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

Re: [Xen-devel] [PATCH v8] x86/mem-sharing: mem-sharing a range of memory



On 27/07/16 10:01, George Dunlap wrote:
> On 26/07/16 23:43, Andrew Cooper wrote:
>> On 26/07/2016 16:49, George Dunlap wrote:
>>> On Tue, Jul 26, 2016 at 4:22 PM, Tamas K Lengyel
>>> <tamas.lengyel@xxxxxxxxxxxx> wrote:
>>>> On Tue, Jul 26, 2016 at 3:12 AM, George Dunlap <george.dunlap@xxxxxxxxxx> 
>>>> wrote:
>>>>> On Wed, Jul 20, 2016 at 7:01 PM, Tamas K Lengyel
>>>>> <tamas.lengyel@xxxxxxxxxxxx> wrote:
>>>>>> Currently mem-sharing can be performed on a page-by-page basis from the 
>>>>>> control
>>>>>> domain. However, this process is quite wasteful when a range of pages 
>>>>>> have to
>>>>>> be deduplicated.
>>>>>>
>>>>>> This patch introduces a new mem_sharing memop for range sharing where
>>>>>> the user doesn't have to separately nominate each page in both the 
>>>>>> source and
>>>>>> destination domain, and the looping over all pages happen in the 
>>>>>> hypervisor.
>>>>>> This significantly reduces the overhead of sharing a range of memory.
>>>>>>
>>>>>> Signed-off-by: Tamas K Lengyel <tamas.lengyel@xxxxxxxxxxxx>
>>>>>> Acked-by: Wei Liu <wei.liu2@xxxxxxxxxx>
>>>>>> Reviewed-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
>>>>>> ---
>>>>>> Cc: Ian Jackson <ian.jackson@xxxxxxxxxxxxx>
>>>>>> Cc: George Dunlap <george.dunlap@xxxxxxxxxxxxx>
>>>>>> Cc: Jan Beulich <jbeulich@xxxxxxxx>
>>>>>> Cc: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
>>>>>>
>>>>>> v8: style fixes and minor adjustments
>>>>>> ---
>>>>>>  tools/libxc/include/xenctrl.h        |  15 ++++
>>>>>>  tools/libxc/xc_memshr.c              |  19 +++++
>>>>>>  tools/tests/mem-sharing/memshrtool.c |  22 ++++++
>>>>>>  xen/arch/x86/mm/mem_sharing.c        | 140 
>>>>>> +++++++++++++++++++++++++++++++++++
>>>>>>  xen/include/public/memory.h          |  10 ++-
>>>>>>  5 files changed, 205 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/tools/libxc/include/xenctrl.h 
>>>>>> b/tools/libxc/include/xenctrl.h
>>>>>> index e904bd5..3782eff 100644
>>>>>> --- a/tools/libxc/include/xenctrl.h
>>>>>> +++ b/tools/libxc/include/xenctrl.h
>>>>>> @@ -2334,6 +2334,21 @@ int xc_memshr_add_to_physmap(xc_interface *xch,
>>>>>>                      domid_t client_domain,
>>>>>>                      unsigned long client_gfn);
>>>>>>
>>>>>> +/* Allows to deduplicate a range of memory of a client domain. Using
>>>>>> + * this function is equivalent of calling xc_memshr_nominate_gfn for 
>>>>>> each gfn
>>>>>> + * in the two domains followed by xc_memshr_share_gfns.
>>>>>> + *
>>>>>> + * May fail with -EINVAL if the source and client domain have different
>>>>>> + * memory size or if memory sharing is not enabled on either of the 
>>>>>> domains.
>>>>>> + * May also fail with -ENOMEM if there isn't enough memory available to 
>>>>>> store
>>>>>> + * the sharing metadata before deduplication can happen.
