Re: [PATCH v3 19/21] memblock: make HugeTLB bootmem allocation work with KHO

> Gigantic huge page allocation is somewhat broken currently when KHO is
> used.
> 
> Firstly, they break KHO scratch size accounting. RSRV_KERN is used to
> track how much memory is reserved for use by the kernel. Since
> hugetlb::alloc_bootmem() calls the memblock_alloc*() APIs, the hugepages
> allocated also get marked as RSRV_KERN.
> 
> Allocations marked RSRV_KERN are used by KHO to calculate how much
> scratch space it should reserve to make sure the next kernel has enough
> memory to boot when it is in scratch-only phase. Counting hugepages in
> that blows up scratch size, and can lead to the scratch allocation
> failing, making KHO unusable. This will show up when huge pages make up
> more than 50% of the system, which is a fairly common use case.
> 
> Secondly, while not supported right now, huge pages are user memory and
> can be preserved via KHO. The scratch spaces should not have any
> preserved memory. Allocating hugepages from scratch (on a KHO boot) can
> lead to them being un-preservable.
> 
> Introduce memblock_alloc_hugetlb(). This lets memblock tailor to the
> needs of hugetb without exposing those details to the general allocation
> routines.
> 
> First, it does not use mirrored memory for hugetlb. Mirrored memory is a
> limited resource that is best saved for kernel data structures, not user
> memory.
> 
> Second, if the free memory area found by memblock_find_in_range_node()
> is a part of a KHO scratch area, the free area is not used. Allocation
> is retried starting after the free area to ensure no hugepages come from
> KHO scratch.
> 
> Third, it simplifies the argument list by baking in some hugetlb
> assumptions like alignment and exact_nid. This also simplifies
> allocation logic in alloc_bootmem().
> 
> Also introduce MEMBLOCK_RSRV_HUGETLB to mark reservations made for
> HugeTLB. This will be used by KHO in future patches to correctly
> calculate scratch sizes.
> 
> Refactor some of the preparation logic like kmemleak tracking and
> accepting memory into a separate helper memblock_prep_allocation(), and
> use it from both memblock_alloc_hugetlb() and the usual
> memblock_alloc_range_nid().
> 
> Add a stub for kho_scratch_overlap to memblock tests to make sure it
> compiles.
> 
> Signed-off-by: Pratyush Yadav (Google) <[email protected]>
>
> diff --git a/include/linux/memblock.h b/include/linux/memblock.h
> index f142c8a8cd5b7..678fe466529a2 100644
> --- a/include/linux/memblock.h
> +++ b/include/linux/memblock.h
> @@ -51,6 +51,7 @@ extern unsigned long long max_possible_pfn;
>   * memory reservations yet, so we get scratch memory from the previous
>   * kernel that we know is good to use. It is the only memory that
>   * allocations may happen from in this phase.
> + * @MEMBLOCK_RSRV_HUGETLB: memory is reserved for hugetlb pages
>   */
>  enum memblock_flags {
>  	MEMBLOCK_NONE		= 0x0,	/* No special request */
> @@ -61,6 +62,7 @@ enum memblock_flags {
>  	MEMBLOCK_RSRV_NOINIT	= 0x10,	/* don't initialize struct pages */
>  	MEMBLOCK_RSRV_KERN	= 0x20,	/* memory reserved for kernel use */
>  	MEMBLOCK_KHO_SCRATCH	= 0x40,	/* scratch memory for kexec handover */
> +	MEMBLOCK_RSRV_HUGETLB	= 0x80, /* memory reserved for hugetlb pages */
>  };
>  
>  /**
> @@ -420,6 +422,7 @@ void *memblock_alloc_try_nid_raw(phys_addr_t size, phys_addr_t align,
>  void *memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align,
>  			     phys_addr_t min_addr, phys_addr_t max_addr,
>  			     int nid);
> +void *memblock_alloc_hugetlb(phys_addr_t size, int nid, bool exact_nid);
>  
>  static __always_inline void *memblock_alloc(phys_addr_t size, phys_addr_t align)
>  {
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 571212b80835e..ab4afc818e8cb 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -3033,29 +3033,21 @@ static __init void *alloc_bootmem(struct hstate *h, int nid, bool node_exact)
>  	if (hugetlb_early_cma(h))
>  		m = hugetlb_cma_alloc_bootmem(h, &listnode, node_exact);
>  	else {
> -		if (node_exact)
> -			m = memblock_alloc_exact_nid_raw(huge_page_size(h),
> -				huge_page_size(h), 0,
> -				MEMBLOCK_ALLOC_ACCESSIBLE, nid);
> -		else {
> -			m = memblock_alloc_try_nid_raw(huge_page_size(h),
> -				huge_page_size(h), 0,
> -				MEMBLOCK_ALLOC_ACCESSIBLE, nid);
> +		m = memblock_alloc_hugetlb(huge_page_size(h), nid, node_exact);
> +		if (m) {
> +			m->flags = 0;
> +			m->cma = NULL;
> +
>  			/*
>  			 * For pre-HVO to work correctly, pages need to be on
>  			 * the list for the node they were actually allocated
>  			 * from. That node may be different in the case of
> -			 * fallback by memblock_alloc_try_nid_raw. So,
> +			 * fallback by memblock_alloc_hugetlb_bootmem. So,
>  			 * extract the actual node first.
>  			 */
> -			if (m)
> +			if (!node_exact)
>  				listnode = early_pfn_to_nid(PHYS_PFN(__pa(m)));
>  		}
> -
> -		if (m) {
> -			m->flags = 0;
> -			m->cma = NULL;
> -		}
>  	}
>  
>  	if (m) {
> diff --git a/mm/memblock.c b/mm/memblock.c
> index 8b2e551435ae9..ba6d887ea18eb 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -1506,6 +1506,32 @@ int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
>  	return 0;
>  }
>  
> +static void memblock_prep_allocation(phys_addr_t start, phys_addr_t size,
> +				     bool leaktrace)

