[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH V7 10/29] machine: memfd-alloc option
From: |
Steven Sistare |
Subject: |
Re: [PATCH V7 10/29] machine: memfd-alloc option |
Date: |
Thu, 10 Mar 2022 13:18:35 -0500 |
User-agent: |
Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:91.0) Gecko/20100101 Thunderbird/91.6.2 |
On 3/10/2022 12:28 PM, Steven Sistare wrote:
> On 3/10/2022 11:00 AM, Igor Mammedov wrote:
>> On Thu, 10 Mar 2022 10:36:08 -0500
>> Steven Sistare <steven.sistare@oracle.com> wrote:
>>
>>> On 3/8/2022 2:20 AM, Igor Mammedov wrote:
>>>> On Tue, 8 Mar 2022 01:50:11 -0500
>>>> "Michael S. Tsirkin" <mst@redhat.com> wrote:
>>>>
>>>>> On Mon, Mar 07, 2022 at 09:41:44AM -0500, Steven Sistare wrote:
>>>>>> On 3/4/2022 5:41 AM, Igor Mammedov wrote:
>>>>>>> On Thu, 3 Mar 2022 12:21:15 -0500
>>>>>>> "Michael S. Tsirkin" <mst@redhat.com> wrote:
>>>>>>>
>>>>>>>> On Wed, Dec 22, 2021 at 11:05:15AM -0800, Steve Sistare wrote:
>>>>>>>>> Allocate anonymous memory using memfd_create if the memfd-alloc
>>>>>>>>> machine
>>>>>>>>> option is set.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
>>>>>>>>> ---
>>>>>>>>> hw/core/machine.c | 19 +++++++++++++++++++
>>>>>>>>> include/hw/boards.h | 1 +
>>>>>>>>> qemu-options.hx | 6 ++++++
>>>>>>>>> softmmu/physmem.c | 47
>>>>>>>>> ++++++++++++++++++++++++++++++++++++++---------
>>>>>>>>> softmmu/vl.c | 1 +
>>>>>>>>> trace-events | 1 +
>>>>>>>>> util/qemu-config.c | 4 ++++
>>>>>>>>> 7 files changed, 70 insertions(+), 9 deletions(-)
>>>>>>>>>
>>>>>>>>> diff --git a/hw/core/machine.c b/hw/core/machine.c
>>>>>>>>> index 53a99ab..7739d88 100644
>>>>>>>>> --- a/hw/core/machine.c
>>>>>>>>> +++ b/hw/core/machine.c
>>>>>>>>> @@ -392,6 +392,20 @@ static void machine_set_mem_merge(Object *obj,
>>>>>>>>> bool value, Error **errp)
>>>>>>>>> ms->mem_merge = value;
>>>>>>>>> }
>>>>>>>>>
>>>>>>>>> +static bool machine_get_memfd_alloc(Object *obj, Error **errp)
>>>>>>>>> +{
>>>>>>>>> + MachineState *ms = MACHINE(obj);
>>>>>>>>> +
>>>>>>>>> + return ms->memfd_alloc;
>>>>>>>>> +}
>>>>>>>>> +
>>>>>>>>> +static void machine_set_memfd_alloc(Object *obj, bool value, Error
>>>>>>>>> **errp)
>>>>>>>>> +{
>>>>>>>>> + MachineState *ms = MACHINE(obj);
>>>>>>>>> +
>>>>>>>>> + ms->memfd_alloc = value;
>>>>>>>>> +}
>>>>>>>>> +
>>>>>>>>> static bool machine_get_usb(Object *obj, Error **errp)
>>>>>>>>> {
>>>>>>>>> MachineState *ms = MACHINE(obj);
>>>>>>>>> @@ -829,6 +843,11 @@ static void machine_class_init(ObjectClass *oc,
>>>>>>>>> void *data)
>>>>>>>>> object_class_property_set_description(oc, "mem-merge",
>>>>>>>>> "Enable/disable memory merge support");
>>>>>>>>>
>>>>>>>>> + object_class_property_add_bool(oc, "memfd-alloc",
>>>>>>>>> + machine_get_memfd_alloc, machine_set_memfd_alloc);
>>>>>>>>> + object_class_property_set_description(oc, "memfd-alloc",
>>>>>>>>> + "Enable/disable allocating anonymous memory using
>>>>>>>>> memfd_create");
>>>>>>>>> +
>>>>>>>>> object_class_property_add_bool(oc, "usb",
>>>>>>>>> machine_get_usb, machine_set_usb);
>>>>>>>>> object_class_property_set_description(oc, "usb",
>>>>>>>>> diff --git a/include/hw/boards.h b/include/hw/boards.h
>>>>>>>>> index 9c1c190..a57d7a0 100644
>>>>>>>>> --- a/include/hw/boards.h
