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

Re: [Xen-devel] [PATCH 3 of 9] x86/mm: Refactor possibly deadlocking get_gfn calls



At 14:51 -0500 on 01 Feb (1328107915), Andres Lagar-Cavilla wrote:
>  xen/arch/x86/hvm/emulate.c    |  35 +++++++--------
>  xen/arch/x86/mm/mem_sharing.c |  28 +++++++------
>  xen/include/asm-x86/p2m.h     |  91 
> +++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 122 insertions(+), 32 deletions(-)
> 
> 
> When calling get_gfn multiple times on different gfn's in the same function, 
> we
> can easily deadlock if p2m lookups are locked. Thus, refactor these calls to
> enforce simple deadlock-avoidance rules:
>  - Lowest-numbered domain first
>  - Lowest-numbered gfn first

This is a good idea, and I like the get_two_gfns() abstraction, but: 
 - I think the two_gfns struct should proabbly just live on the stack
   instead of malloc()ing it up every time.  It's not very big.
 - the implementation of get_two_gfns() seems to be very complex; I'm
   sure it could be simplified.  At the very least, you could avoid a
   bit of duplication by just deciding once which order to do the gets
   in and then running all the setu and get code once.

Cheers,

Tim.

> Signed-off-by: Andres Lagar-Cavilla <andres@xxxxxxxxxxxxxxx>
> 
> diff -r 27031a8a4eff -r 3de7e43b130a xen/arch/x86/hvm/emulate.c
> --- a/xen/arch/x86/hvm/emulate.c
> +++ b/xen/arch/x86/hvm/emulate.c
> @@ -660,12 +660,13 @@ static int hvmemul_rep_movs(
>  {
>      struct hvm_emulate_ctxt *hvmemul_ctxt =
>          container_of(ctxt, struct hvm_emulate_ctxt, ctxt);
> -    unsigned long saddr, daddr, bytes, sgfn, dgfn;
> +    unsigned long saddr, daddr, bytes;
>      paddr_t sgpa, dgpa;
>      uint32_t pfec = PFEC_page_present;
> -    p2m_type_t p2mt;
> +    p2m_type_t sp2mt, dp2mt;
>      int rc, df = !!(ctxt->regs->eflags & X86_EFLAGS_DF);
>      char *buf;
> +    struct two_gfns *tg;
>  
>      rc = hvmemul_virtual_to_linear(
>          src_seg, src_offset, bytes_per_rep, reps, hvm_access_read,
> @@ -693,26 +694,25 @@ static int hvmemul_rep_movs(
>      if ( rc != X86EMUL_OKAY )
>          return rc;
>  
> -    /* XXX In a fine-grained p2m locking scenario, we need to sort this
> -     * get_gfn's, or else we might deadlock */
> -    sgfn = sgpa >> PAGE_SHIFT;
> -    (void)get_gfn(current->domain, sgfn, &p2mt);
> -    if ( !p2m_is_ram(p2mt) && !p2m_is_grant(p2mt) )
> +    tg = get_two_gfns(current->domain, sgpa >> PAGE_SHIFT, &sp2mt, NULL, 
> NULL,
> +                      current->domain, dgpa >> PAGE_SHIFT, &dp2mt, NULL, 
> NULL,
> +                      p2m_guest);
> +    if ( !tg )
> +        return X86EMUL_UNHANDLEABLE;
> +
> +    if ( !p2m_is_ram(sp2mt) && !p2m_is_grant(sp2mt) )
>      {
>          rc = hvmemul_do_mmio(
>              sgpa, reps, bytes_per_rep, dgpa, IOREQ_READ, df, NULL);
> -        put_gfn(current->domain, sgfn);
> +        put_two_gfns(&tg);
>          return rc;
>      }
>  
> -    dgfn = dgpa >> PAGE_SHIFT;
> -    (void)get_gfn(current->domain, dgfn, &p2mt);
> -    if ( !p2m_is_ram(p2mt) && !p2m_is_grant(p2mt) )
> +    if ( !p2m_is_ram(dp2mt) && !p2m_is_grant(dp2mt) )
>      {
>          rc = hvmemul_do_mmio(
>              dgpa, reps, bytes_per_rep, sgpa, IOREQ_WRITE, df, NULL);
> -        put_gfn(current->domain, sgfn);
> -        put_gfn(current->domain, dgfn);
> +        put_two_gfns(&tg);
>          return rc;
>      }
>  