Let's call this kmemleak_trace

> +{
> +	/*
> +	 * Skip kmemleak for those places like kasan_init() and
> +	 * early_pgtable_alloc() due to high volume.
> +	 */
> +	if (leaktrace)
> +		/*
> +		 * Memblock allocated blocks are never reported as
> +		 * leaks. This is because many of these blocks are
> +		 * only referred via the physical address which is
> +		 * not looked up by kmemleak.
> +		 */
> +		kmemleak_alloc_phys(start, size, 0);
> +
> +	/*
> +	 * Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP,
> +	 * require memory to be accepted before it can be used by the
> +	 * guest.
> +	 *
> +	 * Accept the memory of the allocated buffer.
> +	 */
> +	accept_memory(start, size);
> +}
> +
>  /**
>   * memblock_alloc_range_nid - allocate boot memory block
>   * @size: size of memory block to be allocated in bytes
> @@ -1580,28 +1606,7 @@ phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
>  	return 0;
>  
>  done:
> -	/*
> -	 * Skip kmemleak for those places like kasan_init() and
> -	 * early_pgtable_alloc() due to high volume.
> -	 */
> -	if (end != MEMBLOCK_ALLOC_NOLEAKTRACE)
> -		/*
> -		 * Memblock allocated blocks are never reported as
> -		 * leaks. This is because many of these blocks are
> -		 * only referred via the physical address which is
> -		 * not looked up by kmemleak.
> -		 */
> -		kmemleak_alloc_phys(found, size, 0);
> -
> -	/*
> -	 * Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP,
> -	 * require memory to be accepted before it can be used by the
> -	 * guest.
> -	 *
> -	 * Accept the memory of the allocated buffer.
> -	 */
> -	accept_memory(found, size);
> -
> +	memblock_prep_allocation(found, size, end != MEMBLOCK_ALLOC_NOLEAKTRACE);
>  	return found;
>  }
>  
> @@ -1756,6 +1761,69 @@ void * __init memblock_alloc_try_nid_raw(
>  				       false);
>  }
>  
> +/**
> + * memblock_alloc_hugetlb - allocate boot memory for HugeTLB pages
> + * @size:      size of the memory to be allocated in bytes
> + * @nid:       nid of the free memory to find, %NUMA_NO_NODE for any node
> + * @exact_nid: only allocate from the specified nid. If %false, the specified
> + *             nid is tried first, and then all nodes are tried as fallback.
> + *
> + * HugeTLB pages are always aligned by their size, so the alignment matches
> + * @size. Since the memory is for userspace, mirrored memory is not used. The
> + * memory is not zeroed. Does not panic if request cannot be satisfied.
> + *
> + * Return:
> + * Virtual address of allocated memory block on success, %NULL on failure.
> + */
> +void * __init memblock_alloc_hugetlb(phys_addr_t size, int nid, bool exact_nid)
> +{
> +	enum memblock_flags flags = choose_memblock_flags();
> +	phys_addr_t addr, start = 0, end = MEMBLOCK_ALLOC_ACCESSIBLE;
> +
> +	memblock_dbg("%s: %llu bytes, nid=%d, exact_nid=%d %pS\n", __func__,
> +		     (u64)size, nid, exact_nid, (void *)_RET_IP_);
> +
> +	/* Don't waste mirrored memory on HugeTLB pages. */
> +	flags &= ~MEMBLOCK_MIRROR;
> +retry:
> +	/* HugeTLB pages are always aligned by their size. */
> +	addr = memblock_find_in_range_node(size, size, start, end, nid, flags);
> +	if (addr)
> +		goto found;
> +
> +	/* Try all nodes if allowed. */
> +	if (numa_valid_node(nid) && !exact_nid) {
> +		nid = NUMA_NO_NODE;
> +		goto retry;
> +	}
> +
> +	/* Found nothing... :-( */
> +	return NULL;
> +
> +found:
> +	/*
> +	 * HugeTLB pages can be preserved with KHO and no preserved memory can
> +	 * be in scratch. So retry if found address overlaps with scratch.
> +	 *
> +	 * Scratch areas are normally not very large, so this shouldn't take too
> +	 * many retries.
> +	 */
> +	if (kho_scratch_overlap(addr, size)) {
> +		if (memblock_bottom_up())
> +			start = addr + size;
> +		else
> +			start = addr - size;
> +
> +		goto retry;
> +	}
> +
> +	if (__memblock_reserve(addr, size, nid, MEMBLOCK_RSRV_KERN | MEMBLOCK_RSRV_HUGETLB))
> +		return NULL;
> +
> +	memblock_prep_allocation(addr, size, true);
> +	return phys_to_virt(addr);
> +}
> +
>  /**
>   * memblock_alloc_try_nid - allocate boot memory block
>   * @size: size of memory block to be allocated in bytes
> diff --git a/tools/testing/memblock/internal.h b/tools/testing/memblock/internal.h
> index b6b1d147fd750..e86bb9000b22f 100644
> --- a/tools/testing/memblock/internal.h
> +++ b/tools/testing/memblock/internal.h
> @@ -66,4 +66,9 @@ static inline void init_deferred_page(unsigned long pfn, int nid)
>  
>  #define __SetPageReserved(p)	((void)(p))
>  
> +static inline bool kho_scratch_overlap(phys_addr_t phys, size_t size)
> +{
> +	return false;
> +}
> +

This should be in the patch that exposes kho_scratch_overlap() to kexec_handover.h

-- 
Sincerely yours,
Mike.