[PATCH v3 6/6] x86/mm/KASLR: Do not adapt the size of the direct mapping section for SGI UV system

[Baoquan He]

On SGI UV system, kernel often hangs when KASLR is enabled. Disabling
KASLR makes kernel work well.

The back trace is:

kernel BUG at arch/x86/mm/init_64.c:311!
invalid opcode: 0000 [#1] SMP
[...]
RIP: 0010:__init_extra_mapping+0x188/0x196
[...]
Call Trace:
 init_extra_mapping_uc+0x13/0x15
 map_high+0x67/0x75
 map_mmioh_high_uv3+0x20a/0x219
 uv_system_init_hub+0x12d9/0x1496
 uv_system_init+0x27/0x29
 native_smp_prepare_cpus+0x28d/0x2d8
 kernel_init_freeable+0xdd/0x253
 ? rest_init+0x80/0x80
 kernel_init+0xe/0x110
 ret_from_fork+0x2c/0x40

This is because the SGI UV system need map its MMIOH region to the direct
mapping section, and the mapping happens in rest_init() which is much
later than the calling of kernel_randomize_memory() to do mm KASLR. So
mm KASLR can't count in the size of the MMIOH region when calculate the
needed size of address space for the direct mapping section.

When KASLR is disabled, there are 64TB address space for both system RAM
and the MMIOH regions to share. When KASLR is enabled, the current code
of mm KASLR only reserves the actual size of system RAM plus extra 10TB
for the direct mapping. Thus later the MMIOH mapping could go beyond
the upper bound of the direct mapping to step into VMALLOC or VMEMMAP area.
Then BUG_ON() in __init_extra_mapping() will be triggered.

E.g on the SGI UV3 machine where this bug was reported , there are two
MMIOH regions:

[    1.519001] UV: Map MMIOH0_HI 0xffc00000000 - 0x100000000000
[    1.523001] UV: Map MMIOH1_HI 0x100000000000 - 0x200000000000

They are [16TB-16G, 16TB) and [16TB, 32TB). On this machine, 512G RAM are
spread out to 1TB regions. Then above two SGI MMIOH regions also will be
mapped into the direct mapping section.

To fix it, we need check if it's SGI UV system by calling
is_early_uv_system() in kernel_randomize_memory(). If yes, do not adapt
thesize of the direct mapping section, just keep it as is, e.g in level-4
paging mode, 64TB.

Signed-off-by: Baoquan He <bhe@xxxxxxxxxx>
---
 arch/x86/mm/kaslr.c | 57 +++++++++++++++++++++++++++++++++------------
 1 file changed, 42 insertions(+), 15 deletions(-)

diff --git a/arch/x86/mm/kaslr.c b/arch/x86/mm/kaslr.c
index ca12ed4e5239..754b5da91d43 100644
--- a/arch/x86/mm/kaslr.c
+++ b/arch/x86/mm/kaslr.c
@@ -29,6 +29,7 @@
 #include <asm/pgtable.h>
 #include <asm/setup.h>
 #include <asm/kaslr.h>
+#include <asm/uv/uv.h>
 
 #include "mm_internal.h"
 
@@ -113,15 +114,51 @@ static inline bool kaslr_memory_enabled(void)
 	return kaslr_enabled() && !IS_ENABLED(CONFIG_KASAN);
 }
 
+/*
+ * Even though a huge virtual address space is reserved for the direct
+ * mapping of physical memory, e.g in 4-level pageing mode, it's 64TB,
+ * rare system can own enough physical memory to use it up, most are
+ * even less than 1TB. So with KASLR enabled, we adapt the size of
+ * direct mapping area to size of actual physical memory plus the
+ * configured padding CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING.
+ * The left part will be taken out to join memory randomization.
+ *
+ * Note that UV system is an exception, its MMIOH region need be mapped
+ * into the direct mapping area too, while the size can't be got until
+ * rest_init() calling. Hence for UV system, do not adapt the size
+ * of direct mapping area.
+ */
+static inline unsigned long calc_direct_mapping_size(void)
+{
+	unsigned long size_tb, memory_tb;
+
+	/*
+	 * Update Physical memory mapping to available and
+	 * add padding if needed (especially for memory hotplug support).
+	 */
+	memory_tb = DIV_ROUND_UP(max_pfn << PAGE_SHIFT, 1UL << TB_SHIFT) +
+		CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING;
+
+	size_tb = 1 << (MAX_PHYSMEM_BITS - TB_SHIFT);
+
+	/*
+	 * Adapt phyiscal memory region size based on available memory if
+	 * it's not UV system.
+	 */
+	if (memory_tb < size_tb && !is_early_uv_system())
+		size_tb = memory_tb;
+
+	return size_tb;
+}
+
 /* Initialize base and padding for each memory region randomized with KASLR */
 void __init kernel_randomize_memory(void)
 {
-	size_t i;
-	unsigned long vaddr_start, vaddr;
-	unsigned long rand, memory_tb;
-	struct rnd_state rand_state;
+	unsigned long vaddr_start, vaddr, rand;
 	unsigned long remain_entropy;
 	unsigned long vmemmap_size;
+	struct rnd_state rand_state;
+	size_t i;
 
 	vaddr_start = pgtable_l5_enabled() ? __PAGE_OFFSET_BASE_L5 : __PAGE_OFFSET_BASE_L4;
 	vaddr = vaddr_start;
@@ -138,20 +175,10 @@ void __init kernel_randomize_memory(void)
 	if (!kaslr_memory_enabled())
 		return;
 
-	kaslr_regions[0].size_tb = 1 << (MAX_PHYSMEM_BITS - TB_SHIFT);
+	kaslr_regions[0].size_tb = calc_direct_mapping_size();
 	kaslr_regions[1].size_tb = VMALLOC_SIZE_TB;
 
-	/*
-	 * Update Physical memory mapping to available and
-	 * add padding if needed (especially for memory hotplug support).
-	 */
 	BUG_ON(kaslr_regions[0].base != &page_offset_base);
-	memory_tb = DIV_ROUND_UP(max_pfn << PAGE_SHIFT, 1UL << TB_SHIFT) +
-		CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING;
-
-	/* Adapt phyiscal memory region size based on available memory */
-	if (memory_tb < kaslr_regions[0].size_tb)
-		kaslr_regions[0].size_tb = memory_tb;
 
 	/*
 	 * Calculate how many TB vmemmap region needs, and align to
-- 
2.17.2