Re: [PATCH 3/3] x86/crash: Allocate enough low memory when crashkernel=high

[Baoquan He]

On 06/10/15 at 10:48am, Yinghai Lu wrote:
> On Wed, Jun 10, 2015 at 12:04 AM, Baoquan He <bhe@xxxxxxxxxx> wrote:
> >
> > In fact in this case, it doesn't matter how much memory the system has
> > since most of them is above 4G and only 72M is reserved for dma/swiotlb
> > in kdump kernel. And it doesn't matter much how many pci/pcie cards the
> > system has since it matters how greedy those devices expects dma
> > allocation.
> >
> > Before Joerg posted this patchset v1, our customer complained about
> > this too. That means this is not individual case. I think it makes sense
> > to increase the default low-memory to make kdump succeed, it's
> > definitely better than letting user get a failed kdump and then know
> > they need specify more low-memory manually. If user clearly know their
> > system doesn't need so much low-memory, specifying crashkernel=xx,low
> > works for them.
> 
> Now intel 4socket or 8socket system mostly have [2G,4G) for mmio low.
> also I saw one system had [1.5G, 4G) for mmio.

Really? How does it work supporting so many pcie devices but with
so little DMA32 space? Do they support dma above 4G, or hw iommu is
default? Surely for both tight low-memory is not a worry.

> 
> So under 4G low ram is tight, with changing default to 256M is just lazy
> and punishing the other system that does not need that.
> those systems do not need to have "crashkernel=xx,low" before.

OK. Seems defaulting to whom is controversial. Defaulting to 72M kdump
on some systems will die. Defaulting to 256M some special systems will
be tight on dma/dma32 memory, does their kdump die? If not I vote for
256M. 256M is from Joerg's experimental data, I remember he said his
machine need around 200M, then why not 256M. Otherwise it need be
increased sooner or later.

> 
> should have other smart way to detect and update the default value.
> even dmi blacklist way is better than just changing default value.

I am not sure about this. All systems expecting more low-memory need be
checked out?

> 
> Thanks
> 
> Yinghai
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/