| Summary: | Bitbake complains about mmap_min_addr for an unknown reason | ||
|---|---|---|---|
| Product: | [Build System, Metadata & Runtime] OE-Core | Reporter: | Darren Hart <dvhart> |
| Component: | core | Assignee: | Edwin Zhai <edwin.zhai> |
| Status: | RESOLVED FIXED | QA Contact: | |
| Severity: | normal | ||
| Priority: | Medium | CC: | dexuan.cui, ke.yu, kevin.tian, Martin.Jansa, meta.mr.watcher, meta.watcher, richard.purdie, sgw, tom.rini |
| Version: | 0.9 | ||
| Target Milestone: | 1.0 M4 | ||
| Hardware: | x86 | ||
| OS: | Multiple | ||
| Whiteboard: | |||
| OS type for building Yocto: | --- | Type of Regression: | --- |
| Verified: | Documentation change: | --- | |
|
Description
Darren Hart
2010-11-19 14:33:14 UTC
*** Bug 662 has been marked as a duplicate of this bug. *** Edwin, Please confirm that the mmap_min_addr patch is no longer needed with the latest qemu and if so, then remove this sanity check. (In reply to comment #2) > Edwin, Please confirm that the mmap_min_addr patch is no longer needed with the > latest qemu and if so, then remove this sanity check. Saul, I'm not in the mmap_min_addr context, but I assume if I can build qemuarm minimal image, and run without these checks, then it's okay. Right? So, when I brought this up to Richard before, his answer was (to paraphrase) that he had a recollection of someone saying there were still corner cases and that it wasn't 100% fixed upstream (despite upstream saying it was fixed). (In reply to comment #4) > So, when I brought this up to Richard before, his answer was (to paraphrase) > that he had a recollection of someone saying there were still corner cases and > that it wasn't 100% fixed upstream (despite upstream saying it was fixed). The corner cases are when user account running bitbake (and then qemu) is not allowed to read /proc/sys/vm/mmap_min_addr and adjust mmap_min_addr in qemu accordingly. That's why OE still has http://git.openembedded.org/cgit.cgi/openembedded/commit/?id=93a439a42b538be1f96c9250122504bfbdbd8d7b And that's why this check will fail to read /proc/sys/vm/mmap_min_addr on some kernels and report "maybe false positive", see: http://www.mail-archive.com/openembedded-devel@lists.openembedded.org/msg02340.html And here is response on qemu-devel: http://www.mail-archive.com/qemu-devel@nongnu.org/msg31804.html but c581deda322080e8beb88b2e468d4af54454e4b3 afaik didn't solve it completely (In reply to comment #5) > (In reply to comment #4) > > So, when I brought this up to Richard before, his answer was (to paraphrase) > > that he had a recollection of someone saying there were still corner cases and > > that it wasn't 100% fixed upstream (despite upstream saying it was fixed). > > The corner cases are when user account running bitbake (and then qemu) is not > allowed to read /proc/sys/vm/mmap_min_addr and adjust mmap_min_addr in qemu > accordingly. > > That's why OE still has > http://git.openembedded.org/cgit.cgi/openembedded/commit/?id=93a439a42b538be1f96c9250122504bfbdbd8d7b > > And that's why this check will fail to read /proc/sys/vm/mmap_min_addr on some > kernels and report "maybe false positive", see: > http://www.mail-archive.com/openembedded-devel@lists.openembedded.org/msg02340.html > > And here is response on qemu-devel: > http://www.mail-archive.com/qemu-devel@nongnu.org/msg31804.html > but c581deda322080e8beb88b2e468d4af54454e4b3 afaik didn't solve it completely Martin, Thanks for your reply. It's more clear now. There is in fact 2 issues: 1. qemu linux-user doesn't reload it well, so we need set the mmap_min_addr to 0. And qemu c581deda322080e8beb88b2e468d4af54454e4b3 fix it (already in 0.13.0). 2. Another issue is user on some distribution can't access the mmap_min_addr, so your patch assume 65536(from most of popular distribution) in such case. Is it right? I think issue 1 depends on 2, but not sure if these 2 fixes conflict? So proper solution is adding Martin's patch in case that end user has no access of the /proc/..../mmap_min_addr and removed our checks. If so I'll have a test to see if qemuarm can be built and run well with this solution. Another question is: does our check in sanity.bbclass have such access issue? (In reply to comment #6) > (In reply to comment #5) > > (In reply to comment #4) > > > So, when I brought this up to Richard before, his answer was (to paraphrase) > > > that he had a recollection of someone saying there were still corner cases and > > > that it wasn't 100% fixed upstream (despite upstream saying it was fixed). > > > > The corner cases are when user account running bitbake (and then qemu) is not > > allowed to read /proc/sys/vm/mmap_min_addr and adjust mmap_min_addr in qemu > > accordingly. > > > > That's why OE still has > > http://git.openembedded.org/cgit.cgi/openembedded/commit/?id=93a439a42b538be1f96c9250122504bfbdbd8d7b > > > > And that's why this check will fail to read /proc/sys/vm/mmap_min_addr on some > > kernels and report "maybe false positive", see: > > http://www.mail-archive.com/openembedded-devel@lists.openembedded.org/msg02340.html > > > > And here is response on qemu-devel: > > http://www.mail-archive.com/qemu-devel@nongnu.org/msg31804.html > > but c581deda322080e8beb88b2e468d4af54454e4b3 afaik didn't solve it completely > Martin, > Thanks for your reply. It's more clear now. There is in fact 2 issues: > 1. qemu linux-user doesn't reload it well, so we need set the mmap_min_addr to > 0. And qemu c581deda322080e8beb88b2e468d4af54454e4b3 fix it (already in > 0.13.0). > 2. Another issue is user on some distribution can't access the mmap_min_addr, > so your patch assume 65536(from most of popular distribution) in such case. > Is it right? > I think issue 1 depends on 2, but not sure if these 2 fixes conflict? Now in poky we have qemu 0.13.0, and we also have the patch: http://git.pokylinux.org/cgit/cgit.cgi/poky/commit/?id=ea63409781dc9a5de2d852ec044e55dc82e31a96 "qemu-0.13.0: Add patch to avoid mmap_min_addr * This patch is taken from OE commit 40e293342ca76921904a43b03b635d9219432edf (From OE-Core rev: 11d76595e036f46906859b59dc06094b2e979771) Signed-off-by: Khem Raj <raj.khem@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>" So I think the bug was actually fixed? (In reply to comment #7) > So proper solution is adding Martin's patch in case that end user has no access > of the /proc/..../mmap_min_addr and removed our checks. > If so I'll have a test to see if qemuarm can be built and run well with this > solution. > Another question is: does our check in sanity.bbclass have such access issue? The code in sanity.bbclass is: if data.getVar('TARGET_ARCH', e.data, True) == "arm": # This path is no longer user-readable in modern (very recent) Linux try: if os.path.exists("/proc/sys/vm/mmap_min_addr"): f = file("/proc/sys/vm/mmap_min_addr", "r") if (f.read().strip() != "0"): messages = messages + "/proc/sys/vm/mmap_min_addr is not 0. This will cause problems with qemu so please fix the value (as root).\n\nTo fix this in later reboots, set vm.mmap_min_addr = 0 in /etc/sysctl.conf.\n" f.close() except: pass So I think bitbake can ignore the exception safely. I have tested arm build without Raj's patch, and everything works well. In theory, Raj's patch should not cause any regression. I'll make a new test with Raj's patch, then send pull request if no regression. Fixed in poky master @ 0bc244181fce67725ddd0981481e5611b700cf51 |