Re: [PATCH v2 2/4] Memory: Enable writeback for given memory region

[Alex Bennée]

Beata Michalska <address@hidden> writes:

> On Wed, 6 Nov 2019 at 12:20, Richard Henderson
> <address@hidden> wrote:
>>
>> On 11/6/19 12:40 AM, Beata Michalska wrote:
>> > +void qemu_ram_writeback(RAMBlock *block, ram_addr_t start, ram_addr_t 
>> > length)
>> > +{
>> > +    void *addr = ramblock_ptr(block, start);
>> > +
>> > +    /*
>> > +     * The requested range might spread up to the very end of the block
>> > +     */
>> > +    if ((start + length) > block->used_length) {
>> > +        qemu_log("%s: sync range outside the block boundaries: "
>> > +                     "start: " RAM_ADDR_FMT " length: " RAM_ADDR_FMT
>> > +                     " block length: " RAM_ADDR_FMT " Narrowing down ..." 
>> > ,
>> > +                     __func__, start, length, block->used_length);
>> > +        length = block->used_length - start;
>> > +    }
>>
>> qemu_log_mask w/ GUEST_ERROR?  How do we expect the length to overflow?
>
> In theory it shouldn't, at least with current usage.
> I guess the probe_access will make sure of that.
> This was more of a precaution to enable catching potential/future misuses
> aka debugging purpose. I can get rid of that it that's playing too
> safe.

If the internal code might get it wrong and that would be a bug then the
g_assert(), if the values are ultimately from the guest then log with
GUEST_ERROR as Richard suggests.

<snip>

--
Alex Bennée