> @@ -730,8 +730,7 @@ static int hvmemul_rep_movs(
>       */
>      if ( ((dgpa + bytes_per_rep) > sgpa) && (dgpa < (sgpa + bytes)) )
>      {
> -        put_gfn(current->domain, sgfn);
> -        put_gfn(current->domain, dgfn);
> +        put_two_gfns(&tg);
>          return X86EMUL_UNHANDLEABLE;
>      }
>  
> @@ -743,8 +742,7 @@ static int hvmemul_rep_movs(
>      buf = xmalloc_bytes(bytes);
>      if ( buf == NULL )
>      {
> -        put_gfn(current->domain, sgfn);
> -        put_gfn(current->domain, dgfn);
> +        put_two_gfns(&tg);
>          return X86EMUL_UNHANDLEABLE;
>      }
>  
> @@ -757,8 +755,7 @@ static int hvmemul_rep_movs(
>          rc = hvm_copy_to_guest_phys(dgpa, buf, bytes);
>  
>      xfree(buf);
> -    put_gfn(current->domain, sgfn);
> -    put_gfn(current->domain, dgfn);
> +    put_two_gfns(&tg);
>  
>      if ( rc == HVMCOPY_gfn_paged_out )
>          return X86EMUL_RETRY;
> diff -r 27031a8a4eff -r 3de7e43b130a xen/arch/x86/mm/mem_sharing.c
> --- a/xen/arch/x86/mm/mem_sharing.c
> +++ b/xen/arch/x86/mm/mem_sharing.c
> @@ -718,11 +718,13 @@ int mem_sharing_share_pages(struct domai
>      int ret = -EINVAL;
>      mfn_t smfn, cmfn;
>      p2m_type_t smfn_type, cmfn_type;
> +    struct two_gfns *tg;
>  
> -    /* XXX if sd == cd handle potential deadlock by ordering
> -     * the get_ and put_gfn's */
> -    smfn = get_gfn(sd, sgfn, &smfn_type);
> -    cmfn = get_gfn(cd, cgfn, &cmfn_type);
> +    tg = get_two_gfns(sd, sgfn, &smfn_type, NULL, &smfn,
> +                      cd, cgfn, &cmfn_type, NULL, &cmfn,
> +                      p2m_query);
> +    if ( !tg )
> +        return -ENOMEM;
>  
>      /* This tricky business is to avoid two callers deadlocking if 
>       * grabbing pages in opposite client/source order */
> @@ -819,8 +821,7 @@ int mem_sharing_share_pages(struct domai
>      ret = 0;
>      
>  err_out:
> -    put_gfn(cd, cgfn);
> -    put_gfn(sd, sgfn);
> +    put_two_gfns(&tg);
>      return ret;
>  }
>  
> @@ -834,11 +835,13 @@ int mem_sharing_add_to_physmap(struct do
>      struct gfn_info *gfn_info;
>      struct p2m_domain *p2m = p2m_get_hostp2m(cd);
>      p2m_access_t a;
> -    
> -    /* XXX if sd == cd handle potential deadlock by ordering
> -     * the get_ and put_gfn's */
> -    smfn = get_gfn_query(sd, sgfn, &smfn_type);
> -    cmfn = get_gfn_type_access(p2m, cgfn, &cmfn_type, &a, p2m_query, NULL);
> +    struct two_gfns *tg;
> +
> +    tg = get_two_gfns(sd, sgfn, &smfn_type, NULL, &smfn,
> +                      cd, cgfn, &cmfn_type, &a, &cmfn,
> +                      p2m_query);
> +    if ( !tg )
> +        return -ENOMEM;
>  
>      /* Get the source shared page, check and lock */
>      ret = XEN_DOMCTL_MEM_SHARING_S_HANDLE_INVALID;
> @@ -886,8 +889,7 @@ int mem_sharing_add_to_physmap(struct do
>  err_unlock:
>      mem_sharing_page_unlock(spage);
>  err_out:
> -    put_gfn(cd, cgfn);
> -    put_gfn(sd, sgfn);
> +    put_two_gfns(&tg);
>      return ret;
>  }
>  
> diff -r 27031a8a4eff -r 3de7e43b130a xen/include/asm-x86/p2m.h
> --- a/xen/include/asm-x86/p2m.h
> +++ b/xen/include/asm-x86/p2m.h
> @@ -378,6 +378,97 @@ static inline unsigned long mfn_to_gfn(s
>          return mfn_x(mfn);
>  }
>  
> +/* Deadlock-avoidance scheme when calling get_gfn on different gfn's */
> +struct two_gfns {
> +    struct domain  *first_domain;
> +    unsigned long   first_gfn;
> +    struct domain  *second_domain;