>>>>>>>>> +++ b/include/hw/boards.h
>>>>>>>>> @@ -327,6 +327,7 @@ struct MachineState {
>>>>>>>>> char *dt_compatible;
>>>>>>>>> bool dump_guest_core;
>>>>>>>>> bool mem_merge;
>>>>>>>>> + bool memfd_alloc;
>>>>>>>>> bool usb;
>>>>>>>>> bool usb_disabled;
>>>>>>>>> char *firmware;
>>>>>>>>> diff --git a/qemu-options.hx b/qemu-options.hx
>>>>>>>>> index 7d47510..33c8173 100644
>>>>>>>>> --- a/qemu-options.hx
>>>>>>>>> +++ b/qemu-options.hx
>>>>>>>>> @@ -30,6 +30,7 @@ DEF("machine", HAS_ARG, QEMU_OPTION_machine, \
>>>>>>>>> " vmport=on|off|auto controls emulation of vmport
>>>>>>>>> (default: auto)\n"
>>>>>>>>> " dump-guest-core=on|off include guest memory in
>>>>>>>>> a core dump (default=on)\n"
>>>>>>>>> " mem-merge=on|off controls memory merge support
>>>>>>>>> (default: on)\n"
>>>>>>>>> + " memfd-alloc=on|off controls allocating
>>>>>>>>> anonymous guest RAM using memfd_create (default: off)\n"
>>>>>>>>
>>>>>>>> Question: are there any disadvantages associated with using
>>>>>>>> memfd_create? I guess we are using up an fd, but that seems minor. Any
>>>>>>>> reason not to set to on by default? maybe with a fallback option to
>>>>>>>> disable that?
>>>>>>
>>>>>> Old Linux host kernels, circa 4.1, do not support huge pages for shared
>>>>>> memory.
>>>>>> Also, the tunable to enable huge pages for share memory is different
>>>>>> than for
>>>>>> anon memory, so there could be performance loss if it is not set
>>>>>> correctly.
>>>>>> /sys/kernel/mm/transparent_hugepage/enabled
>>>>>> vs
>>>>>> /sys/kernel/mm/transparent_hugepage/shmem_enabled
>>>>>
>>>>> I guess we can test this when launching the VM, and select
>>>>> a good default.
>>>>>
>>>>>> It might make sense to use memfd_create by default for the secondary
>>>>>> segments.
>>>>>
>>>>> Well there's also KSM now you mention it.
>>>>
>>>> then another quest, is there downside to always using memfd_create
>>>> without any knobs being involved?
>>>
>>> Lower performance if small pages are used (but Michael suggests qemu could
>>> automatically check the tunable and use anon memory instead)
>>>
>>> KSM (same page merging) is not supported for shared memory, so
>>> ram_block_add ->
>>> memory_try_enable_merging will not enable it.
>>>
>>> In both cases, I expect the degradation would be negligible if memfd_create
>>> is
>>> only automatically applied to the secondary segments, which are typically
>>> small.
>>> But, someone's secondary segment could be larger, and it is time consuming
>>> to
>>> prove innocence when someone claims your change caused their performance
>>> regression.
>>
>> Adding David as memory subsystem maintainer, maybe he will a better
>> idea instead of introducing global knob that would also magically alter
>> backends' behavior despite of its their configured settings.
>
> OK, in ram_block_add I can set the RAM_SHARED flag based on the
> memory-backend object's
> shared flag. I already set the latter in create_default_memdev when
> memfd-alloc is
> specified. With that change, we do not override configured settings. Users
> can no longer
> use memory-backend-ram for CPR, and must change all memory-backend-ram to
> memory-backend-memfd
> in the command-line arguments. That is fine.
>
> With that change, are you OK with this patch?
