Re: [PATCH v10 3/6] fs/proc/task_mmu: Implement IOCTL to get and/or the clear info about PTEs

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, 23 Feb 2023 at 10:23, Muhammad Usama Anjum
<usama.anjum@xxxxxxxxxxxxx> wrote:
>
> On 2/23/23 1:41 PM, Michał Mirosław wrote:
> > On Thu, 23 Feb 2023 at 07:44, Muhammad Usama Anjum
> > <usama.anjum@xxxxxxxxxxxxx> wrote:
> >>
> >> On 2/22/23 4:48 PM, Michał Mirosław wrote:
> >>> On Wed, 22 Feb 2023 at 12:06, Muhammad Usama Anjum
> >>> <usama.anjum@xxxxxxxxxxxxx> wrote:
> > [...]
> >>>>>>> BTW, I think I assumed that both conditions (all flags in
> >>>>>>> required_flags and at least one in anyof_flags is present) need to be
> >>>>>>> true for the page to be selected - is this your intention?
> >>>>>> All the masks are optional. If all or any of the 3 masks are specified, the
> >>>>>> page flags must pass these masks to get selected.
> >>>>>
> >>>>> This explanation contradicts in part the introductory paragraph, but
> >>>>> this version seems more useful as you can pass all masks zero to have
> >>>>> all pages selected.
> >>>> Sorry, I wrote it wrongly. (All the masks are not optional.) Let me
> >>>> rephrase. All or at least any 1 of the 3 masks (required, any, exclude)
> >>>> must be specified. The return_mask must always be specified. Error is
> >>>> returned if all 3 masks (required, anyof, exclude) are zero or return_mask
> >>>> is zero.
> >>>
> >>> Why do you need those restrictions? I'd guess it is valid to request a
> >>> list of all pages with zero return_mask - this will return a compact
> >>> list of used ranges of the virtual address space.
> >> At the time, we are supporting 4 flags (PAGE_IS_WRITTEN, PAGE_IS_FILE,
> >> PAGE_IS_PRESENT and PAGE_IS_SWAPPED). The idea is that user mention his
> >> flags of interest in the return_mask. If he wants only 1 flag, he'll
> >> specify it. Definitely if user wants only 1 flag, initially it doesn't make
> >> any sense to mention in the return mask. But we want uniformity. If user
> >> want, 2 or more flags in returned, return_mask becomes compulsory. So to
> >> keep things simple and generic for any number of flags of interest
> >> returned, the return_mask must be specified even if the flag of interest is
> >> only 1.
> >
> > I'm not sure why do we want uniformity in the case of 1 flag? If a
> > user specifies a single required flag, I'd expect he doesn't need to
> > look at the flags returned as those will duplicate the information
> > from mere presence of a page. A user might also require a single flag,
> > but want all of them returned. Both requests - return 1 flag and
> > return 0 flags would give meaningful output, so why force one way or
> > the other? Allowing two will also enable users to express the intent:
> > they need either just a list of pages, or they need a list with
> > per-page flags - the need would follow from the code structure or
> > other factors.
> We can add as much flexibility as much people ask by keeping code simple.
> But it is going to be dirty to add error check which detects if return_mask
> = 0 and if there is only 1 flag of interest mentioned by the user. The
> following mentioned error check is essential to return deterministic
> output. Do you think this case is worth it to support and we don't want to
> go with the generality for both 1 or more flag cases?
>
> if (return_mask == 0 && hweight_long(required_mask | any_mask) != 1)
>         return error;

Why would you want to add this error check? If a user requires
multiple flags but cares only about a list of matching pages, then it
would be natural to express this intent as return_mask = 0.

