Hi, On Saturday, 14 January 2006 12:29, Pavel Machek wrote: > > > > > > > It returns the number of pages. Well, it should be written explicitly, > > > > > > > so I'll fix that. > > > > > > > > > > > > Please always talk to the kernel in bytes. Pagesize is only a kernel > > > > > > internal unit. Sth. like off64_t is fine. > > > > > > > > > > These are values returned by the kernel, actually. Of course I can convert them > > > > > to bytes before sending to the user space, if that's preferrable. > > > > > > > > > > Pavel, what do you think? > > > > > > > > Bytes, I'd say. It would be nice if preffered image size was in bytes, > > > > too, for consistency. > > > > > > OK > > > > Having actually tried to do that I see two reasons for keeping the image size > > in megs. > > > > First, if that was in bytes, I'd have to pass it via a pointer, because > > unsigned long might overflow on i386. Then I'd have to use get_user() > > Actually unsigned long is okay. We can't do images > 1.5GB, anyway, > on i386. Right. In fact 1/2 of lowmem is the limit. > > to read the value. However, afterwards I'd have to rescale that value > > to megs for swsusp_shrink_memory(). It's just easier to pass the value > > in megs using the last argument of ioctl() directly (which is consistent > > with the /sys/power/image_size thing, BTW). > > Well, I'd be inclined to make image_size in bytes, too. Having > each ioctl/sys file in different units seems wrong. I'll add these changes to the userland interface patch, then. There won't be too many of them, I think. > > Second, if that's in bytes, it would suggest that the memory-shrinking > > mechanism had byte granularity (ie. way off). > > Yep, but it is not that bad. > > > There also is a reason for which SNAPSHOT_AVAIL_SWAP should return > > the number of pages, IMO. Namely, if that's in pages, the number is directly > > comparable with the number of image pages which the suspending > > utility can read from the image header. Otherwise it would have to rescale > > one of these values using PAGE_SIZE, but that's exactly what we'd like > > to avoid. > > I see.... We could put #bytes into image header (unsigned long) :-). > > Its not too bad one way or another, because uswsusp tools are > intimately tied to kernel, anyway. Exactly. :-) Greetings, Rafael
