@@ -26,6 +26,7 @@ void flush_user_icache_range_asm(unsigned long, unsigned long);
void flush_kernel_icache_range_asm(unsigned long, unsigned long);
void flush_user_dcache_range_asm(unsigned long, unsigned long);
void flush_kernel_dcache_range_asm(unsigned long, unsigned long);
+void purge_kernel_dcache_range_asm(unsigned long, unsigned long);
void flush_kernel_dcache_page_asm(void *);
void flush_kernel_icache_page(void *);
@@ -465,10 +465,14 @@ EXPORT_SYMBOL(copy_user_page);
int __flush_tlb_range(unsigned long sid, unsigned long start,
unsigned long end)
{
- unsigned long flags, size;
+ unsigned long flags;
- size = (end - start);
- if (size >= parisc_tlb_flush_threshold) {
+#ifdef CONFIG_SMP
+ if (!arch_irqs_disabled() &&
+ end - start >= parisc_tlb_flush_threshold) {
+#else
+ 1f (end - start >= parisc_tlb_flush_threshold) {
+#endif
flush_tlb_all();
return 1;
}
@@ -539,13 +543,15 @@ void flush_cache_mm(struct mm_struct *mm)
struct vm_area_struct *vma;
pgd_t *pgd;
- /* Flush the TLB to avoid speculation if coherency is required. */
- if (parisc_requires_coherency())
- flush_tlb_all();
-
/* Flushing the whole cache on each cpu takes forever on
rp3440, etc. So, avoid it if the mm isn't too big. */
+#ifdef CONFIG_SMP
+ if (!arch_irqs_disabled() &&
+ mm_total_size(mm) >= parisc_cache_flush_threshold) {
+#else
if (mm_total_size(mm) >= parisc_cache_flush_threshold) {
+#endif
+ flush_tlb_all();
flush_cache_all();
return;
}
@@ -553,9 +559,9 @@ void flush_cache_mm(struct mm_struct *mm)
if (mm->context == mfsp(3)) {
for (vma = mm->mmap; vma; vma = vma->vm_next) {
flush_user_dcache_range_asm(vma->vm_start, vma->vm_end);
- if ((vma->vm_flags & VM_EXEC) == 0)
- continue;
- flush_user_icache_range_asm(vma->vm_start, vma->vm_end);
+ if (vma->vm_flags & VM_EXEC)
+ flush_user_icache_range_asm(vma->vm_start, vma->vm_end);
+ flush_tlb_range(vma, vma->vm_start, vma->vm_end);
}
return;
}
@@ -581,14 +587,13 @@ void flush_cache_mm(struct mm_struct *mm)
void flush_cache_range(struct vm_area_struct *vma,
unsigned long start, unsigned long end)
{
- BUG_ON(!vma->vm_mm->context);
-
- /* Flush the TLB to avoid speculation if coherency is required. */
- if (parisc_requires_coherency())
+#ifdef CONFIG_SMP
+ if (!arch_irqs_disabled() &&
+ end - start >= parisc_cache_flush_threshold) {
+#else
+ if (end - start >= parisc_cache_flush_threshold) {
+#endif
flush_tlb_range(vma, start, end);
-
- if ((end - start) >= parisc_cache_flush_threshold
- || vma->vm_mm->context != mfsp(3)) {
flush_cache_all();
return;
}
@@ -596,6 +601,7 @@ void flush_cache_range(struct vm_area_struct *vma,
flush_user_dcache_range_asm(start, end);
if (vma->vm_flags & VM_EXEC)
flush_user_icache_range_asm(start, end);
+ flush_tlb_range(vma, start, end);
}
void
@@ -604,8 +610,7 @@ flush_cache_page(struct vm_area_struct *vma, unsigned long vmaddr, unsigned long
BUG_ON(!vma->vm_mm->context);
if (pfn_valid(pfn)) {
- if (parisc_requires_coherency())
- flush_tlb_page(vma, vmaddr);
+ flush_tlb_page(vma, vmaddr);
__flush_cache_page(vma, vmaddr, PFN_PHYS(pfn));
}
}
@@ -613,21 +618,41 @@ flush_cache_page(struct vm_area_struct *vma, unsigned long vmaddr, unsigned long
void flush_kernel_vmap_range(void *vaddr, int size)
{
unsigned long start = (unsigned long)vaddr;
