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

Re: [Xen-devel] [PATCH v2 3/3] x86/ioreq server: Add HVMOP to map guest ram with p2m_ioreq_server to an ioreq server





On 4/8/2016 9:33 PM, Andrew Cooper wrote:
On 31/03/16 11:53, Yu Zhang wrote:
A new HVMOP - HVMOP_map_mem_type_to_ioreq_server, is added to
let one ioreq server claim/disclaim its responsibility for the
handling of guest pages with p2m type p2m_ioreq_server. Users
of this HVMOP can specify whether the p2m_ioreq_server is supposed
to handle write accesses or read ones or both in a parameter named
flags. For now, we only support one ioreq server for this p2m type,
so once an ioreq server has claimed its ownership, subsequent calls
of the HVMOP_map_mem_type_to_ioreq_server will fail. Users can also
disclaim the ownership of guest ram pages with this p2m type, by
triggering this new HVMOP, with ioreq server id set to the current
owner's and flags parameter set to 0.

For now, both HVMOP_map_mem_type_to_ioreq_server and p2m_ioreq_server
are only supported for HVMs with HAP enabled.

Note that flags parameter(if not 0) of this HVMOP only indicates
which kind of memory accesses are to be forwarded to an ioreq server,
it has impact on the access rights of guest ram pages, but are not
the same. Due to hardware limitations, if only write operations are
to be forwarded, read ones will be performed at full speed, with
no hypervisor intervention. But if read ones are to be forwarded to
an ioreq server, writes will inevitably be trapped into hypervisor,
which means significant performance impact.

Also note that this HVMOP_map_mem_type_to_ioreq_server will not
change the p2m type of any guest ram page, until HVMOP_set_mem_type
is triggered. So normally the steps should be the backend driver
first claims its ownership of guest ram pages with p2m_ioreq_server
type, and then sets the memory type to p2m_ioreq_server for specified
guest ram pages.

Signed-off-by: Paul Durrant <paul.durrant@xxxxxxxxxx>
Signed-off-by: Yu Zhang <yu.c.zhang@xxxxxxxxxxxxxxx>
Cc: Keir Fraser <keir@xxxxxxx>
Cc: Jan Beulich <jbeulich@xxxxxxxx>
Cc: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
Cc: George Dunlap <george.dunlap@xxxxxxxxxxxxx>
Cc: Jun Nakajima <jun.nakajima@xxxxxxxxx>
Cc: Kevin Tian <kevin.tian@xxxxxxxxx>
Cc: Tim Deegan <tim@xxxxxxx>
---
  xen/arch/x86/hvm/emulate.c       | 125 +++++++++++++++++++++++++++++++++++++--
  xen/arch/x86/hvm/hvm.c           |  95 +++++++++++++++++++++++++++--
  xen/arch/x86/mm/hap/nested_hap.c |   2 +-
  xen/arch/x86/mm/p2m-ept.c        |  14 ++++-
  xen/arch/x86/mm/p2m-pt.c         |  25 +++++---
  xen/arch/x86/mm/p2m.c            |  82 +++++++++++++++++++++++++
  xen/arch/x86/mm/shadow/multi.c   |   3 +-
  xen/include/asm-x86/p2m.h        |  36 +++++++++--
  xen/include/public/hvm/hvm_op.h  |  37 ++++++++++++
  9 files changed, 395 insertions(+), 24 deletions(-)

diff --git a/xen/arch/x86/hvm/emulate.c b/xen/arch/x86/hvm/emulate.c
index ddc8007..77a4793 100644
--- a/xen/arch/x86/hvm/emulate.c
+++ b/xen/arch/x86/hvm/emulate.c
@@ -94,11 +94,69 @@ static const struct hvm_io_handler null_handler = {
      .ops = &null_ops
  };

+static int mem_read(const struct hvm_io_handler *io_handler,
+                    uint64_t addr,
+                    uint32_t size,
+                    uint64_t *data)
+{
+    struct domain *currd = current->domain;
+    unsigned long gmfn = paddr_to_pfn(addr);
+    unsigned long offset = addr & ~PAGE_MASK;
+    struct page_info *page = get_page_from_gfn(currd, gmfn, NULL, P2M_UNSHARE);
+    uint8_t *p;
+
+    if ( !page )
+        return X86EMUL_UNHANDLEABLE;
+
+    p = __map_domain_page(page);
+    p += offset;
+    memcpy(data, p, size);

What happens when offset + size crosses the page boundary?


The 'size' is set in hvmemul_linear_mmio_access(), to insure offset + size will not cross the page boundary.


diff --git a/xen/include/public/hvm/hvm_op.h b/xen/include/public/hvm/hvm_op.h
index a1eae52..d46f186 100644
--- a/xen/include/public/hvm/hvm_op.h
+++ b/xen/include/public/hvm/hvm_op.h
@@ -489,6 +489,43 @@ struct xen_hvm_altp2m_op {
  typedef struct xen_hvm_altp2m_op xen_hvm_altp2m_op_t;
  DEFINE_XEN_GUEST_HANDLE(xen_hvm_altp2m_op_t);

+#if defined(__XEN__) || defined(__XEN_TOOLS__)
+
+/*
+ * HVMOP_map_mem_type_to_ioreq_server : map or unmap the IOREQ Server <id>
+ *                                      to specific memroy type <type>
+ *                                      for specific accesses <flags>
+ *
+ * Note that if only write operations are to be forwarded to an ioreq server,
+ * read operations will be performed with no hypervisor intervention. But if
+ * flags indicates that read operations are to be forwarded to an ioreq server,
+ * write operations will inevitably be trapped into hypervisor, whether they
+ * are emulated by hypervisor or forwarded to ioreq server depends on the flags
+ * setting. This situation means significant performance impact.
+ */
+#define HVMOP_map_mem_type_to_ioreq_server 26
+struct xen_hvm_map_mem_type_to_ioreq_server {
+    domid_t domid;      /* IN - domain to be serviced */
+    ioservid_t id;      /* IN - ioreq server id */
+    hvmmem_type_t type; /* IN - memory type */

hvmmem_type_t is an enum and doesn't have a fixed width.  It can't be
used in the public API.

You also have some implicit padding holes as a result of the layout.

Oh, guess I should use uint16_t hvmmem_type(any maybe some pad if
necessary).
Thanks for pointing this out, Andrew. :)

~Andrew

+    uint32_t flags;     /* IN - types of accesses to be forwarded to the
+                           ioreq server. flags with 0 means to unmap the
+                           ioreq server */
+#define _HVMOP_IOREQ_MEM_ACCESS_READ 0
+#define HVMOP_IOREQ_MEM_ACCESS_READ \
+    (1u << _HVMOP_IOREQ_MEM_ACCESS_READ)
+
+#define _HVMOP_IOREQ_MEM_ACCESS_WRITE 1
+#define HVMOP_IOREQ_MEM_ACCESS_WRITE \
+    (1u << _HVMOP_IOREQ_MEM_ACCESS_WRITE)
+};
+typedef struct xen_hvm_map_mem_type_to_ioreq_server
+    xen_hvm_map_mem_type_to_ioreq_server_t;
+DEFINE_XEN_GUEST_HANDLE(xen_hvm_map_mem_type_to_ioreq_server_t);
+
+#endif /* defined(__XEN__) || defined(__XEN_TOOLS__) */
+
+
  #endif /* __XEN_PUBLIC_HVM_HVM_OP_H__ */

  /*



B.R.
Yu

_______________________________________________
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®.