[PATCH 2/5] mm/memremap_pages: Introduce memremap_compat_align()

From: Dan Williams
Date: Thu Jan 30 2020 - 15:22:14 EST


The "sub-section memory hotplug" facility allows memremap_pages() users
like libnvdimm to compensate for hardware platforms like x86 that have a
section size larger than their hardware memory mapping granularity. The
compensation that sub-section support affords is being tolerant of
physical memory resources shifting by units smaller (64MiB on x86) than
the memory-hotplug section size (128 MiB). Where the platform
physical-memory mapping granularity is limited by the number and
capability of address-decode-registers in the memory controller.

While the sub-section support allows memremap_pages() to operate on
sub-section (2MiB) granularity, the Power architecture may still
require 16MiB alignment on "!radix_enabled()" platforms.

In order for libnvdimm to be able to detect and manage this per-arch
limitation, introduce memremap_compat_align() as a common minimum
alignment across all driver-facing memory-mapping interfaces, and let
Power override it to 16MiB in the "!radix_enabled()" case.

The assumption / requirement for 16MiB to be a viable
memremap_compat_align() value is that Power does not have platforms
where its equivalent of address-decode-registers never hardware remaps a
persistent memory resource on smaller than 16MiB boundaries.

Based on an initial patch by Aneesh.

Link: http://lore.kernel.org/r/CAPcyv4gBGNP95APYaBcsocEa50tQj9b5h__83vgngjq3ouGX_Q@xxxxxxxxxxxxxx
Reported-by: Aneesh Kumar K.V <aneesh.kumar@xxxxxxxxxxxxx>
Reported-by: Jeff Moyer <jmoyer@xxxxxxxxxx>
Cc: Benjamin Herrenschmidt <benh@xxxxxxxxxxxxxxxxxxx>
Cc: Paul Mackerras <paulus@xxxxxxxxx>
Cc: Michael Ellerman <mpe@xxxxxxxxxxxxxx>
Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
---
arch/powerpc/include/asm/io.h | 10 ++++++++++
drivers/nvdimm/pfn_devs.c | 2 +-
include/linux/io.h | 23 +++++++++++++++++++++++
include/linux/mmzone.h | 1 +
4 files changed, 35 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/include/asm/io.h b/arch/powerpc/include/asm/io.h
index a63ec938636d..0fa2dc483008 100644
--- a/arch/powerpc/include/asm/io.h
+++ b/arch/powerpc/include/asm/io.h
@@ -734,6 +734,16 @@ extern void __iomem * __ioremap_at(phys_addr_t pa, void *ea,
unsigned long size, pgprot_t prot);
extern void __iounmap_at(void *ea, unsigned long size);

+#ifdef CONFIG_SPARSEMEM
+static inline unsigned long memremap_compat_align(void)
+{
+ if (radix_enabled())
+ return SUBSECTION_SIZE;
+ return (1UL << mmu_psize_defs[mmu_linear_psize].shift);
+}
+#define memremap_compat_align memremap_compat_align
+#endif
+
/*
* When CONFIG_PPC_INDIRECT_PIO is set, we use the generic iomap implementation
* which needs some additional definitions here. They basically allow PIO
diff --git a/drivers/nvdimm/pfn_devs.c b/drivers/nvdimm/pfn_devs.c
index b94f7a7e94b8..a5c25cb87116 100644
--- a/drivers/nvdimm/pfn_devs.c
+++ b/drivers/nvdimm/pfn_devs.c
@@ -750,7 +750,7 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn)
start = nsio->res.start;
size = resource_size(&nsio->res);
npfns = PHYS_PFN(size - SZ_8K);
- align = max(nd_pfn->align, (1UL << SUBSECTION_SHIFT));
+ align = max(nd_pfn->align, SUBSECTION_SIZE);
end_trunc = start + size - ALIGN_DOWN(start + size, align);
if (nd_pfn->mode == PFN_MODE_PMEM) {
/*
diff --git a/include/linux/io.h b/include/linux/io.h
index 35e8d84935e0..ccd34519fad3 100644
--- a/include/linux/io.h
+++ b/include/linux/io.h
@@ -6,6 +6,7 @@
#ifndef _LINUX_IO_H
#define _LINUX_IO_H

+#include <linux/mmzone.h>
#include <linux/types.h>
#include <linux/init.h>
#include <linux/bug.h>
@@ -79,6 +80,28 @@ void *devm_memremap(struct device *dev, resource_size_t offset,
size_t size, unsigned long flags);
void devm_memunmap(struct device *dev, void *addr);

+#ifndef memremap_compat_align
+#ifdef CONFIG_SPARSEMEM
+/*
+ * Minimum compatible alignment of the resource (start, end) across
+ * memremap interfaces (i.e. memremap + memremap_pages)
+ */
+static inline unsigned long memremap_compat_align(void)
+{
+ return SUBSECTION_SIZE;
+}
+#else /* CONFIG_SPARSEMEM */
+/*
+ * No ZONE_DEVICE / memremap_pages() support so the minimum mapping
+ * granularity is a single page.
+ */
+static inline unsigned long memremap_compat_align(void)
+{
+ return PAGE_SIZE;
+}
+#endif /* CONFIG_SPARSEMEM */
+#endif /* memremap_compat_align */
+
#ifdef CONFIG_PCI
/*
* The PCI specifications (Rev 3.0, 3.2.5 "Transaction Ordering and
diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 89d8ff06c9ce..b0de83620cd7 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -1171,6 +1171,7 @@ static inline unsigned long section_nr_to_pfn(unsigned long sec)
#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)

#define SUBSECTION_SHIFT 21
+#define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)

#define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
#define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)