arch/riscv/include/asm/kvm_gstage.h | 3 + arch/riscv/kvm/gstage.c | 100 ++++++++++++++++++++++++++++ arch/riscv/kvm/mmu.c | 12 +++- 3 files changed, 114 insertions(+), 1 deletion(-)
From: Jessica Liu <liu.xuemei1@zte.com.cn>
Use block mapping if backed by a THP, as implemented in architectures
like ARM and x86_64.
Signed-off-by: Jessica Liu <liu.xuemei1@zte.com.cn>
---
arch/riscv/include/asm/kvm_gstage.h | 3 +
arch/riscv/kvm/gstage.c | 100 ++++++++++++++++++++++++++++
arch/riscv/kvm/mmu.c | 12 +++-
3 files changed, 114 insertions(+), 1 deletion(-)
diff --git a/arch/riscv/include/asm/kvm_gstage.h b/arch/riscv/include/asm/kvm_gstage.h
index 595e2183173e..cc67fb2d2d42 100644
--- a/arch/riscv/include/asm/kvm_gstage.h
+++ b/arch/riscv/include/asm/kvm_gstage.h
@@ -69,4 +69,7 @@ void kvm_riscv_gstage_wp_range(struct kvm_gstage *gstage, gpa_t start, gpa_t end
void kvm_riscv_gstage_mode_detect(void);
+long kvm_riscv_gstage_thp_adjust(struct kvm *kvm, struct kvm_memory_slot *memslot,
+ unsigned long hva, kvm_pfn_t *pfnp, gpa_t *gpa);
+
#endif
diff --git a/arch/riscv/kvm/gstage.c b/arch/riscv/kvm/gstage.c
index 24c270d6d0e2..98494b4b4652 100644
--- a/arch/riscv/kvm/gstage.c
+++ b/arch/riscv/kvm/gstage.c
@@ -77,6 +77,106 @@ static int gstage_level_to_page_size(u32 level, unsigned long *out_pgsize)
return 0;
}
+static int gstage_get_user_mapping_size(struct kvm *kvm, u64 addr)
+{
+ pte_t *ptepp;
+ u32 ptep_level;
+ unsigned long out_pgsize;
+ struct kvm_gstage gstage = {
+ .pgd = kvm->mm->pgd
+ };
+
+ if (!kvm_riscv_gstage_get_leaf(&gstage, addr, &ptepp, &ptep_level))
+ return -EFAULT;
+
+ if (gstage_level_to_page_size(ptep_level, &out_pgsize))
+ return -EFAULT;
+
+ return out_pgsize;
+}
+
+static bool gstage_supports_huge_mapping(struct kvm_memory_slot *memslot, unsigned long hva)
+{
+ gpa_t gpa_start;
+ hva_t uaddr_start, uaddr_end;
+ size_t size;
+
+ size = memslot->npages * PAGE_SIZE;
+ uaddr_start = memslot->userspace_addr;
+ uaddr_end = uaddr_start + size;
+
+ gpa_start = memslot->base_gfn << PAGE_SIZE;
+
+ /*
+ * Pages belonging to memslots that don't have the same alignment
+ * within a PMD for userspace and GPA cannot be mapped with g-stage
+ * PMD entries, because we'll end up mapping the wrong pages.
+ *
+ * Consider a layout like the following:
+ *
+ * memslot->userspace_addr:
+ * +-----+--------------------+--------------------+---+
+ * |abcde|fgh vs-stage block | vs-stage block tv|xyz|
+ * +-----+--------------------+--------------------+---+
+ *
+ * memslot->base_gfn << PAGE_SHIFT:
+ * +---+--------------------+--------------------+-----+
+ * |abc|def g-stage block | g-stage block |tvxyz|
+ * +---+--------------------+--------------------+-----+
+ *
+ * If we create those g-stage blocks, we'll end up with this incorrect
+ * mapping:
+ * d -> f
+ * e -> g
+ * f -> h
+ */
+ if ((gpa_start & (PMD_SIZE - 1)) != (uaddr_start & (PMD_SIZE - 1)))
+ return false;
+
+ /*
+ * Next, let's make sure we're not trying to map anything not covered
+ * by the memslot. This means we have to prohibit block size mappings
+ * for the beginning and end of a non-block aligned and non-block sized
+ * memory slot (illustrated by the head and tail parts of the
+ * userspace view above containing pages 'abcde' and 'xyz',
+ * respectively).
+ *
+ * Note that it doesn't matter if we do the check using the
+ * userspace_addr or the base_gfn, as both are equally aligned (per
+ * the check above) and equally sized.
