Hi Russell,
On 5/18/21 8:06 PM, Russell King (Oracle) wrote:
[Please note: This e-mail is from an EXTERNAL e-mail address]
On Tue, May 18, 2021 at 12:29:32PM +0100, Russell King (Oracle) wrote:
On Tue, May 18, 2021 at 07:12:54PM +0800, Yanfei Xu wrote:
The value of "vmalloc=" set in cmdline is always 8M more than the value
of "VmallocTotal" in meminfo. When use the "vmalloc=" parameter, user
expect to get the size what they input, and no need to consider the 8M
"hole" hided in codes. This commit make real vmalloc size equal to value
of "vmalloc=" in cmdline.
Also, the commit will reduce the size of vmalloc printed in boot message
by 8M when the size set in cmdline is irrational.
Hi,
I think I'd like to do several cleanups with this:
1. change vmalloc_min to be an unsigned long.
2. exclude VMALLOC_OFFSET from vmalloc_min, moving it into
adjust_lowmem_bounds where vmalloc_min is used.
3. rename vmalloc_min to be vmalloc_start
4. enforce vmalloc_start to be a multiple of 2MiB
5. in early_vmalloc(), calculate vmalloc_max as:
VMALLOC_END - (PAGE_OFFSET + SZ_32M + VMALLOC_OFFSET)
and use that to set the upper bound of vmalloc_reserve (which is
something your patch doesn't do, which I think is a bug.
Thoughts?
I've slightly modified the above idea, and will shortly follow up with
some patches to show the idea a bit better...
Thanks for making it better!
Regards,
Yanfei
--
RMK's Patch system: https://urldefense.com/v3/__https://www.armlinux.org.uk/developer/patches/__;!!AjveYdw8EvQ!Mz6E44Glfsb6CPijsZ50Ofca9g8UAgL2MRQpaxa18ZTAIrVQpH0IcyqGDwHSWphEoo8$ FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!