Sorry, I mis-read my own code in ram_block_add. The existing code is correct
and does
not alter any backend's behavior. It only sets the shared flag when the ram
is *not*
being allocated for a backend:
if (!object_dynamic_cast(parent, TYPE_MEMORY_BACKEND)) {
new_block->flags |= RAM_SHARED;
}
- Steve
>>>>>>>> I am concerned that it's actually a kind of memory backend, this flag
>>>>>>>> seems to instead be closer to the deprecated mem-prealloc. E.g.
>>>>>>>> it does not work with a mem path, does it?
>>>>>>
>>>>>> One can still define a memory backend with mempath to create the main
>>>>>> ram segment,
>>>>>> though it must be some form of shared to work with live update. Indeed,
>>>>>> I would
>>>>>> expect most users to specify an explicit memory backend for it. The
>>>>>> secondary
>>>>>> segments would still use memfd_create.
>>>>>>
>>>>>>> (mem path and mem-prealloc are transparently aliased to used memory
>>>>>>> backend
>>>>>>> if I recall it right.)
>>>>>>>
>>>>>>> Steve,
>>>>>>>
>>>>>>> For allocating guest RAM, we switched exclusively to using
>>>>>>> memory-backends
>>>>>>> including initial guest RAM (-m size option) and we have hostmem-memfd
>>>>>>> that uses memfd_create() and I'd rather avoid adding random knobs to
>>>>>>> machine
>>>>>>> for tweaking how RAM should be allocated, we have memory backends for
>>>>>>> this,
>>>>>>> so this patch begs the question: why hostmem-memfd is not sufficient?
>>>>>>> (patch description is rather lacking on rationale behind the patch)
>>>>>>
>>>>>> There is currently no way to specify memory backends for the secondary
>>>>>> memory
>>>>>> segments (vram, roms, etc), and IMO it would be onerous to specify a
>>>>>> backend for
>>>>>> each of them. On x86_64, these include pc.bios, vga.vram, pc.rom,
>>>>>> vga.rom,
>>>>>> /rom@etc/acpi/tables, /rom@etc/table-loader, /rom@etc/acpi/rsdp.
>>
>> MemoryRegion is not the only place where state is stored.
>> If we only talk about fwcfg entries state, it can also reference
>> plain malloced memory allocated elsewhere or make a deep copy internally.
>> Similarly devices also may store state outside of RamBlock framework.
>>
>> How are you dealing with that?
>>
>>>>>>
>>>>>> - Steve
>>>>>>
>>>>>>>>> " aes-key-wrap=on|off controls support for AES
>>>>>>>>> key wrapping (default=on)\n"
>>>>>>>>> " dea-key-wrap=on|off controls support for DEA
>>>>>>>>> key wrapping (default=on)\n"
>>>>>>>>> " suppress-vmdesc=on|off disables self-describing
>>>>>>>>> migration (default=off)\n"
>>>>>>>>> @@ -76,6 +77,11 @@ SRST
>>>>>>>>> supported by the host, de-duplicates identical memory pages
>>>>>>>>> among VMs instances (enabled by default).
>>>>>>>>>
>>>>>>>>> + ``memfd-alloc=on|off``
>>>>>>>>> + Enables or disables allocation of anonymous guest RAM using
>>>>>>>>> + memfd_create. Any associated memory-backend objects are
>>>>>>>>> created with
>>>>>>>>> + share=on. The memfd-alloc default is off.
>>>>>>>>> +
>>>>>>>>> ``aes-key-wrap=on|off``
>>>>>>>>> Enables or disables AES key wrapping support on s390-ccw
>>>>>>>>> hosts.