+ */
+ return (hva >= ALIGN(uaddr_start, PMD_SIZE)) && (hva < ALIGN_DOWN(uaddr_end, PMD_SIZE));
+}
+
+long kvm_riscv_gstage_thp_adjust(struct kvm *kvm, struct kvm_memory_slot *memslot,
+ unsigned long hva, kvm_pfn_t *hfnp, gpa_t *gpa)
+{
+ kvm_pfn_t hfn = *hfnp;
+
+ /*
+ * Make sure the adjustment is done only for THP pages. Also make
+ * sure that the HVA and GPA are sufficiently aligned and that the
+ * block map is contained within the memslot.
+ */
+ if (gstage_supports_huge_mapping(memslot, hva)) {
+ int sz = gstage_get_user_mapping_size(kvm, hva);
+
+ if (sz < 0)
+ return sz;
+
+ if (sz < PMD_SIZE)
+ return PAGE_SIZE;
+
+ *gpa &= PMD_MASK;
+ hfn &= ~(PTRS_PER_PMD - 1);
+ *hfnp = hfn;
+
+ return PMD_SIZE;
+ }
+
+ return PAGE_SIZE;
+}
+
bool kvm_riscv_gstage_get_leaf(struct kvm_gstage *gstage, gpa_t addr,
pte_t **ptepp, u32 *ptep_level)
{
diff --git a/arch/riscv/kvm/mmu.c b/arch/riscv/kvm/mmu.c
index 525fb5a330c0..f70cf721ebb8 100644
--- a/arch/riscv/kvm/mmu.c
+++ b/arch/riscv/kvm/mmu.c
@@ -337,7 +337,8 @@ int kvm_riscv_mmu_map(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
struct kvm_mmu_memory_cache *pcache = &vcpu->arch.mmu_page_cache;
bool logging = (memslot->dirty_bitmap &&
!(memslot->flags & KVM_MEM_READONLY)) ? true : false;
- unsigned long vma_pagesize, mmu_seq;
+ unsigned long mmu_seq;
+ long vma_pagesize;
struct kvm_gstage gstage;
struct page *page;
@@ -416,6 +417,15 @@ int kvm_riscv_mmu_map(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
if (mmu_invalidate_retry(kvm, mmu_seq))
goto out_unlock;
+ /* check if we are backed by a THP and thus use block mapping if possible */
+ if (vma_pagesize == PAGE_SIZE) {
+ vma_pagesize = kvm_riscv_gstage_thp_adjust(kvm, memslot, hva, &hfn, &gpa);
+ if (vma_pagesize < 0) {
+ ret = vma_pagesize;
+ goto out_unlock;
+ }
+ }
+
if (writable) {
mark_page_dirty_in_slot(kvm, memslot, gfn);
ret = kvm_riscv_gstage_map_page(&gstage, pcache, gpa, hfn << PAGE_SHIFT,
--
2.27.0
On 9/28/2025 1:14 PM, liu.xuemei1@zte.com.cn wrote:
> +static bool gstage_supports_huge_mapping(struct kvm_memory_slot *memslot, unsigned long hva)
> +{
> + gpa_t gpa_start;
> + hva_t uaddr_start, uaddr_end;
> + size_t size;
> +
> + size = memslot->npages * PAGE_SIZE;
> + uaddr_start = memslot->userspace_addr;
> + uaddr_end = uaddr_start + size;
> +
> + gpa_start = memslot->base_gfn << PAGE_SIZE;
looks wrong why << PAGE_SIZE ? typo
> +
> + /*
> + * Pages belonging to memslots that don't have the same alignment
> + * within a PMD for userspace and GPA cannot be mapped with g-stage
> + * PMD entries, because we'll end up mapping the wrong pages.
> + *
> + * Consider a layout like the following:
> + *
> + * memslot->userspace_addr:
> + * +-----+--------------------+--------------------+---+
> + *|abcde|fgh vs-stage block | vs-stage block tv|xyz|
> + * +-----+--------------------+--------------------+---+
> + *
> + * memslot->base_gfn << PAGE_SHIFT:
> + * +---+--------------------+--------------------+-----+
> + *|abc|def g-stage block | g-stage block |tvxyz|
> + * +---+--------------------+--------------------+-----+
> + *
> + * If we create those g-stage blocks, we'll end up with this incorrect
> + * mapping:
> + * d -> f
> + * e -> g
> + * f -> h
> + */
> + if ((gpa_start & (PMD_SIZE - 1)) != (uaddr_start & (PMD_SIZE - 1)))
> + return false;
> +
> + /*
> + * Next, let's make sure we're not trying to map anything not covered
> + * by the memslot. This means we have to prohibit block size mappings
> + * for the beginning and end of a non-block aligned and non-block sized
> + * memory slot (illustrated by the head and tail parts of the
> + * userspace view above containing pages 'abcde' and 'xyz',
> + * respectively).
