riscv: fix flush_tlb_range() end address for flush_tlb_page()
authorPaul Walmsley <paul.walmsley@sifive.com>
Thu, 8 Aug 2019 02:07:34 +0000 (19:07 -0700)
committerPaul Walmsley <paul.walmsley@sifive.com>
Tue, 13 Aug 2019 21:03:52 +0000 (14:03 -0700)
The RISC-V kernel implementation of flush_tlb_page() when CONFIG_SMP
is set is wrong.  It passes zero to flush_tlb_range() as the final
address to flush, but it should be at least 'addr'.

Some other Linux architecture ports use the beginning address to
flush, plus PAGE_SIZE, as the final address to flush.  This might
flush slightly more than what's needed, but it seems unlikely that
being more clever would improve anything.  So let's just take that
implementation for now.

While here, convert the macro into a static inline function, primarily
to avoid unintentional multiple evaluations of 'addr'.

This second version of the patch fixes a coding style issue found by
Christoph Hellwig <hch@lst.de>.

Reported-by: Andreas Schwab <schwab@suse.de>
Signed-off-by: Paul Walmsley <paul.walmsley@sifive.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
arch/riscv/include/asm/tlbflush.h

index 687dd19735a7e6483afb98a319376706ed426cef..4d9bbe8438bf6a7cff301cfd2c6fec0d268a607c 100644 (file)
@@ -53,10 +53,17 @@ static inline void remote_sfence_vma(struct cpumask *cmask, unsigned long start,
 }
 
 #define flush_tlb_all() sbi_remote_sfence_vma(NULL, 0, -1)
-#define flush_tlb_page(vma, addr) flush_tlb_range(vma, addr, 0)
+
 #define flush_tlb_range(vma, start, end) \
        remote_sfence_vma(mm_cpumask((vma)->vm_mm), start, (end) - (start))
-#define flush_tlb_mm(mm) \
+
+static inline void flush_tlb_page(struct vm_area_struct *vma,
+                                 unsigned long addr)
+{
+       flush_tlb_range(vma, addr, addr + PAGE_SIZE);
+}
+
+#define flush_tlb_mm(mm)                               \
        remote_sfence_vma(mm_cpumask(mm), 0, -1)
 
 #endif /* CONFIG_SMP */