linux-kernel - Re: [PATCH V2 2/7] xen/grants: support allocating consecutive grants

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <b2d22549-cdde-1e52-18fd-0ff923b86a5d@gmail.com>
Date:   Mon, 16 May 2022 21:30:21 +0300
From:   Oleksandr <olekstysh@...il.com>
To:     Boris Ostrovsky <boris.ostrovsky@...cle.com>
Cc:     Juergen Gross <jgross@...e.com>, xen-devel@...ts.xenproject.org,
        linux-kernel@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
        Stefano Stabellini <sstabellini@...nel.org>,
        Julien Grall <julien@....org>,
        Oleksandr Tyshchenko <oleksandr_tyshchenko@...m.com>,
        "Michael S. Tsirkin" <mst@...hat.com>,
        Christoph Hellwig <hch@...radead.org>
Subject: Re: [PATCH V2 2/7] xen/grants: support allocating consecutive grants


On 16.05.22 19:00, Boris Ostrovsky wrote:


Hello Boris


>
>
> On 5/16/22 1:59 AM, Juergen Gross wrote:
>> On 14.05.22 04:34, Boris Ostrovsky wrote:
>>>
>>>
>>> On 5/13/22 1:33 AM, Juergen Gross wrote:
>>>> On 12.05.22 22:01, Boris Ostrovsky wrote:
>>>>>
>>>>> On 5/7/22 2:19 PM, Oleksandr Tyshchenko wrote:
>>>
>>>>>> +/* Rebuilds the free grant list and tries to find count 
>>>>>> consecutive entries. */
>>>>>> +static int get_free_seq(unsigned int count)
>>>>>> +{
>>>>>> +    int ret = -ENOSPC;
>>>>>> +    unsigned int from, to;
>>>>>> +    grant_ref_t *last;
>>>>>> +
>>>>>> +    gnttab_free_tail_ptr = &gnttab_free_head;
>>>>>> +    last = &gnttab_free_head;
>>>>>> +
>>>>>> +    for (from = find_first_bit(gnttab_free_bitmap, gnttab_size);
>>>>>> +         from < gnttab_size;
>>>>>> +         from = find_next_bit(gnttab_free_bitmap, gnttab_size, 
>>>>>> to + 1)) {
>>>>>> +        to = find_next_zero_bit(gnttab_free_bitmap, gnttab_size,
>>>>>> +                    from + 1);
>>>>>> +        if (ret < 0 && to - from >= count) {
>>>>>> +            ret = from;
>>>>>> +            bitmap_clear(gnttab_free_bitmap, ret, count);
>>>>>> +            from += count;
>>>>>> +            gnttab_free_count -= count;
>>>>>
>>>>>
>>>>> IIUIC we can have multiple passes over this, meaning that the 
>>>>> gnttab_free_count may be decremented more than once. Is that 
>>>>> intentional?
>>>>
>>>> After the first pass decrementing gnttab_free_cnt, ret will no
>>>> longer be less than zero, so this can be hit only once.
>>>
>>> Oh, yes, of course.
>>>
>>>>
>>>>>
>>>>>
>>>>>> +            if (from == to)
>>>>>> +                continue;
>>>>>> +        }
>>>>>> +
>>>>>> +        while (from < to) {
>>>>>> +            *last = from;
>>>>>> +            last = __gnttab_entry(from);
>>>>>> +            gnttab_last_free = from;
>>>>>> +            from++;
>>>>>> +        }
>>>>>
>>>>>
>>>>> I have been looking at this loop and I can't understand what it is 
>>>>> doing ;-( Can you enlighten me?
>>>>
>>>> It is recreating the free list in order to have it properly sorted.
>>>> This is needed to make sure that the free tail has the maximum
>>>> possible size (you can take the tail off the list without having
>>>> to worry about breaking the linked list because of references into
>>>> the tail).
>>>
>>>
>>> So let's say we have the (one-dimensional) table of length 13
>>>
>>> idx    ..    2    3  ...  10  11  12
>>>
>>> grant       12   11        2  -1   3
>>>
>>>
>>> and gnttab_free_head is 10. I.e. the free list is 2, 12, 3, 11.
>>
>> You meant 10, 2, 12, 3, 11, I guess?
>>
>>>
>>> What will this look like after the 2 iterations of the outer loop?
>>
>> idx    ..    2    3  ...  10  11  12
>>
>> grant        3   10       11  12  -1
>>
>> with gnttab_free_head being 2, i.e the free list is now 2, 3, 10, 11, 
>> 12.
>
>
>
> OK, thanks, that helped. I couldn't link the free chunks in my head
>
>
> With the error handling in gnttab_init() fixed

yes, this is a diff that I am going to apply for the next version:


[snip]

@@ -1596,19 +1601,20 @@ static int gnttab_expand(unsigned int req_entries)
  int gnttab_init(void)
  {
         int i;
-       unsigned long max_nr_grant_frames;
+       unsigned long max_nr_grant_frames, max_nr_grefs;
         unsigned int max_nr_glist_frames, nr_glist_frames;
         int ret;

         gnttab_request_version();
         max_nr_grant_frames = gnttab_max_grant_frames();
+       max_nr_grefs = max_nr_grant_frames *
+ gnttab_interface->grefs_per_grant_frame;
         nr_grant_frames = 1;

         /* Determine the maximum number of frames required for the
          * grant reference free list on the current hypervisor.
          */
-       max_nr_glist_frames = (max_nr_grant_frames *
- gnttab_interface->grefs_per_grant_frame / RPP);
+       max_nr_glist_frames = max_nr_grefs / RPP;

         gnttab_list = kmalloc_array(max_nr_glist_frames,
                                     sizeof(grant_ref_t *),
@@ -1625,8 +1631,7 @@ int gnttab_init(void)
                 }
         }

-       i = gnttab_interface->grefs_per_grant_frame * max_nr_grant_frames;
-       gnttab_free_bitmap = bitmap_zalloc(i, GFP_KERNEL);
+       gnttab_free_bitmap = bitmap_zalloc(max_nr_grefs, GFP_KERNEL);
         if (!gnttab_free_bitmap) {
                 ret = -ENOMEM;
                 goto ini_nomem;


>
>
> Reviewed-by: Boris Ostrovsky <boris.ostrovsky@...cle.com>

Thanks!



>
>
>
> -boris

-- 
Regards,

Oleksandr Tyshchenko