Re: [PATCH kvm-unit-tests 00/11] arm/arm64: mmu fixes and feature

From: David Hildenbrand <david@redhat.com>
To: Andrew Jones <drjones@redhat.com>
Cc: kvm@vger.kernel.org, pbonzini@redhat.com, rkrcmar@redhat.com,
	cdall@linaro.org, lvivier@redhat.com, thuth@redhat.com
Subject: Re: [PATCH kvm-unit-tests 00/11] arm/arm64: mmu fixes and feature
Date: Wed, 17 Jan 2018 10:16:43 +0100	[thread overview]
Message-ID: <f19f4a5b-2097-01bb-8498-ba604cff5acb@redhat.com> (raw)
In-Reply-To: <20180117091202.dwg35eh43t2nluhm@kamzik.brq.redhat.com>

On 17.01.2018 10:12, Andrew Jones wrote:
> On Tue, Jan 16, 2018 at 08:50:27PM +0100, David Hildenbrand wrote:
>> On 16.01.2018 19:53, Andrew Jones wrote:
>>> The fixes are getting tests to run on 32-bit ARM again and also
>>> fixing virt_to_phys(). The feature is to allow a unit test to start
>>> with the MMU disabled. Couple minor cleanups thrown in too.
>>>
>>> I tested x86 and ppc64 over TCG, but not s390 (please-test: David).
>>>
>>> Andrew Jones (11):
>>>   arm/arm64: cleanup alloc.h includes
>>>   arm/arm64: add pgtable to thread_info
>>>   arm/arm64: fix virt_to_phys
>>>   arm/arm64: flush page table cache when installing entries
>>>   arm/arm64: setup: don't allow gaps in phys range
>>>   phys_alloc: ensure we account all allocations
>>>   page_alloc: allow initialization before setup_vm call
>>>   page_alloc: add yet another memalign
>>>   lib/auxinfo: add flags field
>>>   arm/arm64: allow setup_vm to be skipped
>>>   arm/arm64: sieve should start with the mmu off
>>>
>>>  arm/Makefile.common       |  5 ++++-
>>>  arm/cstart.S              | 19 ++++++++++++++++++-
>>>  arm/cstart64.S            | 17 ++++++++++++++++-
>>>  lib/alloc_page.c          | 33 +++++++++++++++++++++++++++++++++
>>>  lib/alloc_page.h          |  2 ++
>>>  lib/alloc_phys.c          | 10 ++++++++++
>>>  lib/alloc_phys.h          |  6 ++++--
>>>  lib/arm/asm/mmu.h         |  5 +++++
>>>  lib/arm/asm/page.h        |  8 +++-----
>>>  lib/arm/asm/pgtable.h     | 16 ++++++++++++----
>>>  lib/arm/asm/setup.h       |  1 -
>>>  lib/arm/asm/thread_info.h |  1 +
>>>  lib/arm/mmu.c             | 43 ++++++++++++++++++++++++++++++++++++++-----
>>>  lib/arm/processor.c       |  1 +
>>>  lib/arm/setup.c           | 22 ++++++++++++++--------
>>>  lib/arm/smp.c             |  7 ++++++-
>>>  lib/arm64/asm/mmu.h       |  5 +++++
>>>  lib/arm64/asm/page.h      | 10 +++-------
>>>  lib/arm64/asm/pgtable.h   | 12 ++++++++++--
>>>  lib/arm64/processor.c     |  1 +
>>>  lib/arm64/spinlock.c      |  1 +
>>>  lib/auxinfo.c             | 13 ++++++++++++-
>>>  lib/auxinfo.h             |  6 ++++++
>>>  lib/vmalloc.c             |  9 ++++++---
>>>  24 files changed, 211 insertions(+), 42 deletions(-)
>>>
>>
>> lib/libcflat.a(alloc_page.o): In function `page_memalign':
>> /home/dhildenb/git/kvm-unit-tests/lib/alloc_page.c:146: undefined
>> reference to `__clzdi2'
>> collect2: error: ld returned 1 exit status
>> make: *** [/home/dhildenb/git/kvm-unit-tests/s390x/Makefile:44:
>> s390x/selftest.elf] Error 1
>>
> 
> Eh, I guess the s390 compiler doesn't have that builtin. I should
> properly add an fls() to asm/bitops.h for each arch then. s390 can
> use a generic implementation, like the kernel's generic binary search,
> __fls.

I wonder if that is a compiler bug. Will try with an older (!cross)
compiler.

> 
> Thanks,
> drew
> 

-- 

Thanks,

David / dhildenb