[PATCH] arm64: Force NO_BLOCK_MAPPINGS if crashkernel reservation is required

[Catalin Marinas]

On Fri, Nov 20, 2020 at 09:25:22AM +0530, Anshuman Khandual wrote:
> On 11/19/20 11:25 PM, Catalin Marinas wrote:
> > mem_init() currently relies on knowing the boundaries of the crashkernel
> > reservation to map such region with page granularity for later
> > unmapping via set_memory_valid(..., 0). If the crashkernel reservation
> > is deferred, such boundaries are not known when the linear mapping is
> > created. Simply parse the command line for "crashkernel" and, if found,
> > create the linear map with NO_BLOCK_MAPPINGS.
> > 
> > Signed-off-by: Catalin Marinas <catalin.marinas at arm.com>
> > Cc: James Morse <james.morse at arm.com>
> > Cc: Nicolas Saenz Julienne <nsaenzjulienne at suse.de>
> > ---
> > 
> > Following the online (and offline) discussion with James and Nicolas,
> > this aims to avoid issues with moving the reserve_crashkernel() call to
> > after the memory map has been created (the ZONE_DMA patches).
> > 
> > https://lore.kernel.org/r/e60d643e-4879-3fc3-737d-2c145332a6d7@arm.com
> > 
> >  arch/arm64/mm/mmu.c | 37 ++++++++++++++++---------------------
> >  1 file changed, 16 insertions(+), 21 deletions(-)
> > 
> > diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> > index d7fe72ee678a..dd214157a026 100644
> > --- a/arch/arm64/mm/mmu.c
> > +++ b/arch/arm64/mm/mmu.c
> > @@ -469,6 +469,21 @@ void __init mark_linear_text_alias_ro(void)
> >  			    PAGE_KERNEL_RO);
> >  }
> >  
> > +static bool crash_mem_map __initdata;
> > +
> > +static int __init enable_crash_mem_map(char *arg)
> > +{
> > +	/*
> > +	 * Proper parameter parsing is done by reserve_crashkernel(). We only
> > +	 * need to know if the linear map has to avoid block mappings so that
> > +	 * the crashkernel reservations can be unmapped later.
> > +	 */
> > +	crash_mem_map = true;
> > +
> > +	return 0;
> > +}
> > +early_param("crashkernel", enable_crash_mem_map);
> 
> Should not the crash kernel cmdline parameter gets parsed enough, just to
> ensure it is atleast a valid one ? Otherwise an invalid cmdline request
> can prevent the block mapping. In that case, the kernel will neither have
> a crash kernel nor the block mapping.

Does it actually matter? If people pass random strings on the command
line, they should expect some side-effects. That's about the intention
to have a crashkernel. We can't fully validate the crashkernel parameter
until we know the zones layout.

Also note that with defconfig, we don't get block mappings anyway
(rodata_full).

-- 
Catalin