[RFC PATCH] powerpc/64s/radix: introduce option to disable broadcast tlbie

Subsystems: linux for powerpc (32-bit and 64-bit), the rest

STALE2615d

4 messages, 3 authors, 2019-08-14 · open the first message on its own page

[RFC PATCH] powerpc/64s/radix: introduce option to disable broadcast tlbie

From: Nicholas Piggin <npiggin@gmail.com>
Date: 2019-07-31 13:50:05

This is an initial hack of a quick option to disable use of the tlbie
instruction. This takes the simplest possible initial pass of just
replacing low level tlbie functions with IPIs. This means it's not as
performant as it could be if we spend some time optmizing it, but on
the other hand having a 1:1 replacement of tlbie is simple and can be
useful for comparisons so I think it's the right initial approach.

It's not entirely complete, doesn't deal with accelerators (reverts to
tlbie), not all the boot code is converted, kernel space invalidations
not converted, and KVM not converted, also radix only to start with. We
can start to add more cases if this will be useful.

Thanks,
Nick
---
 arch/powerpc/mm/book3s64/radix_tlb.c | 149 +++++++++++++++++++++++++--
 1 file changed, 140 insertions(+), 9 deletions(-)
diff --git a/arch/powerpc/mm/book3s64/radix_tlb.c b/arch/powerpc/mm/book3s64/radix_tlb.c
index 71f7fede2fa4..56ceecbd3d5c 100644
--- a/arch/powerpc/mm/book3s64/radix_tlb.c
+++ b/arch/powerpc/mm/book3s64/radix_tlb.c
@@ -11,6 +11,7 @@
 #include <linux/mmu_context.h>
 #include <linux/sched/mm.h>
 
+#include <asm/debugfs.h>
 #include <asm/ppc-opcode.h>
 #include <asm/tlb.h>
 #include <asm/tlbflush.h>
@@ -285,6 +286,30 @@ static inline void _tlbie_pid(unsigned long pid, unsigned long ric)
 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
 }
 
+struct tlbiel_pid {
+	unsigned long pid;
+	unsigned long ric;
+};
+
+static void do_tlbiel_pid(void *info)
+{
+	struct tlbiel_pid *t = info;
+
+	if (t->ric == RIC_FLUSH_TLB)
+		_tlbiel_pid(t->pid, RIC_FLUSH_TLB);
+	else if (t->ric == RIC_FLUSH_PWC)
+		_tlbiel_pid(t->pid, RIC_FLUSH_PWC);
+	else
+		_tlbiel_pid(t->pid, RIC_FLUSH_ALL);
+}
+
+static inline void _tlbiel_pid_broadcast(const struct cpumask *cpus,
+				unsigned long pid, unsigned long ric)
+{
+	struct tlbiel_pid t = { .pid = pid, .ric = ric };
+	on_each_cpu_mask(cpus, do_tlbiel_pid, &t, 1);
+}
+
 static inline void _tlbiel_lpid(unsigned long lpid, unsigned long ric)
 {
 	int set;
@@ -420,6 +445,61 @@ static __always_inline void _tlbie_va(unsigned long va, unsigned long pid,
 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
 }
 
