| manjo | kamal__, looks like you backported ARM: fix alignment of keystone page table fixup do you recall what issue this fixed ? | 16:29 |
|---|---|---|
| apw | manjo, from the description i'd say it makes sure the end of the kernel data space is mapped properly, which i would to mean random builds working and randoms ones not | 16:33 |
| manjo | apw, ok I have a similar situation ... so was trying to understand that patch | 16:34 |
| manjo | apw, was looking for any bug report so that I can compare the trace | 16:35 |
| manjo | apw, the bug that it ref to is of no help | 16:35 |
| apw | the bug would be a stable being applied en-toto bug i assume | 16:35 |
| apw | the original is no more verbose, and has no bug associated | 16:35 |
| manjo | bah | 16:36 |
| apw | but as it is random end of data space alignment issues, it woud literally be anything that happens to be linked there which would be missing | 16:36 |
| kamal__ | manjo, yeah, that commit was just "cc: stable", hence I merged it... there is/was no bug reference for it. | 16:36 |
| apw | i doubt it is constrained to any particular thing | 16:36 |
| manjo | apw, so in my case a dkms module does a get_user_pages() but it is unable to pin it because *pmd = 00000000 | 16:36 |
| manjo | kamal__, ack I feard that might be the case | 16:37 |
| === kamal__ is now known as kamal | ||
| manjo | but the code worked in a previous build but no in the current build | 16:37 |
| apw | manjo, have you a crash stack for it? | 16:39 |
| manjo | pmed you the bug | 16:40 |
| apw | that stack could just be saying you asked it to map something from userspace which does not exist | 16:48 |
| manjo | right so it seems like there is nothing wrong kernel wise ... but the user tells me that they did not see failures like this before | 16:49 |
| apw | that tells you nothing other than they used to be luckier, in reality. code which should and indeed seems impossible to have worked ever | 16:50 |
| apw | often works for years before exploding and showing you they are passing the wrong thing | 16:50 |
| manjo | did you see my comment #2 ? that is what I was trying to understand as to why their module is trying to pin invalid page table | 16:51 |
| apw | well that is simply defined by the address you pass in, it is saying "you asked me to pin a page in this range, but there isn't even a pmd for it, so there is no page EFAULT" | 16:52 |
| manjo | I want to completely rule out that any merge from ubuntu to this flavor kernel did not intro any regressions | 16:52 |
| apw | from that trace, i would say the driver reported that the get_user_pages failed, and then went on to use the address anyhow as if it did not | 16:53 |
| manjo | :) yeah it looks messed up ... trying to untangle the spaghetti so that I can put the ball on their court | 16:55 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!