Re: [Qemu-devel] [PATCH v2] migration: avoid filling ignore-shared ramblock when in incoming migration

[Catherine Ho]

Hi Peter Xu

On Thu, 4 Apr 2019 at 12:25, Peter Xu <address@hidden> wrote:

> On Wed, Apr 03, 2019 at 11:21:47PM +0800, Catherine Ho wrote:
> > Hi Peter Xu
> >
> > On Wed, 3 Apr 2019 at 10:25, Peter Xu <address@hidden> wrote:
> >
> > > On Tue, Apr 02, 2019 at 11:30:01AM -0400, Catherine Ho wrote:
> > > > Commit 18269069c310 ("migration: Introduce ignore-shared capability")
> > > > addes ignore-shared capability to bypass the shared ramblock (e,g,
> > > > membackend + numa node). It does good to live migration.
> > > >
> > > > This commit expectes that QEMU doesn't write to guest RAM until
> > > > VM starts, but it does on aarch64 qemu:
> > > > Backtrace:
> > > > 1  0x000055f4a296dd84 in address_space_write_rom_internal () at
> > > exec.c:3458
> > > > 2  0x000055f4a296de3a in address_space_write_rom () at exec.c:3479
> > > > 3  0x000055f4a2d519ff in rom_reset () at hw/core/loader.c:1101
> > > > 4  0x000055f4a2d475ec in qemu_devices_reset () at hw/core/reset.c:69
> > > > 5  0x000055f4a2c90a28 in qemu_system_reset () at vl.c:1675
> > > > 6  0x000055f4a2c9851d in main () at vl.c:4552
> > > >
> > > > Actually, on arm64 virt marchine, ramblock "dtb" will be filled into
> ram
> > > > during rom_reset. In ignore-shared incoming case, this rom filling
> > > > is not required since all the data has been stored in memory backend
> > > file.
> > > >
> > > > Fixes: commit 18269069c310 ("migration: Introduce ignore-shared
> > > capability")
> > > > Signed-off-by: Catherine Ho <address@hidden>
> > > > Suggested-by: Yury Kotov <address@hidden>
> > >
> > > (note: IIUC normally you should have your signed-off to be the last
> > >  line before the suggested-by :)
> > >
> > > About the patch content, I have had a question on whether we should
> > > need to check ignore-shared at all... That question lies in:
> > >
> > > https://patchwork.kernel.org/patch/10859889/#22546487
> > >
> > > And if my understanding was correct above, IMHO the patch could be as
> > > simply be as "if (runstate_check(RUN_STATE_INMIGRATE)) return;" at [1]
> > > below.
> > >
> > >
> > Thanks， but I thought this method would break the x86 rom_reset logic
> during
> > RUN_STATE_INMIGRATE.
> > Please see the debugging patch and log lines below:
> > diff --git a/hw/core/loader.c b/hw/core/loader.c
> > index fe5cb24122..b0c871af26 100644
> > --- a/hw/core/loader.c
> > +++ b/hw/core/loader.c
> > @@ -1086,8 +1086,9 @@ int rom_add_option(const char *file, int32_t
> > bootindex)
> >  static void rom_reset(void *unused)
> >  {
> >      Rom *rom;
> > -
> >      QTAILQ_FOREACH(rom, &roms, next) {
> > +        if (runstate_check(RUN_STATE_INMIGRATE))
> > +           printf("rom name=%s\n",rom->name);
> >          if (rom->fw_file) {
> >              continue;
> >          }
> >
> > rom name=kvmvapic.bin
> > rom name=linuxboot_dma.bin
> > rom name=bios-256k.bin
> > rom name=etc/acpi/tables
> > rom name=etc/table-loader
> > rom name=etc/acpi/rsdp
>
> Hi, Catherine,
>
> I only see that rom names were dumped.  Could you help explain what is
> broken?  Thanks,
>
> Sorry, I have another concern here. What if there is no memory_backend
file?
If there is no memory backend file (i.e. without -object
memory-backend-file,id=dimm1,size=512M,mem-path=/path/memory)

Should the rom blobs still be written into ram in such case?

B.R.
Catherine

> --
> Peter Xu
>