+struct tlbiel_va {
+	unsigned long pid;
+	unsigned long va;
+	unsigned long psize;
+	unsigned long ric;
+};
+
+static void do_tlbiel_va(void *info)
+{
+	struct tlbiel_va *t = info;
+
+	if (t->ric == RIC_FLUSH_TLB)
+		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_TLB);
+	else if (t->ric == RIC_FLUSH_PWC)
+		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_PWC);
+	else
+		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_ALL);
+}
+
+static inline void _tlbiel_va_broadcast(const struct cpumask *cpus,
+				unsigned long va, unsigned long pid,
+				unsigned long psize, unsigned long ric)
+{
+	struct tlbiel_va t = { .va = va, .pid = pid, .psize = psize, .ric = ric };
+	on_each_cpu_mask(cpus, do_tlbiel_va, &t, 1);
+}
+
+struct tlbiel_va_range {
+	unsigned long pid;
+	unsigned long start;
+	unsigned long end;
+	unsigned long page_size;
+	unsigned long psize;
+	bool also_pwc;
+};
+
+static void do_tlbiel_va_range(void *info)
+{
+	struct tlbiel_va_range *t = info;
+
+	_tlbiel_va_range(t->start, t->end, t->pid, t->page_size,
+				    t->psize, t->also_pwc);
+}
+
+static inline void _tlbiel_va_range_broadcast(const struct cpumask *cpus,
+				unsigned long start, unsigned long end,
+				unsigned long pid, unsigned long page_size,
+				unsigned long psize, bool also_pwc)
+{
+	struct tlbiel_va_range t = { .start = start, .end = end,
+				.pid = pid, .page_size = page_size,
+				.psize = psize, .also_pwc = also_pwc };
+	on_each_cpu_mask(cpus, do_tlbiel_va_range, &t, 1);
+}
+
 static __always_inline void _tlbie_lpid_va(unsigned long va, unsigned long lpid,
 			      unsigned long psize, unsigned long ric)
 {
@@ -524,6 +604,12 @@ static bool mm_needs_flush_escalation(struct mm_struct *mm)
 	return false;
 }
 
+static bool tlbie_enabled = true;
+static bool use_tlbie(void)
+{
+	return tlbie_enabled;
+}
+
 #ifdef CONFIG_SMP
 static void do_exit_flush_lazy_tlb(void *arg)
 {
@@ -582,8 +668,10 @@ void radix__flush_tlb_mm(struct mm_struct *mm)
 
 		if (mm_needs_flush_escalation(mm))
 			_tlbie_pid(pid, RIC_FLUSH_ALL);
-		else
+		else if (use_tlbie())
 			_tlbie_pid(pid, RIC_FLUSH_TLB);
+		else
+			_tlbiel_pid_broadcast(mm_cpumask(mm), pid, RIC_FLUSH_TLB);
 	} else {
 local:
 		_tlbiel_pid(pid, RIC_FLUSH_TLB);
@@ -609,7 +697,10 @@ static void __flush_all_mm(struct mm_struct *mm, bool fullmm)
 				goto local;
 			}
 		}
-		_tlbie_pid(pid, RIC_FLUSH_ALL);
+		if (mm_needs_flush_escalation(mm) || use_tlbie())
+			_tlbie_pid(pid, RIC_FLUSH_ALL);
+		else
+			_tlbiel_pid_broadcast(mm_cpumask(mm), pid, RIC_FLUSH_ALL);
 	} else {
 local:
 		_tlbiel_pid(pid, RIC_FLUSH_ALL);
@@ -644,7 +735,11 @@ void radix__flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
 			exit_flush_lazy_tlbs(mm);
 			goto local;
 		}
-		_tlbie_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
+		if (mm_needs_flush_escalation(mm) || use_tlbie())
+			_tlbie_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
+		else
+			_tlbiel_va_broadcast(mm_cpumask(mm),
+				       vmaddr, pid, psize, RIC_FLUSH_TLB);
 	} else {
 local:
 		_tlbiel_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
@@ -731,8 +826,11 @@ static inline void __radix__flush_tlb_range(struct mm_struct *mm,
 		} else {
 			if (mm_needs_flush_escalation(mm))
 				_tlbie_pid(pid, RIC_FLUSH_ALL);
-			else
+			else if (use_tlbie())
 				_tlbie_pid(pid, RIC_FLUSH_TLB);
+			else
+				_tlbiel_pid_broadcast(mm_cpumask(mm),
+						pid, RIC_FLUSH_TLB);
 		}
 	} else {
 		bool hflush = flush_all_sizes;
@@ -757,8 +855,8 @@ static inline void __radix__flush_tlb_range(struct mm_struct *mm,
 				gflush = false;
 		}
 
