Re: [PATCH v3] arm64: mm: fix linear mapping mem access performance degradation

[Catalin Marinas <catalin.marinas@arm.com>]

On Thu, Jun 30, 2022 at 06:50:22PM +0800, Guanghui Feng wrote:
> +static void init_pmd_remap(pud_t *pudp, unsigned long addr, unsigned long end,
> +			   phys_addr_t phys, pgprot_t prot,
> +			   phys_addr_t (*pgtable_alloc)(int), int flags)
> +{
> +	unsigned long next;
> +	pmd_t *pmdp;
> +	phys_addr_t map_offset;
> +	pmdval_t pmdval;
> +
> +	pmdp = pmd_offset(pudp, addr);
> +	do {
> +		next = pmd_addr_end(addr, end);
> +
> +		if (!pmd_none(*pmdp) && pmd_sect(*pmdp)) {
> +			phys_addr_t pte_phys = pgtable_alloc(PAGE_SHIFT);
> +			pmd_clear(pmdp);
> +			pmdval = PMD_TYPE_TABLE | PMD_TABLE_UXN;
> +			if (flags & NO_EXEC_MAPPINGS)
> +				pmdval |= PMD_TABLE_PXN;
> +			__pmd_populate(pmdp, pte_phys, pmdval);
> +			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);

This doesn't follow the architecture requirements for "break before
make" when changing live page tables. While it may work, it risks
triggering a TLB conflict abort. The correct sequence normally is:

	pmd_clear();
	flush_tlb_kernel_range();
	__pmd_populate();

However, do we have any guarantees that the kernel doesn't access the
pmd range being unmapped temporarily? The page table itself might live
in one of these sections, so set_pmd() etc. can get a translation fault.

-- 
Catalin