>>>>>> + */
>>>>>> +int xc_memshr_range_share(xc_interface *xch,
>>>>>> +                          domid_t source_domain,
>>>>>> +                          domid_t client_domain,
>>>>>> +                          uint64_t start,
>>>>>> +                          uint64_t end);
>>>>>> +
>>>>>>  /* Debug calls: return the number of pages referencing the shared frame 
>>>>>> backing
>>>>>>   * the input argument. Should be one or greater.
>>>>>>   *
>>>>>> diff --git a/tools/libxc/xc_memshr.c b/tools/libxc/xc_memshr.c
>>>>>> index deb0aa4..2b871c7 100644
>>>>>> --- a/tools/libxc/xc_memshr.c
>>>>>> +++ b/tools/libxc/xc_memshr.c
>>>>>> @@ -181,6 +181,25 @@ int xc_memshr_add_to_physmap(xc_interface *xch,
>>>>>>      return xc_memshr_memop(xch, source_domain, &mso);
>>>>>>  }
>>>>>>
>>>>>> +int xc_memshr_range_share(xc_interface *xch,
>>>>>> +                          domid_t source_domain,
>>>>>> +                          domid_t client_domain,
>>>>>> +                          uint64_t start,
>>>>>> +                          uint64_t end)
>>>>>> +{
>>>>>> +    xen_mem_sharing_op_t mso;
>>>>>> +
>>>>>> +    memset(&mso, 0, sizeof(mso));
>>>>>> +
>>>>>> +    mso.op = XENMEM_sharing_op_range_share;
>>>>>> +
>>>>>> +    mso.u.range.client_domain = client_domain;
>>>>>> +    mso.u.range.start = start;
>>>>>> +    mso.u.range.end = end;
>>>>>> +
>>>>>> +    return xc_memshr_memop(xch, source_domain, &mso);
>>>>>> +}
>>>>>> +
>>>>>>  int xc_memshr_domain_resume(xc_interface *xch,
>>>>>>                              domid_t domid)
>>>>>>  {
>>>>>> diff --git a/tools/tests/mem-sharing/memshrtool.c 
>>>>>> b/tools/tests/mem-sharing/memshrtool.c
>>>>>> index 437c7c9..2af6a9e 100644
>>>>>> --- a/tools/tests/mem-sharing/memshrtool.c
>>>>>> +++ b/tools/tests/mem-sharing/memshrtool.c
>>>>>> @@ -24,6 +24,8 @@ static int usage(const char* prog)
>>>>>>      printf("  nominate <domid> <gfn>  - Nominate a page for 
>>>>>> sharing.\n");
>>>>>>      printf("  share <domid> <gfn> <handle> <source> <source-gfn> 
>>>>>> <source-handle>\n");
>>>>>>      printf("                          - Share two pages.\n");
>>>>>> +    printf("  range <source-domid> <destination-domid> <start-gfn> 
>>>>>> <end-gfn>\n");
>>>>>> +    printf("                          - Share pages between domains in 
>>>>>> range.\n");
>>>>>>      printf("  unshare <domid> <gfn>   - Unshare a page by grabbing a 
>>>>>> writable map.\n");
>>>>>>      printf("  add-to-physmap <domid> <gfn> <source> <source-gfn> 
>>>>>> <source-handle>\n");
>>>>>>      printf("                          - Populate a page in a domain 
>>>>>> with a shared page.\n");
>>>>>> @@ -180,6 +182,26 @@ int main(int argc, const char** argv)
>>>>>>          }
>>>>>>          printf("Audit returned %d errors.\n", rc);
>>>>>>      }
>>>>>> +    else if( !strcasecmp(cmd, "range") )
>>>>>> +    {
>>>>>> +        domid_t sdomid, cdomid;
>>>>>> +        int rc;
>>>>>> +        uint64_t start, end;
>>>>>> +
>>>>>> +        if ( argc != 6 )
>>>>>> +            return usage(argv[0]);
>>>>>>
>>>>>> +        sdomid = strtol(argv[2], NULL, 0);
>>>>>> +        cdomid = strtol(argv[3], NULL, 0);
>>>>>> +        start = strtoul(argv[4], NULL, 0);
>>>>>> +        end = strtoul(argv[5], NULL, 0);
>>>>>> +