-		asm volatile("ptesync": : :"memory");
 		if (local) {
+			asm volatile("ptesync": : :"memory");
 			__tlbiel_va_range(start, end, pid, page_size, mmu_virtual_psize);
 			if (hflush)
 				__tlbiel_va_range(hstart, hend, pid,
@@ -767,7 +865,8 @@ static inline void __radix__flush_tlb_range(struct mm_struct *mm,
 				__tlbiel_va_range(gstart, gend, pid,
 						PUD_SIZE, MMU_PAGE_1G);
 			asm volatile("ptesync": : :"memory");
-		} else {
+		} else if (use_tlbie()) {
+			asm volatile("ptesync": : :"memory");
 			__tlbie_va_range(start, end, pid, page_size, mmu_virtual_psize);
 			if (hflush)
 				__tlbie_va_range(hstart, hend, pid,
@@ -777,6 +876,15 @@ static inline void __radix__flush_tlb_range(struct mm_struct *mm,
 						PUD_SIZE, MMU_PAGE_1G);
 			fixup_tlbie();
 			asm volatile("eieio; tlbsync; ptesync": : :"memory");
+		} else {
+			_tlbiel_va_range_broadcast(mm_cpumask(mm),
+					start, end, pid, page_size, mmu_virtual_psize, false);
+			if (hflush)
+				_tlbiel_va_range_broadcast(mm_cpumask(mm),
+					hstart, hend, pid, PMD_SIZE, MMU_PAGE_2M, false);
+			if (gflush)
+				_tlbiel_va_range_broadcast(mm_cpumask(mm),
+					gstart, gend, pid, PUD_SIZE, MMU_PAGE_1G, false);
 		}
 	}
 	preempt_enable();
@@ -969,13 +1077,22 @@ static __always_inline void __radix__flush_tlb_range_psize(struct mm_struct *mm,
 			if (mm_needs_flush_escalation(mm))
 				also_pwc = true;
 
-			_tlbie_pid(pid, also_pwc ? RIC_FLUSH_ALL : RIC_FLUSH_TLB);
+			if (use_tlbie())
+				_tlbie_pid(pid,
+					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
+			else
+				_tlbiel_pid_broadcast(mm_cpumask(mm), pid,
+					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
+
 		}
 	} else {
 		if (local)
 			_tlbiel_va_range(start, end, pid, page_size, psize, also_pwc);
-		else
+		else if (mm_needs_flush_escalation(mm) || use_tlbie())
 			_tlbie_va_range(start, end, pid, page_size, psize, also_pwc);
+		else
+			_tlbiel_va_range_broadcast(mm_cpumask(mm),
+					start, end, pid, page_size, psize, also_pwc);
 	}
 	preempt_enable();
 }
@@ -1017,7 +1134,11 @@ void radix__flush_tlb_collapsed_pmd(struct mm_struct *mm, unsigned long addr)
 			exit_flush_lazy_tlbs(mm);
 			goto local;
 		}
-		_tlbie_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
+		if (mm_needs_flush_escalation(mm) || use_tlbie())
+			_tlbie_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
+		else
+			_tlbiel_va_range_broadcast(mm_cpumask(mm),
+					addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
 	} else {
 local:
 		_tlbiel_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
@@ -1100,3 +1221,13 @@ extern void radix_kvm_prefetch_workaround(struct mm_struct *mm)
 }
 EXPORT_SYMBOL_GPL(radix_kvm_prefetch_workaround);
 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
+
+static int __init radix_tlb_setup(void)
+{
+	debugfs_create_bool("tlbie_enabled", 0600,
+			powerpc_debugfs_root,
+			&tlbie_enabled);
+
+	return 0;
+}
+arch_initcall(radix_tlb_setup);
-- 
2.22.0