+ unsigned long end = start + size;
- if ((unsigned long)size > parisc_cache_flush_threshold)
+#ifdef CONFIG_SMP
+ if (!arch_irqs_disabled() &&
+ (unsigned long)size >= parisc_cache_flush_threshold) {
+#else
+ if ((unsigned long)size >= parisc_cache_flush_threshold) {
+#endif
+ flush_tlb_kernel_range(start, end);
flush_data_cache();
- else
- flush_kernel_dcache_range_asm(start, start + size);
+ return;
+ }
+
+ flush_kernel_dcache_range_asm(start, end);
+ flush_tlb_kernel_range(start, end);
}
EXPORT_SYMBOL(flush_kernel_vmap_range);
void invalidate_kernel_vmap_range(void *vaddr, int size)
{
unsigned long start = (unsigned long)vaddr;
+ unsigned long end = start + size;
- if ((unsigned long)size > parisc_cache_flush_threshold)
+#ifdef CONFIG_SMP
+ if (!arch_irqs_disabled() &&
+ (unsigned long)size >= parisc_cache_flush_threshold) {
+#else
+ if ((unsigned long)size >= parisc_cache_flush_threshold) {
+#endif
+ flush_tlb_kernel_range(start, end);
flush_data_cache();
- else
- flush_kernel_dcache_range_asm(start, start + size);
+ return;
+ }
+
+ purge_kernel_dcache_range_asm(start, end);
+ flush_tlb_kernel_range(start, end);
}
EXPORT_SYMBOL(invalidate_kernel_vmap_range);
@@ -1110,6 +1110,28 @@ ENTRY_CFI(flush_kernel_dcache_range_asm)
.procend
ENDPROC_CFI(flush_kernel_dcache_range_asm)
+ENTRY_CFI(purge_kernel_dcache_range_asm)
+ .proc
+ .callinfo NO_CALLS
+ .entry
+
+ ldil L%dcache_stride, %r1
+ ldw R%dcache_stride(%r1), %r23
+ ldo -1(%r23), %r21
+ ANDCM %r26, %r21, %r26
+
+1: cmpb,COND(<<),n %r26, %r25,1b
+ pdc,m %r23(%r26)
+
+ sync
+ syncdma
+ bv %r0(%r2)
+ nop
+ .exit
+
+ .procend
+ENDPROC_CFI(purge_kernel_dcache_range_asm)
+
ENTRY_CFI(flush_user_icache_range_asm)
.proc
.callinfo NO_CALLS
The change to flush_kernel_vmap_range() wasn't sufficient to avoid the SMP stalls. The problem is some drivers call these routines with interrupts disabled. Interrupts need to be enabled for flush_tlb_all() and flush_cache_all() to work. This version adds checks to ensure interrupts are not disabled before calling routines that need IPI interrupts. When interrupts are disabled, we now drop into slower code. The attached change fixes the ordering of cache and TLB flushes in several cases. When we flush the cache using the existing PTE/TLB entries, we need to flush the TLB after doing the cache flush. We don't need to do this when we flush the entire instruction and data caches as these flushes don't use the existing TLB entries. The same is true for tmpalias region flushes. The flush_kernel_vmap_range() and invalidate_kernel_vmap_range() routines have been updated. Secondly, we added a new purge_kernel_dcache_range_asm() routine to pacache.S and use it in invalidate_kernel_vmap_range(). Nominally, purges are faster than flushes as the cache lines don't have to be written back to memory. Hopefully, this is sufficient to resolve the remaining problems due to cache speculation. So far, testing indicates that this is the case. I did work up a patch using tmpalias flushes, but there is a performance hit because we need the physical address for each page, and we also need to sequence access to the tmpalias flush code. This increases the probability of stalls. Signed-off-by: John David Anglin <dave.anglin@bell.net>