>>>>>> +        rc = xc_memshr_range_share(xch, sdomid, cdomid, start, end);
>>>>>> +        if ( rc < 0 )
>>>>>> +        {
>>>>>> +            printf("error executing xc_memshr_range_share: %s\n", 
>>>>>> strerror(errno));
>>>>>> +            return rc;
>>>>>> +        }
>>>>>> +    }
>>>>>>      return 0;
>>>>>>  }
>>>>>> diff --git a/xen/arch/x86/mm/mem_sharing.c 
>>>>>> b/xen/arch/x86/mm/mem_sharing.c
>>>>>> index a522423..329fbd9 100644
>>>>>> --- a/xen/arch/x86/mm/mem_sharing.c
>>>>>> +++ b/xen/arch/x86/mm/mem_sharing.c
>>>>>> @@ -1294,6 +1294,58 @@ int relinquish_shared_pages(struct domain *d)
>>>>>>      return rc;
>>>>>>  }
>>>>>>
>>>>>> +static int range_share(struct domain *d, struct domain *cd,
>>>>>> +                       struct mem_sharing_op_range *range)
>>>>>> +{
>>>>>> +    int rc = 0;
>>>>>> +    shr_handle_t sh, ch;
>>>>>> +    unsigned long start = range->_scratchspace ?: range->start;
>>>>>> +
>>>>>> +    while( range->end >= start )
>>>>>> +    {
>>>>>> +        /*
>>>>>> +         * We only break out if we run out of memory as individual 
>>>>>> pages may
>>>>>> +         * legitimately be unsharable and we just want to skip over 
>>>>>> those.
>>>>>> +         */
>>>>>> +        rc = mem_sharing_nominate_page(d, start, 0, &sh);
>>>>>> +        if ( rc == -ENOMEM )
>>>>>> +            break;
>>>>>> +
>>>>>> +        if ( !rc )
>>>>>> +        {
>>>>>> +            rc = mem_sharing_nominate_page(cd, start, 0, &ch);
>>>>>> +            if ( rc == -ENOMEM )
>>>>>> +                break;
>>>>>> +
>>>>>> +            if ( !rc )
>>>>>> +            {
>>>>>> +                /* If we get here this should be guaranteed to succeed. 
>>>>>> */
>>>>>> +                rc = mem_sharing_share_pages(d, start, sh,
>>>>>> +                                             cd, start, ch);
>>>>>> +                ASSERT(!rc);
>>>>>> +            }
>>>>>> +        }
>>>>>> +
>>>>>> +        /* Check for continuation if it's not the last iteration. */
>>>>>> +        if ( range->end >= ++start && hypercall_preempt_check() )
>>>>>> +        {
>>>>>> +            rc = 1;
>>>>>> +            break;
>>>>>> +        }
>>>>>> +    }
>>>>>> +
>>>>>> +    range->_scratchspace = start;
>>>>>> +
>>>>>> +    /*
>>>>>> +     * The last page may fail with -EINVAL, and for range sharing we 
>>>>>> don't
>>>>>> +     * care about that.
>>>>>> +     */
>>>>>> +    if ( range->end < start && rc == -EINVAL )
>>>>>> +        rc = 0;
>>>>>> +
>>>>>> +    return rc;
>>>>>> +}
>>>>>> +
>>>>>>  int mem_sharing_memop(XEN_GUEST_HANDLE_PARAM(xen_mem_sharing_op_t) arg)
>>>>>>  {
>>>>>>      int rc;
>>>>>> @@ -1468,6 +1520,94 @@ int 
>>>>>> mem_sharing_memop(XEN_GUEST_HANDLE_PARAM(xen_mem_sharing_op_t) arg)
>>>>>>          }
>>>>>>          break;
>>>>>>
>>>>>> +        case XENMEM_sharing_op_range_share:
>>>>>> +        {
>>>>>> +            unsigned long max_sgfn, max_cgfn;
>>>>>> +            struct domain *cd;
>>>>>> +
>>>>>> +            rc = -EINVAL;
>>>>>> +            if ( mso.u.range._pad[0] || mso.u.range._pad[1] ||
>>>>>> +                 mso.u.range._pad[2] )
>>>>>> +                 goto out;
>>>>>> +
>>>>>> +            /*
>>>>>> +             * We use _scratchscape for the hypercall continuation 
>>>>>> value.
