On Wed, Nov 26, 2025 at 1:31 PM Ian Forbes <[email protected]> wrote:
>
> On Thu, Nov 20, 2025 at 10:43 AM Zack Rusin <[email protected]> wrote:
>
> >
> > This bit looks suspicious. It looks like if ctx->merge_dups is false
> > and the resource is doomed then we'll try to remove an invalid node
> > (because it has not been added to the hash).
> >
> > Other than that, it looks great.
> >
> >
> > z
>
> It does look suspicious but since the hashtable is intrusive we are
> just nulling out the node which is still zeroed at this point.

BTW hash initialization, you've removed the vmw_sw_context_init that
was calling hash_init and started relying on .res_ht = {}, which while
should work right now because hlist just wants the pointers nulled, it
might break in the future because we're not exactly following the hash
api. I think at the very least we want to call hash_init after
DECLARE_VAL_CONTEXT to properly initialize that hashtable. Anyway, I'd
probably shield that deletion just in case anyway, even if it's just
because it looks confusing.

Did lowering the VMW_RES_HT_ORDER affect any of the benchmarks? e.g.
unigine heaven or valley ?

z

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

Reply via email to