Re: [RFC PATCH] powerpc/64s/radix: introduce option to disable broadcast tlbie

From: Christophe Leroy <hidden>
Date: 2019-07-31 14:03:28


Le 31/07/2019 à 14:32, Nicholas Piggin a écrit :
This is an initial hack of a quick option to disable use of the tlbie
instruction. This takes the simplest possible initial pass of just
replacing low level tlbie functions with IPIs. This means it's not as
performant as it could be if we spend some time optmizing it, but on
the other hand having a 1:1 replacement of tlbie is simple and can be
useful for comparisons so I think it's the right initial approach.
Can you explain why we want to optionnaly disable use of tlbie ?

Christophe
quoted hunk
It's not entirely complete, doesn't deal with accelerators (reverts to
tlbie), not all the boot code is converted, kernel space invalidations
not converted, and KVM not converted, also radix only to start with. We
can start to add more cases if this will be useful.

Thanks,
Nick
---
  arch/powerpc/mm/book3s64/radix_tlb.c | 149 +++++++++++++++++++++++++--
  1 file changed, 140 insertions(+), 9 deletions(-)
diff --git a/arch/powerpc/mm/book3s64/radix_tlb.c b/arch/powerpc/mm/book3s64/radix_tlb.c
index 71f7fede2fa4..56ceecbd3d5c 100644
--- a/arch/powerpc/mm/book3s64/radix_tlb.c
+++ b/arch/powerpc/mm/book3s64/radix_tlb.c
@@ -11,6 +11,7 @@
  #include <linux/mmu_context.h>
  #include <linux/sched/mm.h>
  
+#include <asm/debugfs.h>
  #include <asm/ppc-opcode.h>
  #include <asm/tlb.h>
  #include <asm/tlbflush.h>
@@ -285,6 +286,30 @@ static inline void _tlbie_pid(unsigned long pid, unsigned long ric)
  	asm volatile("eieio; tlbsync; ptesync": : :"memory");
  }
  
+struct tlbiel_pid {
+	unsigned long pid;
+	unsigned long ric;
+};
+
+static void do_tlbiel_pid(void *info)
+{
+	struct tlbiel_pid *t = info;
+
+	if (t->ric == RIC_FLUSH_TLB)
+		_tlbiel_pid(t->pid, RIC_FLUSH_TLB);
+	else if (t->ric == RIC_FLUSH_PWC)
+		_tlbiel_pid(t->pid, RIC_FLUSH_PWC);
+	else
+		_tlbiel_pid(t->pid, RIC_FLUSH_ALL);
+}
+
+static inline void _tlbiel_pid_broadcast(const struct cpumask *cpus,
+				unsigned long pid, unsigned long ric)
+{
+	struct tlbiel_pid t = { .pid = pid, .ric = ric };
+	on_each_cpu_mask(cpus, do_tlbiel_pid, &t, 1);
+}
+
  static inline void _tlbiel_lpid(unsigned long lpid, unsigned long ric)
  {
  	int set;
@@ -420,6 +445,61 @@ static __always_inline void _tlbie_va(unsigned long va, unsigned long pid,
  	asm volatile("eieio; tlbsync; ptesync": : :"memory");
  }
  
