<html> <head> <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1"> </head> <body> <div>Hi Matthew,</div> <div><br> </div> <div>this doesn't work, I've already tried something similar.</div> <div><br> </div> <div>TTM uses the reverse lookup functionality when migrating BOs between system and device memory. And that doesn't seem to work with pages from a shmem file.</div> <div><br> </div> <div>Regards,</div> <div>Christian.</div> <div><br> </div> <hr style="display:inline-block;width:98%" tabindex="-1"> <div id="divRplyFwdMsg" dir="ltr"><font face="Calibri, sans-serif" style="font-size:11pt" color="#000000"><b>Von:</b> Matthew Auld <matthew.william.auld@gmail.com><br> <b>Gesendet:</b> Donnerstag, 9. September 2021 16:56<br> <b>An:</b> Christian König <ckoenig.leichtzumerken@gmail.com>; Koenig, Christian <Christian.Koenig@amd.com><br> <b>Cc:</b> Thomas Hellström <thomas.hellstrom@linux.intel.com>; ML dri-devel <dri-devel@lists.freedesktop.org><br> <b>Betreff:</b> i915 ttm_tt shmem backend</font> <div> </div> </div> <div class="BodyFragment"><font size="2"><span style="font-size:11pt;"> <div class="PlainText">Hi Christian,<br> <br> We are looking into using shmem as a ttm_tt backend in i915 for cached<br> system memory objects. We would also like to make such objects visible<br> to the i915-gem shrinker, so that they may be swapped out or discarded<br> when under memory pressure.<br> <br> One idea for handling this is roughly something like:<br> - Add a new TTM_PAGE_FLAG_SHMEM flag, or similar.<br> - Skip the ttm_pages_allocated accounting on such objects, similar to<br> how FLAG_SG is already handled.<br> - Skip all the page->mapping and page->index related bits, like in<br> tt_add_mapping, since it looks like these are set and used by shmem.<br> Not sure what functionally this might break, but looks like it's maybe<br> only driver specific?<br> - Skip calling into ttm_bo_swap_out/in and just have<br> ttm_populate/unpopulate handle this directly for such objects.<br> - Make such objects visible to the i915-gem shrinker.<br> <br> Does this approach look acceptable?<br> </div> </span></font></div> </body> </html>