>>>>>> +             * Ideally the user sets this to 0 in the beginning but
>>>>>> +             * there is no good way of enforcing that here, so we just 
>>>>>> check
>>>>>> +             * that it's at least in range.
>>>>>> +             */
>>>>>> +            if ( mso.u.range._scratchspace &&
>>>>>> +                 (mso.u.range._scratchspace < mso.u.range.start ||
>>>>>> +                  mso.u.range._scratchspace > mso.u.range.end) )
>>>>>> +                goto out;
>>>>>> +
>>>>>> +            if ( !mem_sharing_enabled(d) )
>>>>>> +                goto out;
>>>>>> +
>>>>>> +            rc = 
>>>>>> rcu_lock_live_remote_domain_by_id(mso.u.range.client_domain,
>>>>>> +                                                   &cd);
>>>>>> +            if ( rc )
>>>>>> +                goto out;
>>>>>> +
>>>>>> +            /*
>>>>>> +             * We reuse XENMEM_sharing_op_share XSM check here as this 
>>>>>> is
>>>>>> +             * essentially the same concept repeated over multiple 
>>>>>> pages.
>>>>>> +             */
>>>>>> +            rc = xsm_mem_sharing_op(XSM_DM_PRIV, d, cd,
>>>>>> +                                    XENMEM_sharing_op_share);
>>>>>> +            if ( rc )
>>>>>> +            {
>>>>>> +                rcu_unlock_domain(cd);
>>>>>> +                goto out;
>>>>>> +            }
>>>>>> +
>>>>>> +            if ( !mem_sharing_enabled(cd) )
>>>>>> +            {
>>>>>> +                rcu_unlock_domain(cd);
>>>>>> +                rc = -EINVAL;
>>>>>> +                goto out;
>>>>>> +            }
>>>>>> +
>>>>>> +            /*
>>>>>> +             * Sanity check only, the client should keep the domains 
>>>>>> paused for
>>>>>> +             * the duration of this op.
>>>>>> +             */
>>>>>> +            if ( !atomic_read(&d->pause_count) ||
>>>>>> +                 !atomic_read(&cd->pause_count) )
>>>>>> +            {
>>>>>> +                rcu_unlock_domain(cd);
>>>>>> +                rc = -EINVAL;
>>>>>> +                goto out;
>>>>>> +            }
>>>>>> +
>>>>>> +            max_sgfn = domain_get_maximum_gpfn(d);
>>>>>> +            max_cgfn = domain_get_maximum_gpfn(cd);
>>>>>> +
>>>>>> +            if ( max_sgfn < mso.u.range.start || max_sgfn < 
>>>>>> mso.u.range.end ||
>>>>>> +                 max_cgfn < mso.u.range.start || max_cgfn < 
>>>>>> mso.u.range.end )
>>>>>> +            {
>>>>>> +                rcu_unlock_domain(cd);
>>>>>> +                rc = -EINVAL;
>>>>>> +                goto out;
>>>>>> +            }
>>>>>> +
>>>>>> +            rc = range_share(d, cd, &mso.u.range);
>>>>>> +            rcu_unlock_domain(cd);
>>>>>> +
>>>>>> +            if ( rc > 0 )
>>>>>> +            {
>>>>>> +                if ( __copy_to_guest(arg, &mso, 1) )
>>>>>> +                    rc = -EFAULT;
>>>>>> +                else
>>>>>> +                    rc = 
>>>>>> hypercall_create_continuation(__HYPERVISOR_memory_op,
>>>>>> +                                                       "lh", 
>>>>>> XENMEM_sharing_op,
>>>>>> +                                                       arg);
>>>>>> +            }
>>>>>> +            else
>>>>>> +                mso.u.range._scratchspace = 0;
>>>>>> +        }
>>>>>> +        break;
