Re: [PATCH] arm64, vmcoreinfo : Append 'MAX_USER_VA_BITS' and 'MAX_PHYSMEM_BITS' to vmcoreinfo

[James Morse <james.morse@arm.com>]

Hi Bhupesh,

On 04/02/2019 14:35, Bhupesh Sharma wrote:
> On 01/31/2019 03:09 AM, Bhupesh Sharma wrote:
>> On 01/30/2019 08:51 PM, James Morse wrote:
>>> On 01/30/2019 12:23 PM, Bhupesh Sharma wrote:
>>>> With ARMv8.2-LVA and LPA architecture extensions, arm64 hardware which
>>>> supports these extensions can support upto 52-bit virtual and 52-bit
>>>> physical addresses respectively.
>>>>
>>>> Since at the moment we enable the support of these extensions via CONFIG
>>>> flags, e.g.
>>>>   - LPA via CONFIG_ARM64_PA_BITS_52
>>>>
>>>> there are no clear mechanisms in user-space right now to
>>>> deteremine these CONFIG flag values and also determine the PARange and
>>>> VARange address values.
>>>> User-space tools like 'makedumpfile' and 'crash-utility' can instead
>>>> use the 'MAX_USER_VA_BITS' and 'MAX_PHYSMEM_BITS' values to determine
>>>> the maximum virtual address and physical address (respectively)
>>>> supported by underlying kernel.
>>>>
>>>> A reference 'makedumpfile' implementation which uses this approach to
>>>> determining the maximum physical address is available in [0].
>>>
>>> Why does it need to know?
>>>
>>> (Suzuki asked the same question on your earlier version)
>>> https://lore.kernel.org/linux-arm-kernel/cff44754-7fe4-efea-bc8e-4dde2277c821@arm.com/


>> I have shared some details (after discussion with our test teams) in reply to
>> the review comments from Suzuki here:
>> http://lists.infradead.org/pipermail/kexec/2019-January/022389.html, and
>> http://lists.infradead.org/pipermail/kexec/2019-January/022390.html
>>
>> Just to summarize, I mentioned in my replies to the review comments tha the
>> makedumpfile implementation (for decoding the PTE) was just as an example,
>> however there can be other user-space applications, for e.g a user-space
>> application running with 48-bit kernel VA and 52-bit user space VA and
>> requesting allocation in 'high' address via a 'hint' to mmap.

But vmcoreinfo is the wrong place to expose this information. (it can be
configured off, and is only accessible to root)


>>>  From your github link it looks like you use this to re-assemble the two bits
>>> of the PFN from the pte. Can't you always do this for 64K pages? CPUs with
>>> the feature always do this too, its not something the kernel turns on.
>>
>> Ok, let me try to give some perspective of a common makedumpfile use-case
>> before I jump into the details:


>> Also hardcoding the PTE calculation to use the high address bit mask always
>> will break the backward compatibility with older kernels (which don't support
>> 52-bit address space extensions).

What would go wrong?

The hardware ignores those bits and supplies zero. As far as I can see the
kernel has always generated zero there.


>> (b). Also x86_64 already has a vmcoreinfo export for 'pgtable_l5_enabled':

So? 5-level page tables is a different feature. I agree you need to know the
number of levels to walk the page-tables, but that isn't how arm64's 52bit stuff
works.


> Ping. Since this patch fixes a regression with user-space tools like
> makedumpfile and crash-utility which are broken since arm64 kernels with 52-bit
> VA and PA support are available (and distributions which enable them), would
> request review comments/ack on this simple change.

Broken how? What goes wrong?

I can see how a not-52bit-aware crash/gdb/whatever would be confused by high
bits being set in the physical address, and possibly throw them away.
But once it supports this for 64K pages, I don't see what can go wrong if those
bits aren't set. Why does it need to know?


Thanks,

James

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel