lkml.org 
[lkml]   [2009]   [May]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 3/6] ksm: change the KSM_REMOVE_MEMORY_REGION ioctl.
Hugh Dickins wrote:
> On Sun, 3 May 2009, Izik Eidus wrote:
>
>
>> This patch change the KSM_REMOVE_MEMORY_REGION ioctl to be specific per
>> memory region (instead of flushing all the registred memory regions inside
>> the file descriptor like it happen now)
>>
>> The previoes api was:
>> user register memory regions using KSM_REGISTER_MEMORY_REGION inside the fd,
>> and then when he wanted to remove just one memory region, he had to remove them
>> all using KSM_REMOVE_MEMORY_REGION.
>>
>> This patch change this beahivor by chaning the KSM_REMOVE_MEMORY_REGION
>> ioctl to recive another paramter that it is the begining of the virtual
>> address that is wanted to be removed.
>>
>> (user can still remove all the memory regions all at once, by just closing
>> the file descriptor)
>>
>> Signed-off-by: Izik Eidus <ieidus@redhat.com>
>>
>
> I realize that it's ridiculous to break my silence

Hello ! :)

> with a comment
> on this particular patch, when I've not yet commented on KSM as a
> whole. (In the last few days I have at last managed to set aside
> some time to give KSM the attention it deserves, but I'm still
> not yet through and ready to comment.)
>
> However, although this patch is on the right lines (certainly you
> should be allowing to remove individual regions rather than just
> all at once), I believe the patch is seriously broken and corrupting
> as is, so thought I'd better speak up now.
>
> remove_mm_from_hash_and_tree(slot->mm) is still doing its own
> silly loop through the slots:
> list_for_each_entry(slot, &slots, link)
> if (slot->mm == mm)
> break;
> So it will be operating on whatever it finds first

I just started to write big answer that go over the code path to show
why you are wrong, and then found the problem.

Thanks i will fix it and resend...


> , in general
> the wrong slot, and I expect havoc to follow once you kfree(slot).
>
> Easily fixed: replace remove_mm_from_hash_and_tree(mm)
> by remove_slot_from_hash_and_tree(slot).
>

Yea, remove_mm_from_hash_and_tree(mm) is surely something that we dont
need that was left from the old code base.

Thanks.
> Hugh
>
>
>> ---
>> mm/ksm.c | 31 +++++++++++++++++++++----------
>> 1 files changed, 21 insertions(+), 10 deletions(-)
>>
>> diff --git a/mm/ksm.c b/mm/ksm.c
>> index 982dfff..c14019f 100644
>> --- a/mm/ksm.c
>> +++ b/mm/ksm.c
>> @@ -561,17 +561,20 @@ static void remove_mm_from_hash_and_tree(struct mm_struct *mm)
>> list_del(&slot->link);
>> }
>>
>> -static int ksm_sma_ioctl_remove_memory_region(struct ksm_sma *ksm_sma)
>> +static int ksm_sma_ioctl_remove_memory_region(struct ksm_sma *ksm_sma,
>> + unsigned long addr)
>> {
>> struct ksm_mem_slot *slot, *node;
>>
>> down_write(&slots_lock);
>> list_for_each_entry_safe(slot, node, &ksm_sma->sma_slots, sma_link) {
>> - remove_mm_from_hash_and_tree(slot->mm);
>> - mmput(slot->mm);
>> - list_del(&slot->sma_link);
>> - kfree(slot);
>> - ksm_sma->nregions--;
>> + if (addr == slot->addr) {
>> + remove_mm_from_hash_and_tree(slot->mm);
>> + mmput(slot->mm);
>> + list_del(&slot->sma_link);
>> + kfree(slot);
>> + ksm_sma->nregions--;
>> + }
>> }
>> up_write(&slots_lock);
>> return 0;
>> @@ -579,12 +582,20 @@ static int ksm_sma_ioctl_remove_memory_region(struct ksm_sma *ksm_sma)
>>
>> static int ksm_sma_release(struct inode *inode, struct file *filp)
>> {
>> + struct ksm_mem_slot *slot, *node;
>> struct ksm_sma *ksm_sma = filp->private_data;
>> - int r;
>>
>> - r = ksm_sma_ioctl_remove_memory_region(ksm_sma);
>> + down_write(&slots_lock);
>> + list_for_each_entry_safe(slot, node, &ksm_sma->sma_slots, sma_link) {
>> + remove_mm_from_hash_and_tree(slot->mm);
>> + mmput(slot->mm);
>> + list_del(&slot->sma_link);
>> + kfree(slot);
>> + }
>> + up_write(&slots_lock);
>> +
>> kfree(ksm_sma);
>> - return r;
>> + return 0;
>> }
>>
>> static long ksm_sma_ioctl(struct file *filp,
>> @@ -607,7 +618,7 @@ static long ksm_sma_ioctl(struct file *filp,
>> break;
>> }
>> case KSM_REMOVE_MEMORY_REGION:
>> - r = ksm_sma_ioctl_remove_memory_region(sma);
>> + r = ksm_sma_ioctl_remove_memory_region(sma, arg);
>> break;
>> }
>>
>> --
>> 1.5.6.5
>>



\
 
 \ /
  Last update: 2009-05-04 22:51    [W:1.201 / U:1.908 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site