On Thu, Sep 30, 2021 at 11:15 AM Jane Chu <[email protected]> wrote: > > Hi, Dan, > > On 9/16/2021 1:33 PM, Dan Williams wrote: > > The new consideration of > > whole_page() errors means that the code in nfit_handle_mce() is wrong > > to assume that the length of the error is just 1 cacheline. When that > > bug is fixed the badblocks range will cover the entire page in > > whole_page() notification cases. > > Does this explain what I saw a while ago: inject two poisons > to the same block, and pwrite clears one poison at a time?
Potentially, yes. If you injected poison over 512 bytes then it could take 2 pwrites() of 256-bytes each to clear that poison. In the DAX case nothing is enforcing sector aligned I/O for pwrite().
