[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH v2 2/2] hw/pflash: implement update buffer for block writes
|
From: |
Peter Maydell |
|
Subject: |
Re: [PATCH v2 2/2] hw/pflash: implement update buffer for block writes |
|
Date: |
Fri, 12 Jan 2024 16:54:20 +0000 |
On Mon, 8 Jan 2024 at 13:06, Philippe Mathieu-Daudé <philmd@linaro.org> wrote:
>
> Hi Gerd,
>
> On 8/1/24 13:53, Philippe Mathieu-Daudé wrote:
> > From: Gerd Hoffmann <kraxel@redhat.com>
> >
> > Add an update buffer where all block updates are staged.
> > Flush or discard updates properly, so we should never see
> > half-completed block writes in pflash storage.
> >
> > Drop a bunch of FIXME comments ;)
> >
> > Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> > Message-ID: <20240105135855.268064-3-kraxel@redhat.com>
> > Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> > ---
> > hw/block/pflash_cfi01.c | 106 ++++++++++++++++++++++++++++++----------
> > 1 file changed, 80 insertions(+), 26 deletions(-)
> >
> > diff --git a/hw/block/pflash_cfi01.c b/hw/block/pflash_cfi01.c
> > index ce63ba43b6..0120462648 100644
> > --- a/hw/block/pflash_cfi01.c
> > +++ b/hw/block/pflash_cfi01.c
> > @@ -80,16 +80,39 @@ struct PFlashCFI01 {
> > uint16_t ident3;
> > uint8_t cfi_table[0x52];
> > uint64_t counter;
> > - unsigned int writeblock_size;
> > + uint32_t writeblock_size;
> > MemoryRegion mem;
> > char *name;
> > void *storage;
> > VMChangeStateEntry *vmstate;
> > bool old_multiple_chip_handling;
> > +
> > + /* block update buffer */
> > + unsigned char *blk_bytes;
>
> I'd rather use a 'void *' type here, but then we need to
> use a (uinptr_t) cast in pflash_data_write().
>
> > + uint32_t blk_offset;
> > };
> >
> > static int pflash_post_load(void *opaque, int version_id);
> >
> > +static bool pflash_blk_write_state_needed(void *opaque)
> > +{
> > + PFlashCFI01 *pfl = opaque;
> > +
> > + return (pfl->blk_offset != -1);
> > +}
> > +
> > +static const VMStateDescription vmstate_pflash_blk_write = {
> > + .name = "pflash_cfi01_blk_write",
> > + .version_id = 1,
> > + .minimum_version_id = 1,
> > + .needed = pflash_blk_write_state_needed,
> > + .fields = (const VMStateField[]) {
> > + VMSTATE_VBUFFER_UINT32(blk_bytes, PFlashCFI01, 0, NULL,
> > writeblock_size),
>
> I don't get the difference with VMSTATE_VBUFFER_ALLOC_UINT32() which
> sets VMS_ALLOC. In this case pflash_cfi01_realize() does the alloc so
> we don't need VMS_ALLOC?
Yes, that's the idea. A VMS_ALLOC vmstate type means "this
block of memory is dynamically sized at runtime, so when the
migration code is doing inbound migration it needs to
allocate a buffer of the right size first (based on some
state struct field we've already migrated) and then put the
incoming data into it". VMS_VBUFFER means "the size of the buffer
isn't a compile-time constant, so we need to fish it out of
some other state struct field". So:
VMSTATE_VBUFFER_UINT32: we need to migrate (a pointer to) an array
of uint32_t; the size of that is in some other struct field,
but it's a runtime constant and we can assume the memory has
already been allocated
VMSTATE_VBUFFER_ALLOC_UINT32: we need to migrate an array
of uint32_t of variable size dependent on the inbound migration
data, and so the migration code must allocate it
thanks
-- PMM