> +    unsigned long   second_gfn;
> +};
> +
> +#define assign_pointers(dest, source)                                       \
> +do {                                                                        \
> +    dest ## _mfn = (source ## mfn) ? (source ## mfn) : &__ ## dest ## _mfn;  
> \
> +    dest ## _a   = (source ## a)   ? (source ## a)   : &__ ## dest ## _a;    
> \
> +    dest ## _t   = (source ## t)   ? (source ## t)   : &__ ## dest ## _t;    
> \
> +} while(0)
> +
> +/* Returns mfn, type and access for potential caller consumption, but any
> + * of those can be NULL */
> +static inline struct two_gfns *get_two_gfns(struct domain *rd, unsigned long 
> rgfn,
> +        p2m_type_t *rt, p2m_access_t *ra, mfn_t *rmfn, struct domain *ld, 
> +        unsigned long lgfn, p2m_type_t *lt, p2m_access_t *la, mfn_t *lmfn,
> +        p2m_query_t q)
> +{
> +    mfn_t           *first_mfn, *second_mfn, __first_mfn, __second_mfn;
> +    p2m_access_t    *first_a, *second_a, __first_a, __second_a;
> +    p2m_type_t      *first_t, *second_t, __first_t, __second_t;
> +    
> +    struct two_gfns *rval = xmalloc(struct two_gfns);
> +    if ( !rval )
> +        return NULL;
> +
> +    if ( rd == ld )
> +    {
> +        rval->first_domain = rval->second_domain = rd;
> +
> +        /* Sort by gfn */
> +        if (rgfn <= lgfn)
> +        {
> +            rval->first_gfn     = rgfn;
> +            rval->second_gfn    = lgfn;
> +            assign_pointers(first, r);
> +            assign_pointers(second, l);
> +        } else {
> +            rval->first_gfn     = lgfn;
> +            rval->second_gfn    = rgfn;
> +            assign_pointers(first, l);
> +            assign_pointers(second, r);
> +        }        
> +    } else {
> +        /* Sort by domain */
> +        if ( rd->domain_id <= ld->domain_id )
> +        {
> +            rval->first_domain  = rd;
> +            rval->first_gfn     = rgfn;
> +            rval->second_domain = ld;
> +            rval->second_gfn    = lgfn;
> +            assign_pointers(first, r);
> +            assign_pointers(second, l);
> +        } else {
> +            rval->first_domain  = ld;
> +            rval->first_gfn     = lgfn;
> +            rval->second_domain = rd;
> +            rval->second_gfn    = rgfn;
> +            assign_pointers(first, l);
> +            assign_pointers(second, r);
> +        }
> +    }
> +
> +    /* Now do the gets */
> +    *first_mfn  = get_gfn_type_access(p2m_get_hostp2m(rval->first_domain), 
> +                                      rval->first_gfn, first_t, first_a, q, 
> NULL);
> +    *second_mfn = get_gfn_type_access(p2m_get_hostp2m(rval->second_domain), 
> +                                      rval->second_gfn, second_t, second_a, 
> q, NULL);
> +
> +    return rval;
> +}
> +
> +static inline void put_two_gfns(struct two_gfns **arg_ptr)
> +{
> +    struct two_gfns *arg;
> +
> +    if ( !arg_ptr || !(*arg_ptr) )
> +        return;
> +
> +    arg = *arg_ptr;
> +    put_gfn(arg->second_domain, arg->second_gfn);
> +    put_gfn(arg->first_domain, arg->first_gfn);
> +
> +    xfree(arg);
> +    *arg_ptr = NULL;
> +}
> +
>  /* Init the datastructures for later use by the p2m code */
>  int p2m_init(struct domain *d);
>  
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-devel

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel


 


Rackspace

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