Re: [PATCH 1/3] memory: add a sparse memory device

["Philippe Mathieu-Daudé" <philmd@redhat.com>]

On 3/11/21 6:36 AM, Alexander Bulekov wrote:
> For testing, it can be useful to simulate an enormous amount of memory
> (e.g. 2^64 RAM). This adds an MMIO device that acts as sparse memory.
> When something writes a nonzero value to a sparse-mem address, we
> allocate a block of memory. This block is kept around, until all of the
> bytes within the block are zero-ed. The device has a very low priority
> (so it can be mapped beneath actual RAM, and virtual device MMIO
> regions).

I'm not convinced we need this, but still added some comments while
reviewing.

> 
> Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
> ---
>  MAINTAINERS         |   1 +
>  hw/mem/meson.build  |   1 +
>  hw/mem/sparse-mem.c | 154 ++++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 156 insertions(+)
>  create mode 100644 hw/mem/sparse-mem.c
> 
> diff --git a/MAINTAINERS b/MAINTAINERS
> index f22d83c178..9e3d8b1401 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -2618,6 +2618,7 @@ R: Thomas Huth <thuth@redhat.com>
>  S: Maintained
>  F: tests/qtest/fuzz/
>  F: scripts/oss-fuzz/
> +F: hw/mem/sparse-mem.c
>  F: docs/devel/fuzzing.rst
>  
>  Register API
> diff --git a/hw/mem/meson.build b/hw/mem/meson.build
> index 0d22f2b572..732f459e0a 100644
> --- a/hw/mem/meson.build
> +++ b/hw/mem/meson.build
> @@ -1,5 +1,6 @@
>  mem_ss = ss.source_set()
>  mem_ss.add(files('memory-device.c'))
> +mem_ss.add(files('sparse-mem.c'))
>  mem_ss.add(when: 'CONFIG_DIMM', if_true: files('pc-dimm.c'))
>  mem_ss.add(when: 'CONFIG_NPCM7XX', if_true: files('npcm7xx_mc.c'))
>  mem_ss.add(when: 'CONFIG_NVDIMM', if_true: files('nvdimm.c'))
> diff --git a/hw/mem/sparse-mem.c b/hw/mem/sparse-mem.c
> new file mode 100644
> index 0000000000..ffda6f76b4
> --- /dev/null
> +++ b/hw/mem/sparse-mem.c
> @@ -0,0 +1,154 @@
> +/*
> + * A sparse memory device
> + *
> + * Copyright Red Hat Inc., 2021
> + *
> + * Authors:
> + *  Alexander Bulekov   <alxndr@bu.edu>
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> + * See the COPYING file in the top-level directory.
> + */
> +
> +#include "qemu/osdep.h"
> +
> +#include "exec/address-spaces.h"
> +#include "hw/qdev-properties.h"
> +
> +#define TYPE_SPARSE_MEM "sparse-mem"
> +#define SPARSE_MEM(obj) OBJECT_CHECK(SparseMemState, (obj), TYPE_SPARSE_MEM)
> +
> +#define SPARSE_BLOCK_SIZE 0x1000
> +
> +typedef struct SparseMemState {
> +    DeviceState parent_obj;
> +    MemoryRegion mmio;
> +    uint64_t baseaddr;
> +    uint64_t length;
> +    uint64_t usage;

usage -> size_used?

> +    uint64_t maxsize;
> +    GHashTable *mapped;
> +} SparseMemState;
> +
> +typedef struct sparse_mem_block {
> +    uint16_t nonzeros;
> +    uint8_t data[SPARSE_BLOCK_SIZE];
> +} sparse_mem_block;
> +

> +static const MemoryRegionOps sparse_mem_ops = {
> +    .read = sparse_mem_read,
> +    .write = sparse_mem_write,
> +    .endianness = DEVICE_LITTLE_ENDIAN,
> +    .valid = {
> +            .min_access_size = 1,
> +            .max_access_size = 8,
> +            .unaligned = false,

Why restrict unaligned accesses?

> +        },
> +};
> +
> +static Property sparse_mem_properties[] = {
> +    /* The base address of the memory */
> +    DEFINE_PROP_UINT64("baseaddr", SparseMemState, baseaddr, 0x0),
> +    /* The length of the sparse memory region */
> +    DEFINE_PROP_UINT64("length", SparseMemState, length, UINT64_MAX),
> +    /* Max amount of actual memory that can be used to back the sparse memory */
> +    DEFINE_PROP_UINT64("maxsize", SparseMemState, maxsize, 0x100000),

0x100000 -> 1 * MiB

> +    DEFINE_PROP_END_OF_LIST(),
> +};
> +
> +static void sparse_mem_realize(DeviceState *dev, Error **errp)
> +{
> +    SparseMemState *s = SPARSE_MEM(dev);

Anyhow, we should restrict this device to QTest accelerator, right?

Maybe:
    if (!qtest_enabled()) {
        error_setg(errp, "sparse_mem device requires QTest");
        return;
    }

> +
> +    assert(s->baseaddr + s->length > s->baseaddr);

Don't you need more than 64-bit to do this check?

> +
> +    s->mapped = g_hash_table_new(NULL, NULL);
> +    memory_region_init_io(&(s->mmio), OBJECT(s), &sparse_mem_ops, s,
> +                          "sparse-mem", s->length);
> +    memory_region_add_subregion_overlap(get_system_memory(), s->baseaddr,
> +                                        &(s->mmio), -100);

mr_add() to sysmem from a non-sysbus device is odd... Maybe it is
acceptable, I don't know enough.

> +}
> +