[PATCH v5 24/25] mm: Clean up the last pieces of page fault accountings
From: Peter Xu
Date: Tue Jul 07 2020 - 18:51:13 EST
Here're the last pieces of page fault accounting that were still done outside
handle_mm_fault() where we still have regs==NULL when calling handle_mm_fault():
arch/powerpc/mm/copro_fault.c: copro_handle_mm_fault
arch/sparc/mm/fault_32.c: force_user_fault
arch/um/kernel/trap.c: handle_page_fault
mm/gup.c: faultin_page
fixup_user_fault
mm/hmm.c: hmm_vma_fault
mm/ksm.c: break_ksm
Some of them has the issue of duplicated accounting for page fault retries.
Some of them didn't do the accounting at all.
This patch cleans all these up by letting handle_mm_fault() to do per-task page
fault accounting even if regs==NULL (though we'll still skip the perf event
accountings). With that, we can safely remove all the outliers now.
There's another functional change in that now we account the page faults to the
caller of gup, rather than the task_struct that passed into the gup code. More
information of this can be found at [1].
After this patch, below things should never be touched again outside
handle_mm_fault():
- task_struct.[maj|min]_flt
- PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN]
[1] https://lore.kernel.org/lkml/CAHk-=wj_V2Tps2QrMn20_W0OJF9xqNh52XSGA42s-ZJ8Y+GyKw@xxxxxxxxxxxxxx/
Signed-off-by: Peter Xu <peterx@xxxxxxxxxx>
---
arch/powerpc/mm/copro_fault.c | 5 -----
arch/um/kernel/trap.c | 4 ----
mm/gup.c | 13 -------------
mm/memory.c | 17 ++++++++++-------
4 files changed, 10 insertions(+), 29 deletions(-)
diff --git a/arch/powerpc/mm/copro_fault.c b/arch/powerpc/mm/copro_fault.c
index 2d0276abe0a6..8acd00178956 100644
--- a/arch/powerpc/mm/copro_fault.c
+++ b/arch/powerpc/mm/copro_fault.c
@@ -76,11 +76,6 @@ int copro_handle_mm_fault(struct mm_struct *mm, unsigned long ea,
BUG();
}
- if (*flt & VM_FAULT_MAJOR)
- current->maj_flt++;
- else
- current->min_flt++;
-
out_unlock:
mmap_read_unlock(mm);
return ret;
diff --git a/arch/um/kernel/trap.c b/arch/um/kernel/trap.c
index 8d9870d76da1..ad12f78bda7e 100644
--- a/arch/um/kernel/trap.c
+++ b/arch/um/kernel/trap.c
@@ -88,10 +88,6 @@ int handle_page_fault(unsigned long address, unsigned long ip,
BUG();
}
if (flags & FAULT_FLAG_ALLOW_RETRY) {
- if (fault & VM_FAULT_MAJOR)
- current->maj_flt++;
- else
- current->min_flt++;
if (fault & VM_FAULT_RETRY) {
flags |= FAULT_FLAG_TRIED;
diff --git a/mm/gup.c b/mm/gup.c
index 80fd1610d43e..71e1d501a1d3 100644
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -893,13 +893,6 @@ static int faultin_page(struct task_struct *tsk, struct vm_area_struct *vma,
BUG();
}
- if (tsk) {
- if (ret & VM_FAULT_MAJOR)
- tsk->maj_flt++;
- else
- tsk->min_flt++;
- }
-
if (ret & VM_FAULT_RETRY) {
if (locked && !(fault_flags & FAULT_FLAG_RETRY_NOWAIT))
*locked = 0;
@@ -1255,12 +1248,6 @@ int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
goto retry;
}
- if (tsk) {
- if (major)
- tsk->maj_flt++;
- else
- tsk->min_flt++;
- }
return 0;
}
EXPORT_SYMBOL_GPL(fixup_user_fault);
diff --git a/mm/memory.c b/mm/memory.c
index bb7ba127661a..ad5eca9dd1ed 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4404,20 +4404,23 @@ static inline void mm_account_fault(struct pt_regs *regs,
*/
major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED);
+ if (major)
+ current->maj_flt++;
+ else
+ current->min_flt++;
+
/*
- * If the fault is done for GUP, regs will be NULL, and we will skip
- * the fault accounting.
+ * If the fault is done for GUP, regs will be NULL. We only do the
+ * accounting for the per thread fault counters who triggered the
+ * fault, and we skip the perf event updates.
*/
if (!regs)
return;
- if (major) {
- current->maj_flt++;
+ if (major)
perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
- } else {
- current->min_flt++;
+ else
perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
- }
}
/*
--
2.26.2