Re: [Xen-devel] [PATCH v2 3/4] libxc: rework vnuma bits in setup_guest

[Wei Liu <wei.liu2@xxxxxxxxxx>]

On Fri, May 29, 2015 at 11:02:16AM -0400, Boris Ostrovsky wrote:
> On 05/29/2015 07:37 AM, Wei Liu wrote:
> >Make the setup process similar to PV counterpart. That is, to allocate a
> >P2M array that covers the whole memory range and start from there. This
> >is clearer than using an array with no holes in it.
> >
> >Also the dummy layout should take MMIO hole into consideration. We might
> >end up having two vmemranges in the dummy layout.
> >
> >Signed-off-by: Wei Liu <wei.liu2@xxxxxxxxxx>
> >Cc: Ian Campbell <ian.campbell@xxxxxxxxxx>
> >Cc: Ian Jackson <ian.jackson@xxxxxxxxxxxxx>
> >---
> >v2: only generate 1 vnode in dummy layout
> >---
> >  tools/libxc/xc_hvm_build_x86.c | 62 
> > ++++++++++++++++++++++++++++++++----------
> >  1 file changed, 47 insertions(+), 15 deletions(-)
> >
> >diff --git a/tools/libxc/xc_hvm_build_x86.c b/tools/libxc/xc_hvm_build_x86.c
> >index df4b7ed..b3855e0 100644
> >--- a/tools/libxc/xc_hvm_build_x86.c
> >+++ b/tools/libxc/xc_hvm_build_x86.c
> >@@ -238,6 +238,7 @@ static int setup_guest(xc_interface *xch,
> >  {
> >      xen_pfn_t *page_array = NULL;
> >      unsigned long i, vmemid, nr_pages = args->mem_size >> PAGE_SHIFT;
> >+    unsigned long p2m_size;
> >      unsigned long target_pages = args->mem_target >> PAGE_SHIFT;
> >      unsigned long entry_eip, cur_pages, cur_pfn;
> >      void *hvm_info_page;
> >@@ -254,8 +255,8 @@ static int setup_guest(xc_interface *xch,
> >      xen_pfn_t special_array[NR_SPECIAL_PAGES];
> >      xen_pfn_t ioreq_server_array[NR_IOREQ_SERVER_PAGES];
> >      uint64_t total_pages;
> >-    xen_vmemrange_t dummy_vmemrange;
> >-    unsigned int dummy_vnode_to_pnode;
> >+    xen_vmemrange_t dummy_vmemrange[2];
> >+    unsigned int dummy_vnode_to_pnode[1];
> >      memset(&elf, 0, sizeof(elf));
> >      if ( elf_init(&elf, image, image_size) != 0 )
> >@@ -275,17 +276,35 @@ static int setup_guest(xc_interface *xch,
> >      if ( args->nr_vmemranges == 0 )
> >      {
> >-        /* Build dummy vnode information */
> >-        dummy_vmemrange.start = 0;
> >-        dummy_vmemrange.end   = args->mem_size;
> >-        dummy_vmemrange.flags = 0;
> >-        dummy_vmemrange.nid   = 0;
> >+        /* Build dummy vnode information
> >+         *
> >+         * Guest physical address space layout:
> >+         * [0, hole_start) [hole_start, 4G) [4G, highmem_end)
> >+         *
> >+         * Of course if there is no high memory, the second vmemrange
> >+         * has no effect on the actual result.
> >+         */
> >+
> >+        dummy_vmemrange[0].start = 0;
> >+        dummy_vmemrange[0].end   = args->lowmem_end;
> >+        dummy_vmemrange[0].flags = 0;
> >+        dummy_vmemrange[0].nid   = 0;
> >          args->nr_vmemranges = 1;
> >-        args->vmemranges = &dummy_vmemrange;
> >-        dummy_vnode_to_pnode = XC_NUMA_NO_NODE;
> >+        if ( args->highmem_end > (1ULL << 32) )
> >+        {
> >+            dummy_vmemrange[1].start = 1ULL << 32;
> >+            dummy_vmemrange[1].end   = args->highmem_end;
> >+            dummy_vmemrange[1].flags = 0;
> >+            dummy_vmemrange[1].nid   = 0;
> >+
> >+            args->nr_vmemranges++;
> >+        }
> >+
> >+        dummy_vnode_to_pnode[0] = XC_NUMA_NO_NODE;
> >          args->nr_vnodes = 1;
> >-        args->vnode_to_pnode = &dummy_vnode_to_pnode;
> >+        args->vmemranges = dummy_vmemrange;
> >+        args->vnode_to_pnode = dummy_vnode_to_pnode;
> 
> 
> I should have asked last time: we are setting args->vmemranges and
> args->vnode_to_pnode to point to something on the stack. Is it safe to
> assume that these pointers will never be used outside of setup_guest()? (and
> if it is, should they be set to NULL before returning, just in case?)
> 

I think this is a good idea. In libxl we won't read those fields back
because it's one way communication. Cannot speak for other toolstack
that builds directly on top of libxc, so let's err on the safe side.

> I also had a very cursory look at their usage and I think I see in some
> cases memory is allocated (and presumably then freed), but I am not sure
> whether this on some other code path.
> 

The memory allocated to those fields are not managed here in libxc.
The only memory allocated here is page_array and it's subsequently
freed.

Wei.

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