> + *
> + * Note that it doesn't matter if we do the check using the
> + * userspace_addr or the base_gfn, as both are equally aligned (per
> + * the check above) and equally sized.
> + */
> + return (hva >= ALIGN(uaddr_start, PMD_SIZE)) && (hva < ALIGN_DOWN(uaddr_end, PMD_SIZE));
> +}
> +
Thanks,
Alok
Hi,
kernel test robot noticed the following build warnings:
[auto build test WARNING on kvm/queue]
[also build test WARNING on kvm/next mst-vhost/linux-next linus/master v6.17-rc7 next-20250926]
[cannot apply to kvm/linux-next]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/liu-xuemei1-zte-com-cn/RISC-V-KVM-Transparent-huge-page-support/20250928-154904
base: https://git.kernel.org/pub/scm/virt/kvm/kvm.git queue
patch link: https://lore.kernel.org/r/20250928154450701hRC3fm00QYFnGiM0_M1No%40zte.com.cn
patch subject: [PATCH] RISC-V: KVM: Transparent huge page support
config: riscv-randconfig-001-20250928 (https://download.01.org/0day-ci/archive/20250928/202509282326.NFfcoD5h-lkp@intel.com/config)
compiler: riscv64-linux-gcc (GCC) 11.5.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20250928/202509282326.NFfcoD5h-lkp@intel.com/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202509282326.NFfcoD5h-lkp@intel.com/
All warnings (new ones prefixed by >>):
arch/riscv/kvm/gstage.c: In function 'gstage_supports_huge_mapping':
>> arch/riscv/kvm/gstage.c:108:39: warning: left shift count >= width of type [-Wshift-count-overflow]
108 | gpa_start = memslot->base_gfn << PAGE_SIZE;
| ^~
vim +108 arch/riscv/kvm/gstage.c
97
98 static bool gstage_supports_huge_mapping(struct kvm_memory_slot *memslot, unsigned long hva)
99 {
100 gpa_t gpa_start;
101 hva_t uaddr_start, uaddr_end;
102 size_t size;
103
104 size = memslot->npages * PAGE_SIZE;
105 uaddr_start = memslot->userspace_addr;
106 uaddr_end = uaddr_start + size;
107
> 108 gpa_start = memslot->base_gfn << PAGE_SIZE;
109
110 /*
111 * Pages belonging to memslots that don't have the same alignment
112 * within a PMD for userspace and GPA cannot be mapped with g-stage
113 * PMD entries, because we'll end up mapping the wrong pages.
114 *
115 * Consider a layout like the following:
116 *
117 * memslot->userspace_addr:
118 * +-----+--------------------+--------------------+---+
119 * |abcde|fgh vs-stage block | vs-stage block tv|xyz|
120 * +-----+--------------------+--------------------+---+
121 *
122 * memslot->base_gfn << PAGE_SHIFT:
123 * +---+--------------------+--------------------+-----+
124 * |abc|def g-stage block | g-stage block |tvxyz|
125 * +---+--------------------+--------------------+-----+
126 *
127 * If we create those g-stage blocks, we'll end up with this incorrect
128 * mapping:
129 * d -> f
130 * e -> g
131 * f -> h
132 */
133 if ((gpa_start & (PMD_SIZE - 1)) != (uaddr_start & (PMD_SIZE - 1)))
134 return false;
135
136 /*
137 * Next, let's make sure we're not trying to map anything not covered
138 * by the memslot. This means we have to prohibit block size mappings
139 * for the beginning and end of a non-block aligned and non-block sized
140 * memory slot (illustrated by the head and tail parts of the
141 * userspace view above containing pages 'abcde' and 'xyz',
142 * respectively).
143 *
144 * Note that it doesn't matter if we do the check using the
145 * userspace_addr or the base_gfn, as both are equally aligned (per
146 * the check above) and equally sized.
147 */
148 return (hva >= ALIGN(uaddr_start, PMD_SIZE)) && (hva < ALIGN_DOWN(uaddr_end, PMD_SIZE));
149 }
150
--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
© 2016 - 2025 Red Hat, Inc.