Hey, Here's a small set of improvements that I was hoping you would consider as part of your Soft-Reserved series[0]; should you think they make sense. It does the following:
Patch 1-2: Add an align sysfs attribute, as opposed to being limited to 2M. This brings parity to static dax regions, which support 1G. Patch 3. Add a module parameter to hmem, to initialize the region as idle with full available_size for child devices. When the region is gonna be partiotined by default and assigned to guests, doesn't help to initialize the region. Also if the majority of the region starts non-idle and it's a region which memmap is bigger than System-Ram, then there not be enough space for the device to be probed successfully. Patch 4: Add an sysfs attribute for range allocation. It is a single entry where you pass a range <start>-<end> and where the ordering plus range length designate the page_offset. It is meant for recreate the same mappings after kexec (retaining the GPA->HPA association), but could also allow application to implement their own allocation strategy. I will submit a separate daxctl counterpart shortly, to help visualize its use. [0] https://lore.kernel.org/linux-mm/159457116473.754248.7879464730875147365.st...@dwillia2-desk3.amr.corp.intel.com Joao Joao Martins (4): device-dax: Make align a per-device property device-dax: Add an 'align' attribute dax/hmem: Introduce dax_hmem.region_idle parameter device-dax: Add a range mapping allocation attribute drivers/dax/bus.c | 166 +++++++++++++++++++++++++++++++++++--- drivers/dax/dax-private.h | 1 + drivers/dax/device.c | 35 ++++---- drivers/dax/hmem/hmem.c | 5 +- 4 files changed, 177 insertions(+), 30 deletions(-) -- 2.17.1 _______________________________________________ Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org To unsubscribe send an email to linux-nvdimm-le...@lists.01.org
