On 2026/5/28 1:00, Lisa Wang wrote:
>> On Mon, Apr 13, 2026 at 07:57:55PM +0800, Miaohe Lin wrote:
> [...snip...]
>>> -
>>> + case MADV_SOFT_DIRTY_SHMEM:
>>> /* The page content should remain unchanged. */
>>> ASSERT_TRUE(check_memory(vaddr, self->page_size));
>>
>> Why we skip check_memory() for case MADV_SOFT_CLEAN_SHMEM?
>
> Because we did not write 0xce to clean_shmem.
> The difference between clean and dirty pagecache is whether we synced
> the dirty pages to disk. However, the difference between clean and dirty
> shmem is whether we initialized it with 0xce. Thus, we need to skip
> check_memory(), which checking 0xce, for clean_shmem.
Got it.
>
>>> + case MADV_HARD_CLEAN_SHMEM:
>>> + case MADV_SOFT_CLEAN_SHMEM:
>>> + /* It is not expected to receive a SIGBUS signal. */
>>> + ASSERT_EQ(setjmp, 0);
> [...snip...]
>>> + if (variant->type == MADV_HARD) {
>>> + check(_metadata, self, addr, MADV_HARD_DIRTY_SHMEM, ret);
>>> + ret = sigsetjmp(signal_jmp_buf, 1);
>>> + if (ret == 0)
>>> + FORCE_READ(*addr);
>>> + check(_metadata, self, addr, READ_ERROR, ret);
>>> + } else {
>>
>> Should we always add a FORCE_READ() just after variant->inject to verify the
>> accessibility
>> to the addr for soft-offline case ?
>>
>> Thanks.
>
> In dirty_shmem, check() already verifies accessibility to the address
> through memcmp(). Thus, we only need FORCE_READ() in clean_shmem. Will
> add this in the next version. Thanks !
Thanks for your work.
.