>>>>>>>>> This feature controls whether AES wrapping keys will be
>>>>>>>>> created
>>>>>>>>> diff --git a/softmmu/physmem.c b/softmmu/physmem.c
>>>>>>>>> index 3524c04..95e2b49 100644
>>>>>>>>> --- a/softmmu/physmem.c
>>>>>>>>> +++ b/softmmu/physmem.c
>>>>>>>>> @@ -41,6 +41,7 @@
>>>>>>>>> #include "qemu/config-file.h"
>>>>>>>>> #include "qemu/error-report.h"
>>>>>>>>> #include "qemu/qemu-print.h"
>>>>>>>>> +#include "qemu/memfd.h"
>>>>>>>>> #include "exec/memory.h"
>>>>>>>>> #include "exec/ioport.h"
>>>>>>>>> #include "sysemu/dma.h"
>>>>>>>>> @@ -1964,35 +1965,63 @@ static void ram_block_add(RAMBlock
>>>>>>>>> *new_block, Error **errp)
>>>>>>>>> const bool shared = qemu_ram_is_shared(new_block);
>>>>>>>>> RAMBlock *block;
>>>>>>>>> RAMBlock *last_block = NULL;
>>>>>>>>> + struct MemoryRegion *mr = new_block->mr;
>>>>>>>>> ram_addr_t old_ram_size, new_ram_size;
>>>>>>>>> Error *err = NULL;
>>>>>>>>> + const char *name;
>>>>>>>>> + void *addr = 0;
>>>>>>>>> + size_t maxlen;
>>>>>>>>> + MachineState *ms = MACHINE(qdev_get_machine());
>>>>>>>>>
>>>>>>>>> old_ram_size = last_ram_page();
>>>>>>>>>
>>>>>>>>> qemu_mutex_lock_ramlist();
>>>>>>>>> - new_block->offset = find_ram_offset(new_block->max_length);
>>>>>>>>> + maxlen = new_block->max_length;
>>>>>>>>> + new_block->offset = find_ram_offset(maxlen);
>>>>>>>>>
>>>>>>>>> if (!new_block->host) {
>>>>>>>>> if (xen_enabled()) {
>>>>>>>>> - xen_ram_alloc(new_block->offset, new_block->max_length,
>>>>>>>>> - new_block->mr, &err);
>>>>>>>>> + xen_ram_alloc(new_block->offset, maxlen, new_block->mr,
>>>>>>>>> &err);
>>>>>>>>> if (err) {
>>>>>>>>> error_propagate(errp, err);
>>>>>>>>> qemu_mutex_unlock_ramlist();
>>>>>>>>> return;
>>>>>>>>> }
>>>>>>>>> } else {
>>>>>>>>> - new_block->host =
>>>>>>>>> qemu_anon_ram_alloc(new_block->max_length,
>>>>>>>>> -
>>>>>>>>> &new_block->mr->align,
>>>>>>>>> - shared, noreserve);
>>>>>>>>> - if (!new_block->host) {
>>>>>>>>> + name = memory_region_name(mr);
>>>>>>>>> + if (ms->memfd_alloc) {
>>>>>>>>> + Object *parent = &mr->parent_obj;
>>>>>>>>> + int mfd = -1; /* placeholder until next
>>>>>>>>> patch */
>>>>>>>>> + mr->align = QEMU_VMALLOC_ALIGN;
>>>>>>>>> + if (mfd < 0) {
>>>>>>>>> + mfd = qemu_memfd_create(name, maxlen + mr->align,
>>>>>>>>> + 0, 0, 0, &err);
>>>>>>>>> + if (mfd < 0) {
>>>>>>>>> + return;
>>>>>>>>> + }
>>>>>>>>> + }
>>>>>>>>> + qemu_set_cloexec(mfd);
>>>>>>>>> + /* The memory backend already set its desired flags.
>>>>>>>>> */
>>>>>>>>> + if (!object_dynamic_cast(parent,
>>>>>>>>> TYPE_MEMORY_BACKEND)) {
>>>>>>>>> + new_block->flags |= RAM_SHARED;
>>>>>>>>> + }
>>>>>>>>> + addr = file_ram_alloc(new_block, maxlen, mfd,
>>>>>>>>> + false, false, 0, errp);
>>>>>>>>> + trace_anon_memfd_alloc(name, maxlen, addr, mfd);
>>>>>>>>> + } else {
>>>>>>>>> + addr = qemu_anon_ram_alloc(maxlen, &mr->align,
>>>>>>>>> + shared, noreserve);
>>>>>>>>> + }
>>>>>>>>> +
>>>>>>>>> + if (!addr) {
>>>>>>>>> error_setg_errno(errp, errno,
>>>>>>>>> "cannot set up guest memory '%s'",
>>>>>>>>> - memory_region_name(new_block->mr));
>>>>>>>>> + name);
>>>>>>>>> qemu_mutex_unlock_ramlist();
>>>>>>>>> return;
>>>>>>>>> }
>>>>>>>>> - memory_try_enable_merging(new_block->host,
>>>>>>>>> new_block->max_length);
>>>>>>>>> + memory_try_enable_merging(addr, maxlen);
>>>>>>>>> + new_block->host = addr;
>>>>>>>>> }
>>>>>>>>> }
>>>>>>>>>
>>>>>>>>> diff --git a/softmmu/vl.c b/softmmu/vl.c
