On Mon, Jan 30, 2023 at 10:47:25PM +0800, Feng Sun wrote:
> Igor Mammedov <imamm...@redhat.com> 于2023年1月24日周二 18:30写道：
> >
> > On Tue, 17 Jan 2023 19:15:21 +0800
> > Sun Feng <loyo...@gmail.com> wrote:
> >
> > > Migrate from aarch64 host with PAGE_SIZE 64k to 4k failed with following
> > > errors:
> > >
> > > qmp_cmd_name: migrate-incoming, arguments: {"uri": "tcp:[::]:49152"}
> > > {"timestamp": {"seconds": 1673922775, "microseconds": 534702}, "event":
> > > "MIGRATION", "data": {"status": "setup"}}
> > > {"timestamp": {"seconds": 1673922776, "microseconds": 53003}, "event":
> > > "MIGRATION", "data": {"status": "active"}}
> > > 2023-01-17T02:32:56.058827Z qemu-system-aarch64: Length too large:
> > > /rom@etc/acpi/rsdp: 0x10000 > 0x1000: Invalid argument
> >
> > this should mention/explain why it's happening.
> >
> > i.e we now have 4k limit for RSDP, but then source somehow managed to start
> > with 64k
> > allocated to for RSDP. It looks like limit isn't working as expected to me.
>
> 4k limit should be romsize limit. I can see Rom '/rom@etc/acpi/rsdp'
> with romsize:4096, datasize:36.
> RAMBlock's used_length is set with datasize aligned to PAGE_SIZE, so
> it become 64k when PAGE_SIZE is 64k.
> ```
> static
> RAMBlock *qemu_ram_alloc_internal(ram_addr_t size, ram_addr_t max_size,
> void (*resized)(const char*,
> uint64_t length,
> void *host),
> void *host, uint32_t ram_flags,
> MemoryRegion *mr, Error **errp)
> {
> RAMBlock *new_block;
> Error *local_err = NULL;
>
> assert((ram_flags & ~(RAM_SHARED | RAM_RESIZEABLE | RAM_PREALLOC |
> RAM_NORESERVE)) == 0);
> assert(!host ^ (ram_flags & RAM_PREALLOC));
>
> size = HOST_PAGE_ALIGN(size);
> max_size = HOST_PAGE_ALIGN(max_size);
> new_block = g_malloc0(sizeof(*new_block));
> new_block->mr = mr;
> new_block->resized = resized;
> new_block->used_length = size;
> ```
> So when migrate to 4k PAGE_SIZE, it will report the errors.
>
> ramblock information for PAGE_SIZE 64k and 4k.
> ```
> # getconf PAGE_SIZE
> 65536
> # virsh qemu-monitor-command testvm --hmp 'info ramblock'
> Block Name PSize Offset
> Used Total
> mach-virt.ram 64 KiB 0x0000000000000000
> 0x0000000040000000 0x0000000040000000
> virt.flash0 64 KiB 0x0000000040000000
> 0x0000000004000000 0x0000000004000000
> virt.flash1 64 KiB 0x0000000044000000
> 0x0000000004000000 0x0000000004000000
> /rom@etc/acpi/tables 64 KiB 0x0000000048040000
> 0x0000000000020000 0x0000000000200000
> 0000:00:01.2:00.0/virtio-net-pci.rom 64 KiB 0x0000000048000000
> 0x0000000000040000 0x0000000000040000
> /rom@etc/table-loader 64 KiB 0x0000000048240000
> 0x0000000000010000 0x0000000000010000
> /rom@etc/acpi/rsdp 64 KiB 0x0000000048280000
> 0x0000000000010000 0x0000000000010000
>
> # getconf PAGE_SIZE
> 4096
> # virsh qemu-monitor-command testvm --hmp 'info ramblock'
> Block Name PSize Offset
> Used Total
> mach-virt.ram 4 KiB 0x0000000000000000
> 0x0000000800000000 0x0000000800000000
> virt.flash0 4 KiB 0x0000000800000000
> 0x0000000004000000 0x0000000004000000
> virt.flash1 4 KiB 0x0000000804000000
> 0x0000000004000000 0x0000000004000000
> /rom@etc/acpi/tables 4 KiB 0x0000000808000000
> 0x0000000000020000 0x0000000000200000
> /rom@etc/table-loader 4 KiB 0x0000000808200000
> 0x0000000000001000 0x0000000000010000
> /rom@etc/acpi/rsdp 4 KiB 0x0000000808240000
> 0x0000000000001000 0x0000000000001000
> ```
Oh interesting. I don't remember why I decided to align in.
What does the following do (warning: completely untested):
diff --git a/softmmu/physmem.c b/softmmu/physmem.c
index cb998cdf23..5c732101b9 100644
--- a/softmmu/physmem.c
+++ b/softmmu/physmem.c
@@ -2154,7 +2154,7 @@ RAMBlock *qemu_ram_alloc_internal(ram_addr_t size,
ram_addr_t max_size,
RAM_NORESERVE)) == 0);
assert(!host ^ (ram_flags & RAM_PREALLOC));
- size = HOST_PAGE_ALIGN(size);
+ // size = HOST_PAGE_ALIGN(size);
max_size = HOST_PAGE_ALIGN(max_size);
new_block = g_malloc0(sizeof(*new_block));
new_block->mr = mr;
