[PATCH 05/13] powerpc/xive: Fix allocation of pages donated to the XIVE controller

[Cédric Le Goater]

The XIVE interrupt controller uses a set of internal tables to handle
interrupt routing. The small tables (SBE, EAT) are direct tables and
allocated by OPAL. The bigger ones (NVT, ENDT) are indirect, i.e., the
first table entries point to a page which contains the XIVE structures
used by HW. For these, OPAL only allocates the first page level and
requests pages to be allocated by Linux when a new entry is inserted.

Make sure that these pages can not be reclaimed. The problem can be
seen while stressing a system with 4K KVM guests, 16 vCPUs each.

Signed-off-by: Cédric Le Goater <clg at kaod.org>
---
 arch/powerpc/sysdev/xive/xive-internal.h | 2 ++
 arch/powerpc/sysdev/xive/common.c        | 2 +-
 arch/powerpc/sysdev/xive/native.c        | 2 +-
 3 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/arch/powerpc/sysdev/xive/xive-internal.h b/arch/powerpc/sysdev/xive/xive-internal.h
index d701af7fb48c..1eacc90f4dcf 100644
--- a/arch/powerpc/sysdev/xive/xive-internal.h
+++ b/arch/powerpc/sysdev/xive/xive-internal.h
@@ -73,6 +73,8 @@ static inline u32 xive_alloc_order(u32 queue_shift)
 	return (queue_shift > PAGE_SHIFT) ? (queue_shift - PAGE_SHIFT) : 0;
 }
 
+#define XIVE_GFP (__GFP_NORETRY | __GFP_NOWARN | __GFP_NOMEMALLOC)
+
 extern bool xive_cmdline_disabled;
 
 #endif /*  __XIVE_INTERNAL_H */
diff --git a/arch/powerpc/sysdev/xive/common.c b/arch/powerpc/sysdev/xive/common.c
index 605238ca65e4..80fd97d764ab 100644
--- a/arch/powerpc/sysdev/xive/common.c
+++ b/arch/powerpc/sysdev/xive/common.c
@@ -1544,7 +1544,7 @@ __be32 *xive_queue_page_alloc(unsigned int cpu, u32 queue_shift)
 	__be32 *qpage;
 
 	alloc_order = xive_alloc_order(queue_shift);
-	pages = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, alloc_order);
+	pages = alloc_pages_node(cpu_to_node(cpu), XIVE_GFP, alloc_order);
 	if (!pages)
 		return ERR_PTR(-ENOMEM);
 	qpage = (__be32 *)page_address(pages);
diff --git a/arch/powerpc/sysdev/xive/native.c b/arch/powerpc/sysdev/xive/native.c
index cb58ec7ce77a..6afb44d0d816 100644
--- a/arch/powerpc/sysdev/xive/native.c
+++ b/arch/powerpc/sysdev/xive/native.c
@@ -643,7 +643,7 @@ static bool xive_native_provision_pages(void)
 		 * XXX TODO: Try to make the allocation local to the node where
 		 * the chip resides.
 		 */
-		p = kmem_cache_alloc(xive_provision_cache, GFP_KERNEL);
+		p = kmem_cache_alloc(xive_provision_cache, XIVE_GFP);
 		if (!p) {
 			pr_err("Failed to allocate provisioning page\n");
 			return false;
-- 
2.26.2