>>>>>>>>> index 620a1f1..ab3648a 100644
>>>>>>>>> --- a/softmmu/vl.c
>>>>>>>>> +++ b/softmmu/vl.c
>>>>>>>>> @@ -2440,6 +2440,7 @@ static void create_default_memdev(MachineState
>>>>>>>>> *ms, const char *path)
>>>>>>>>> object_property_set_str(obj, "mem-path", path, &error_fatal);
>>>>>>>>> }
>>>>>>>>> object_property_set_int(obj, "size", ms->ram_size, &error_fatal);
>>>>>>>>> + object_property_set_bool(obj, "share", ms->memfd_alloc,
>>>>>>>>> &error_fatal);
>>>>>>>>> object_property_add_child(object_get_objects_root(),
>>>>>>>>> mc->default_ram_id,
>>>>>>>>> obj);
>>>>>>>>> /* Ensure backend's memory region name is equal to
>>>>>>>>> mc->default_ram_id */
>>>>>>>>> diff --git a/trace-events b/trace-events
>>>>>>>>> index a637a61..770a9ac 100644
>>>>>>>>> --- a/trace-events
>>>>>>>>> +++ b/trace-events
>>>>>>>>> @@ -45,6 +45,7 @@ ram_block_discard_range(const char *rbname, void
>>>>>>>>> *hva, size_t length, bool need_
>>>>>>>>> # accel/tcg/cputlb.c
>>>>>>>>> memory_notdirty_write_access(uint64_t vaddr, uint64_t ram_addr,
>>>>>>>>> unsigned size) "0x%" PRIx64 " ram_addr 0x%" PRIx64 " size %u"
>>>>>>>>> memory_notdirty_set_dirty(uint64_t vaddr) "0x%" PRIx64
>>>>>>>>> +anon_memfd_alloc(const char *name, size_t size, void *ptr, int fd)
>>>>>>>>> "%s size %zu ptr %p fd %d"
>>>>>>>>>
>>>>>>>>> # gdbstub.c
>>>>>>>>> gdbstub_op_start(const char *device) "Starting gdbstub using device
>>>>>>>>> %s"
>>>>>>>>> diff --git a/util/qemu-config.c b/util/qemu-config.c
>>>>>>>>> index 436ab63..3606e5c 100644
>>>>>>>>> --- a/util/qemu-config.c
>>>>>>>>> +++ b/util/qemu-config.c
>>>>>>>>> @@ -207,6 +207,10 @@ static QemuOptsList machine_opts = {
>>>>>>>>> .type = QEMU_OPT_BOOL,
>>>>>>>>> .help = "enable/disable memory merge support",
>>>>>>>>> },{
>>>>>>>>> + .name = "memfd-alloc",
>>>>>>>>> + .type = QEMU_OPT_BOOL,
>>>>>>>>> + .help = "enable/disable memfd_create for anonymous
>>>>>>>>> memory",
>>>>>>>>> + },{
>>>>>>>>> .name = "usb",
>>>>>>>>> .type = QEMU_OPT_BOOL,
>>>>>>>>> .help = "Set on/off to enable/disable usb",
>>>>>>>>> --
>>>>>>>>> 1.8.3.1
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>
>>>>
>>>
>>
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Steven Sistare, 2022/03/03
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Steven Sistare, 2022/03/03
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Michael S. Tsirkin, 2022/03/03
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Igor Mammedov, 2022/03/04
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Steven Sistare, 2022/03/07
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Michael S. Tsirkin, 2022/03/08
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Igor Mammedov, 2022/03/08
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Steven Sistare, 2022/03/10
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Igor Mammedov, 2022/03/10
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Steven Sistare, 2022/03/10
- Re: [PATCH V7 10/29] machine: memfd-alloc option,
Steven Sistare <=
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Igor Mammedov, 2022/03/11
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Steven Sistare, 2022/03/29
- Re: [PATCH V7 10/29] machine: memfd-alloc option, Daniel P . Berrangé, 2022/03/11
Re: [PATCH V7 10/29] machine: memfd-alloc option, David Hildenbrand, 2022/03/11
Re: [PATCH V7 10/29] machine: memfd-alloc option, David Hildenbrand, 2022/03/11