[Crash-utility] Re: [PATCH v6 3/4] xarray: add large folio support

On Thu, May 7, 2026 at 11:40 PM Huang Shijie <huangsj(a)hygon.cn&gt; wrote: > > We may meet a large folio at two cases: > 1.) when height is 1, the real data layout may looks like this: > -------------------------------------------- > crash> p *(struct xa_node*)0xffff889883a6a910 > $5 = { > shift = 0 '\000', > offset = 0 '\000', > count = 64 '@', > nr_values = 0 '\000', > ................. > slots = { 0xffffea0061d21000, 0x2, 0x2, 0x2, > 0xffffea0121d85c00, 0x12, 0x12, 0x12, > 0xffffea0121d9f400, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, > 0xffffea0121583800, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, > 0xffffea0121833e00, 0x62, 0x62, 0x62, 0x62, 0x62, 0x62, 0x62, > 0xffffea0121eb3800, 0x82, 0x82, 0x82, 0x82, 0x82, 0x82, 0x82, > 0x82, 0x82, 0x82, 0x82, 0x82, 0x82, 0x82, > 0x82, 0x82, 0x82, 0x82, 0x82, 0x82, 0x82, > 0x82, 0x82, 0x82, 0x82, 0x82, 0x82, 0x82, > 0x82, 0x82, 0x82 > }, > ................. > } > -------------------------------------------- > The page orders for above folios: > 0xffffea0061d21000 --> order 2 > 0xffffea0121d85c00 --> order 2 > 0xffffea0121d9f400 --> order 3 > 0xffffea0121583800 --> order 3 > 0xffffea0121833e00 --> order 3 > 0xffffea0121eb3800 --> order 5 > > 2.) when height is not 1, the real data layout may looks like this: > -------------------------------------------- > slot: { 0xffff8a2c866b745a, 0xffff8a2c866b16d2, 0xfffff9fcc4ea4000, 0xfffff9fcc4eb9000, ..} > -------------------------------------------- > The 0xffff8a2c866b745a and 0xffff8a2c866b16d2 is a internal node, not a folio. Don't we need to iterate into the internal node in order to find their leaf nodes?

> The page orders for above folios: > 0xfffff9fcc4ea4000 --> order 6 BTW, could you send me a vmcore with the folio data? When reviewing the patch, I can use the vmcore as a reference to better understand this case.