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

[Xen-devel] [PATCH v2] xen/arm: p2m: Correctly flush TLB in create_p2m_entries



The p2m is shared between VCPUs for each domain. Currently Xen only flush
TLB on the local PCPU. This could result to mismatch between the mapping in the
p2m and TLBs.

Flush TLB entries used by this domain on every PCPU. The flush can also be
moved out of the loop because:
    - ALLOCATE: only called for dom0 RAM allocation, so the flush is never 
called
    - INSERT: if valid = 1 that would means with have replaced a
    page that already belongs to the domain. A VCPU can write on the wrong page.
    This can append for dom0 with the 1:1 mapping because the mapping is not
    removed from the p2m.
    - REMOVE: except for grant-table (replace_grant_host_mapping), each
    call to guest_physmap_remove_page are protected by the callers via a
        get_page -> .... -> guest_physmap_remove_page -> ... -> put_page. So
    the page can't be allocated for another domain until the last put_page.
    - RELINQUISH : the domain is not running anymore so we don't care...

Signed-off-by: Julien Grall <julien.grall@xxxxxxxxxx>

---
    Changes in v2:
        - Switch to the domain for only flush its TLBs entries
        - Move the flush out of the loop

This is a possible bug fix (found by reading the code) for Xen 4.4, I moved the
flush out of the loop which should be safe (see why in the commit message).
Without this patch, the guest can have stale TLB entries when the VCPU is moved
to another PCPU.

Except grant-table (I can't find {get,put}_page for grant-table code???),
all the callers are protected by a get_page before removing the page. So if the
another VCPU is trying to access to this page before the flush, it will just
read/write the wrong page.

The downside of this patch is Xen flushes less TLBs. Instead of flushing all 
TLBs
on the current PCPU, Xen flushes TLBs for a specific VMID on every CPUs. This
should be safe because create_p2m_entries only deal with a specific domain.

I don't think I forget case in this function. Let me know if it's the case.
---
 xen/arch/arm/p2m.c |   24 +++++++++++++++++++-----
 1 file changed, 19 insertions(+), 5 deletions(-)

diff --git a/xen/arch/arm/p2m.c b/xen/arch/arm/p2m.c
index 11f4714..ad6f76e 100644
--- a/xen/arch/arm/p2m.c
+++ b/xen/arch/arm/p2m.c
@@ -238,7 +238,7 @@ static int create_p2m_entries(struct domain *d,
                      int mattr,
                      p2m_type_t t)
 {
-    int rc, flush;
+    int rc;
     struct p2m_domain *p2m = &d->arch.p2m;
     lpae_t *first = NULL, *second = NULL, *third = NULL;
     paddr_t addr;
@@ -246,10 +246,14 @@ static int create_p2m_entries(struct domain *d,
                   cur_first_offset = ~0,
                   cur_second_offset = ~0;
     unsigned long count = 0;
+    unsigned int flush = 0;
     bool_t populate = (op == INSERT || op == ALLOCATE);
 
     spin_lock(&p2m->lock);
 
+    if ( d != current->domain )
+        p2m_load_VTTBR(d);
+
     addr = start_gpaddr;
     while ( addr < end_gpaddr )
     {
@@ -316,7 +320,7 @@ static int create_p2m_entries(struct domain *d,
             cur_second_offset = second_table_offset(addr);
         }
 
-        flush = third[third_table_offset(addr)].p2m.valid;
+        flush |= third[third_table_offset(addr)].p2m.valid;
 
         /* Allocate a new RAM page and attach */
         switch (op) {
@@ -373,9 +377,6 @@ static int create_p2m_entries(struct domain *d,
                 break;
         }
 
-        if ( flush )
-            flush_tlb_all_local();
-
         /* Preempt every 2MiB (mapped) or 32 MiB (unmapped) - arbitrary */
         if ( op == RELINQUISH && count >= 0x2000 )
         {
@@ -392,6 +393,16 @@ static int create_p2m_entries(struct domain *d,
         addr += PAGE_SIZE;
     }
 
+    if ( flush )
+    {
+        /* At the beginning of the function, Xen is updating VTTBR
+         * with the domain where the mappings are created. In this
+         * case it's only necessary to flush TLBs on every CPUs with
+         * the current VMID (our domain).
+         */
+        flush_tlb();
+    }
+
     if ( op == ALLOCATE || op == INSERT )
     {
         unsigned long sgfn = paddr_to_pfn(start_gpaddr);
@@ -409,6 +420,9 @@ out:
     if (second) unmap_domain_page(second);
     if (first) unmap_domain_page(first);
 
+    if ( d != current->domain )
+        p2m_load_VTTBR(current->domain);
+
     spin_unlock(&p2m->lock);
 
     return rc;
-- 
1.7.10.4


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


 


Rackspace

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