+struct tlbiel_va {
+	unsigned long pid;
+	unsigned long va;
+	unsigned long psize;
+	unsigned long ric;
+};
+
+static void do_tlbiel_va(void *info)
+{
+	struct tlbiel_va *t = info;
+
+	if (t->ric == RIC_FLUSH_TLB)
+		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_TLB);
+	else if (t->ric == RIC_FLUSH_PWC)
+		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_PWC);
+	else
+		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_ALL);
+}
+
+static inline void _tlbiel_va_broadcast(const struct cpumask *cpus,
+				unsigned long va, unsigned long pid,
+				unsigned long psize, unsigned long ric)
+{
+	struct tlbiel_va t = { .va = va, .pid = pid, .psize = psize, .ric = ric };
+	on_each_cpu_mask(cpus, do_tlbiel_va, &t, 1);
+}
+
+struct tlbiel_va_range {
+	unsigned long pid;
+	unsigned long start;
+	unsigned long end;
+	unsigned long page_size;
+	unsigned long psize;
+	bool also_pwc;
+};
+
+static void do_tlbiel_va_range(void *info)
+{
+	struct tlbiel_va_range *t = info;
+
+	_tlbiel_va_range(t->start, t->end, t->pid, t->page_size,
+				    t->psize, t->also_pwc);
+}
+
+static inline void _tlbiel_va_range_broadcast(const struct cpumask *cpus,
+				unsigned long start, unsigned long end,
+				unsigned long pid, unsigned long page_size,
+				unsigned long psize, bool also_pwc)
+{
+	struct tlbiel_va_range t = { .start = start, .end = end,
+				.pid = pid, .page_size = page_size,
+				.psize = psize, .also_pwc = also_pwc };
+	on_each_cpu_mask(cpus, do_tlbiel_va_range, &t, 1);
+}
+
  static __always_inline void _tlbie_lpid_va(unsigned long va, unsigned long lpid,
  			      unsigned long psize, unsigned long ric)
  {
@@ -524,6 +604,12 @@ static bool mm_needs_flush_escalation(struct mm_struct *mm)
  	return false;
  }
  
+static bool tlbie_enabled = true;
+static bool use_tlbie(void)
+{
+	return tlbie_enabled;
+}
+
  #ifdef CONFIG_SMP
  static void do_exit_flush_lazy_tlb(void *arg)
  {
@@ -582,8 +668,10 @@ void radix__flush_tlb_mm(struct mm_struct *mm)
  
  		if (mm_needs_flush_escalation(mm))
  			_tlbie_pid(pid, RIC_FLUSH_ALL);
-		else
+		else if (use_tlbie())
  			_tlbie_pid(pid, RIC_FLUSH_TLB);
+		else
+			_tlbiel_pid_broadcast(mm_cpumask(mm), pid, RIC_FLUSH_TLB);
  	} else {
  local:
  		_tlbiel_pid(pid, RIC_FLUSH_TLB);
@@ -609,7 +697,10 @@ static void __flush_all_mm(struct mm_struct *mm, bool fullmm)
  				goto local;
  			}
  		}
-		_tlbie_pid(pid, RIC_FLUSH_ALL);
+		if (mm_needs_flush_escalation(mm) || use_tlbie())
+			_tlbie_pid(pid, RIC_FLUSH_ALL);
+		else
+			_tlbiel_pid_broadcast(mm_cpumask(mm), pid, RIC_FLUSH_ALL);
  	} else {
  local:
  		_tlbiel_pid(pid, RIC_FLUSH_ALL);
@@ -644,7 +735,11 @@ void radix__flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
  			exit_flush_lazy_tlbs(mm);
  			goto local;
  		}
-		_tlbie_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
+		if (mm_needs_flush_escalation(mm) || use_tlbie())
+			_tlbie_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
+		else
+			_tlbiel_va_broadcast(mm_cpumask(mm),
+				       vmaddr, pid, psize, RIC_FLUSH_TLB);
  	} else {
  local:
  		_tlbiel_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
@@ -731,8 +826,11 @@ static inline void __radix__flush_tlb_range(struct mm_struct *mm,
  		} else {
  			if (mm_needs_flush_escalation(mm))
  				_tlbie_pid(pid, RIC_FLUSH_ALL);
-			else
+			else if (use_tlbie())
  				_tlbie_pid(pid, RIC_FLUSH_TLB);
+			else
+				_tlbiel_pid_broadcast(mm_cpumask(mm),
+						pid, RIC_FLUSH_TLB);
  		}
  	} else {
  		bool hflush = flush_all_sizes;
@@ -757,8 +855,8 @@ static inline void __radix__flush_tlb_range(struct mm_struct *mm,
  				gflush = false;
  		}
  