> >>>>>> After taking a while to understand this and compare with already present
> >>>>>> flag system, `negated flags` is comparatively difficult to understand while
> >>>>>> already present flags seem easier.
> >>>>>
> >>>>> Maybe replacing negated_flags in the API with matched_values =
> >>>>> ~negated_flags would make this better?
> >>>>>
> >>>>> We compare having to understand XOR vs having to understand ordering
> >>>>> of required_flags and excluded_flags.
> >>>> There is no ordering in current masks scheme. No mask is preferable. For a
> >>>> page to get selected, all the definitions of the masks must be fulfilled.
> >>>> You have come up with good example that what if required_mask =
> >>>> exclude_mask. In this case, no page will fulfill the criterion and hence no
> >>>> page would be selected. It is user's fault that he isn't understanding the
> >>>> definitions of these masks correctly.
> >>>>
> >>>> Now thinking about it, I can add a error check which would return error if
> >>>> a bit in required and excluded masks matches. Would you like it? Lets put
> >>>> this check in place.
> >>>> (Previously I'd left it for user's wisdom not to do this. If he'll specify
> >>>> same masks in them, he'll get no addresses out of the syscall.)
> >>>
> >>> This error case is (one of) the problems I propose avoiding. You also
> >>> need much more text to describe the requred/excluded flags
> >>> interactions and edge cases than saying that a flag must have a value
> >>> equal to corresponding bit in ~negated_flags to be matched by
> >>> requried/anyof masks.
> >> I've found excluded_mask very intuitive as compared to negated_mask which
> >> is so difficult to understand that I don't know how to use it correctly.
> >> Lets take an example, I want pages which are PAGE_IS_WRITTEN and are not
> >> PAGE_IS_FILE. In addition, the pages must be PAGE_IS_PRESENT or
> >> PAGE_IS_SWAPPED. This can be specified as:
> >>
> >> required_mask = PAGE_IS_WRITTEN
> >> excluded_mask = PAGE_IS_FILE
> >> anyof_mask = PAGE_IS_PRESETNT | PAGE_IS_SWAP
> >>
> >> (a) assume page_flags = 0b1111
> >> skip page as 0b1111 & 0b0010 = true
> >>
> >> (b) assume page_flags = 0b1001
> >> select page as 0b1001 & 0b0010 = false
> >>
> >> It seemed intuitive. Right? How would you achieve same thing with negated_mask?
> >>
> >> required_mask = PAGE_IS_WRITTEN
> >> negated_mask = PAGE_IS_FILE
> >> anyof_mask = PAGE_IS_PRESETNT | PAGE_IS_SWAP
> >>
> >> (1) assume page_flags = 0b1111
> >> tested_flags = 0b1111 ^ 0b0010 = 0b1101
> >>
> >> (2) assume page_flags = 0b1001
> >> tested_flags = 0b1001 ^ 0b0010 = 0b1011
> >>
> >> In (1), we wanted to skip pages which have PAGE_IS_FILE set. But
> >> negated_mask has just masked it and page is still getting tested if it
> >> should be selected and it would get selected. It is wrong.
> >>
> >> In (2), the PAGE_IS_FILE bit of page_flags was 0 and got updated to 1 or
> >> PAGE_IS_FILE in tested_flags.
> >
> > I require flags PAGE_IS_WRITTEN=1, PAGE_IS_FILE=0, so:
> >
> > required_mask = PAGE_IS_WRITTEN | PAGE_IS_FILE;
> > negated_flags = PAGE_IS_FILE; // flags I want zero
> You want PAGE_IS_FILE to be zero and at the same time you are requiring the
> PAGE_IS_FILE. It is confusing.

Ok, I believe the misunderstanding comes from the naming. I "require"
the flag to be a particular value - hence include it in
"required_flags" and specify the required value in ~negated_flags. You
"require" the flag to be set (equal 1) and so include it in
"required_flags" and you "require" the flag to be clear (equal to 0)
so include it in "excluded_flags". Both approaches are correct, but I
would not consider one "easier" than the other. The former is more
general, though - makes any_of also able to match on flags cleared and
removes the possibility of a conflicting case of a flag present in
both sets.

Maybe considered_flags or matched_flags then would make the field
better understandable?

Best Regards
Michał Mirosław





[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux