number of quite
complicated knobs.


	-ss

> For it, per-device creating/destroy interface looks better.
> https://lkml.org/lkml/2014/8/8/142
> Anyway, I cannot expect it happens sooner so I'm not strong
> against your patch(ie, create_device, destroy_devices)
> because I think we could do refactoring it when we need it.
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/