-		asm volatile("ptesync": : :"memory");
  		if (local) {
+			asm volatile("ptesync": : :"memory");
  			__tlbiel_va_range(start, end, pid, page_size, mmu_virtual_psize);
  			if (hflush)
  				__tlbiel_va_range(hstart, hend, pid,
@@ -767,7 +865,8 @@ static inline void __radix__flush_tlb_range(struct mm_struct *mm,
  				__tlbiel_va_range(gstart, gend, pid,
  						PUD_SIZE, MMU_PAGE_1G);
  			asm volatile("ptesync": : :"memory");
-		} else {
+		} else if (use_tlbie()) {
+			asm volatile("ptesync": : :"memory");
  			__tlbie_va_range(start, end, pid, page_size, mmu_virtual_psize);
  			if (hflush)
  				__tlbie_va_range(hstart, hend, pid,
@@ -777,6 +876,15 @@ static inline void __radix__flush_tlb_range(struct mm_struct *mm,
  						PUD_SIZE, MMU_PAGE_1G);
  			fixup_tlbie();
  			asm volatile("eieio; tlbsync; ptesync": : :"memory");
+		} else {
+			_tlbiel_va_range_broadcast(mm_cpumask(mm),
+					start, end, pid, page_size, mmu_virtual_psize, false);
+			if (hflush)
+				_tlbiel_va_range_broadcast(mm_cpumask(mm),
+					hstart, hend, pid, PMD_SIZE, MMU_PAGE_2M, false);
+			if (gflush)
+				_tlbiel_va_range_broadcast(mm_cpumask(mm),
+					gstart, gend, pid, PUD_SIZE, MMU_PAGE_1G, false);
  		}
  	}
  	preempt_enable();
@@ -969,13 +1077,22 @@ static __always_inline void __radix__flush_tlb_range_psize(struct mm_struct *mm,
  			if (mm_needs_flush_escalation(mm))
  				also_pwc = true;
  
-			_tlbie_pid(pid, also_pwc ? RIC_FLUSH_ALL : RIC_FLUSH_TLB);
+			if (use_tlbie())
+				_tlbie_pid(pid,
+					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
+			else
+				_tlbiel_pid_broadcast(mm_cpumask(mm), pid,
+					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
+
  		}
  	} else {
  		if (local)
  			_tlbiel_va_range(start, end, pid, page_size, psize, also_pwc);
-		else
+		else if (mm_needs_flush_escalation(mm) || use_tlbie())
  			_tlbie_va_range(start, end, pid, page_size, psize, also_pwc);
+		else
+			_tlbiel_va_range_broadcast(mm_cpumask(mm),
+					start, end, pid, page_size, psize, also_pwc);
  	}
  	preempt_enable();
  }
@@ -1017,7 +1134,11 @@ void radix__flush_tlb_collapsed_pmd(struct mm_struct *mm, unsigned long addr)
  			exit_flush_lazy_tlbs(mm);
  			goto local;
  		}
-		_tlbie_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
+		if (mm_needs_flush_escalation(mm) || use_tlbie())
+			_tlbie_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
+		else
+			_tlbiel_va_range_broadcast(mm_cpumask(mm),
+					addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
  	} else {
  local:
  		_tlbiel_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
@@ -1100,3 +1221,13 @@ extern void radix_kvm_prefetch_workaround(struct mm_struct *mm)
  }
  EXPORT_SYMBOL_GPL(radix_kvm_prefetch_workaround);
  #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
+
+static int __init radix_tlb_setup(void)
+{
+	debugfs_create_bool("tlbie_enabled", 0600,
+			powerpc_debugfs_root,
+			&tlbie_enabled);
+
+	return 0;
+}
+arch_initcall(radix_tlb_setup);