>>>>>> +
>>>>>>          case XENMEM_sharing_op_debug_gfn:
>>>>>>          {
>>>>>>              unsigned long gfn = mso.u.debug.u.gfn;
>>>>>> diff --git a/xen/include/public/memory.h b/xen/include/public/memory.h
>>>>>> index 29ec571..e0bc018 100644
>>>>>> --- a/xen/include/public/memory.h
>>>>>> +++ b/xen/include/public/memory.h
>>>>>> @@ -465,6 +465,7 @@ DEFINE_XEN_GUEST_HANDLE(xen_mem_access_op_t);
>>>>>>  #define XENMEM_sharing_op_debug_gref        5
>>>>>>  #define XENMEM_sharing_op_add_physmap       6
>>>>>>  #define XENMEM_sharing_op_audit             7
>>>>>> +#define XENMEM_sharing_op_range_share       8
>>>>>>
>>>>>>  #define XENMEM_SHARING_OP_S_HANDLE_INVALID  (-10)
>>>>>>  #define XENMEM_SHARING_OP_C_HANDLE_INVALID  (-9)
>>>>>> @@ -500,7 +501,14 @@ struct xen_mem_sharing_op {
>>>>>>              uint64_aligned_t client_gfn;    /* IN: the client gfn */
>>>>>>              uint64_aligned_t client_handle; /* IN: handle to the client 
>>>>>> page */
>>>>>>              domid_t  client_domain; /* IN: the client domain id */
>>>>>> -        } share;
>>>>>> +        } share;
>>>>>> +        struct mem_sharing_op_range {         /* OP_RANGE_SHARE */
>>>>>> +            uint64_aligned_t start;          /* IN: start gfn. */
>>>>>> +            uint64_aligned_t end;            /* IN: end gfn (inclusive) 
>>>>>> */
>>>>>> +            uint64_aligned_t _scratchspace;  /* Must be set to 0 */
>>>>> Tamas,
>>>>>
>>>>> Why include this "scratchspace" that's not used in the interface,
>>>>> rather than just doing what the memory operations in
>>>>> xen/common/memory.c do, and store it in EAX by shifting it over by
>>>>> MEMOP_EXTENT_SHIFT?  I looked through the history and I see that v1
>>>>> did something very much like that, but it was changed to using the
>>>>> scratch space without any explanation.
>>>>>
>>>>> Having this in the public interface is ugly; and what's worse, it
>>>>> exposes some of the internal mechanism to the guest.
>>>>>
>>>>>  -George
>>>> Well, that's exactly what I did in an earlier version of the patch but
>>>> it was requested that I change it to something like this by Andrew
>>>> (see https://lists.xen.org/archives/html/xen-devel/2015-10/msg00434.html).
>>>> Then over the various iterations it ended up like looking like this.
>>> Oh right -- sorry, I did look but somehow I missed that Andrew had 
>>> requested it.
>>>
>>> I would have read his comment to  mean to put the _scratchspace
>>> variable in the larger structure.  But it has his R-b, so I'll
>>> consider myself answered.
>> I am open to other suggestion wrt naming, but wasn't looking to bikeshed
>> the issue.
> My question wasn't about naming so much as the idea of stashing internal
> implementation mechanism in the public interface in the first place,
> rather than just doing what all the other memory operations do, which is
> stash it in EAX.

Stashing in eax is a gross hack ^W necessary resort (for a previous XSA
iirc) where we needed to retrofit continuations into an interface which
couldn't support them.

Since then, all new interfaces liable to require continuations have been
having explicit room to do so, to avoid stashing in eax.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.