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

Re: [Xen-devel] [PATCH for 4.7] Remove HVMMEM_mmio_write_dm from the public interface.

Thanks Jan. And I admire your rigorous thought. :)

On 4/28/2016 6:57 PM, Jan Beulich wrote:
On 28.04.16 at 12:42, <george.dunlap@xxxxxxxxxx> wrote:
On 28/04/16 11:22, Jan Beulich wrote:
On 28.04.16 at 10:29, <yu.c.zhang@xxxxxxxxxxxxxxx> wrote:
@@ -5529,7 +5527,7 @@ long do_hvm_op(unsigned long op,
             [HVMMEM_ram_rw]  = p2m_ram_rw,
             [HVMMEM_ram_ro]  = p2m_ram_ro,
             [HVMMEM_mmio_dm] = p2m_mmio_dm,
-            [HVMMEM_mmio_write_dm] = p2m_mmio_write_dm
+            [HVMMEM_unused] = p2m_invalid

Why don't you simply delete the old line, without replacement?

Well, I did not delete the old line, because in my coming patch(the
p2m renaming code), I'm planning to introduce the HVMMEM_ioreq_server,
which is HVMMEM_unused+1. And I do not want the check of a.hvmmem_type
against HVMMEN_unused later in this routine appear in that patch.

That might have been slightly cleaner; but we're going to have to put it
back as soon as the development window opens anyway, so I don't really
see the point of going through the effort of respinning the patch again.

Would you be willing to ack this version anyway?

I have no problem doing so (and in fact I have it on my to by
committed list already), it is just looked slightly confusing (and
I had already typed half a reply that this isn't what was discussed
until I properly looked at the next hunk), and hence I wanted to
understand the motivation. And btw., I'm not convinced it would
need to be put there anyway later: I don't view the used
mechanism as a good (read: extensible) one to deal with what
would be holes in the array above. Indeed we can't leave them
uninitialized (as that would mean p2m_ram_rw), but I think we
should better find a way to initialize _all_ unused slots without
requiring an initializer for each of them. Sadly the desire to allow
compilation with clang prohibits the most natural solution:

        static const p2m_type_t memtype[] = {
            [0 ... <upper-bound> - 1] = p2m_invalid,

Not sure if this will compile? Can have a try. :)

            [HVMMEM_ram_rw]  = p2m_ram_rw,
            [HVMMEM_ram_ro]  = p2m_ram_ro,
            [HVMMEM_mmio_dm] = p2m_mmio_dm,

Maybe we could do (altering the second hunk of this patch)

@@ -5553,7 +5551,10 @@ long do_hvm_op(unsigned long op, 
              ((a.first_pfn + a.nr - 1) > domain_get_maximum_gpfn(d)) )
             goto setmemtype_fail;

-        if ( a.hvmmem_type >= ARRAY_SIZE(memtype) )
+        BUILD_BUG_ON(p2m_ram_rw);
+        BUILD_BUG_ON(HVMMEM_ram_rw);
+        if ( a.hvmmem_type >= ARRAY_SIZE(memtype) ||
+             (a.hvmmem_type && !memtype[a.hvmmem_type]) )

I guess by !memtype[a.hvmmem_type] you are trying to check if it's
p2m_invalid? But p2m_ram_rw is 0, and p2m_invalid is 1. So may be it
should be checked like memtype[a.hvmmem_type] < 0 and initialize the
holes with -1.

But I still wonder is this really necessary? Because we only have one
hole in this array in the forseeable future.

             goto setmemtype_fail;

         while ( a.nr > start_iter )



Xen-devel mailing list



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