Re: [RFC PATCH] powerpc/64s/radix: introduce option to disable broadcast tlbie

From: Nicholas Piggin <npiggin@gmail.com>
Date: 2019-07-31 23:09:58

Christophe Leroy's on July 31, 2019 11:56 pm:

Le 31/07/2019 à 14:32, Nicholas Piggin a écrit :
quoted
This is an initial hack of a quick option to disable use of the tlbie
instruction. This takes the simplest possible initial pass of just
replacing low level tlbie functions with IPIs. This means it's not as
performant as it could be if we spend some time optmizing it, but on
the other hand having a 1:1 replacement of tlbie is simple and can be
useful for comparisons so I think it's the right initial approach.
Can you explain why we want to optionnaly disable use of tlbie ?
It's something we've wanted to have more control of for a while.

One is for testing performance, especially on large systems it is not
always best to use tlbie. For example if you have threaded apps that
only run on a few CPUs (e.g., process * thread hybrid process model
server).

There is also concern about coherent accelerators may have high
latency to respond to tlbie, which can block the resource for others.

And we did have a hardware errata with early POWER9 it would have
been nice to disable it until the kernel could be updated with the
workaround.

Thanks,
Nick

Re: [RFC PATCH] powerpc/64s/radix: introduce option to disable broadcast tlbie

From: Michael Ellerman <mpe@ellerman.id.au>
Date: 2019-08-14 02:08:17

Hi Nick,

Just a few comments.

Nicholas Piggin [off-list ref] writes:
quoted hunk
diff --git a/arch/powerpc/mm/book3s64/radix_tlb.c b/arch/powerpc/mm/book3s64/radix_tlb.c
index 71f7fede2fa4..56ceecbd3d5c 100644
--- a/arch/powerpc/mm/book3s64/radix_tlb.c
+++ b/arch/powerpc/mm/book3s64/radix_tlb.c
@@ -285,6 +286,30 @@ static inline void _tlbie_pid(unsigned long pid, unsigned long ric)
 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
 }
 
+struct tlbiel_pid {
+	unsigned long pid;
+	unsigned long ric;
+};
+
+static void do_tlbiel_pid(void *info)
+{
+	struct tlbiel_pid *t = info;
+
+	if (t->ric == RIC_FLUSH_TLB)
+		_tlbiel_pid(t->pid, RIC_FLUSH_TLB);
+	else if (t->ric == RIC_FLUSH_PWC)
+		_tlbiel_pid(t->pid, RIC_FLUSH_PWC);
+	else
+		_tlbiel_pid(t->pid, RIC_FLUSH_ALL);
+}
+
+static inline void _tlbiel_pid_broadcast(const struct cpumask *cpus,
+				unsigned long pid, unsigned long ric)
Can we call these "multicast" instead of "broadcast"?

I think that's more accurate, and avoids confusion with tlbie which
literally does a broadcast (at least architecturally).
quoted hunk
@@ -524,6 +604,12 @@ static bool mm_needs_flush_escalation(struct mm_struct *mm)
 	return false;
 }
 
+static bool tlbie_enabled = true;
+static bool use_tlbie(void)
+{
+	return tlbie_enabled;
+}
No synchronisation, but that's OK. Would probably be good to have
a comment though explaining why.

We could use a static_key but I guess the overhead of a comparison and
branch is in the noise vs the tlbie/tlbiel.
quoted hunk
@@ -1100,3 +1221,13 @@ extern void radix_kvm_prefetch_workaround(struct mm_struct *mm)
 }
 EXPORT_SYMBOL_GPL(radix_kvm_prefetch_workaround);
 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
+
+static int __init radix_tlb_setup(void)
+{
+	debugfs_create_bool("tlbie_enabled", 0600,
+			powerpc_debugfs_root,
+			&tlbie_enabled);
+
+	return 0;
+}
+arch_initcall(radix_tlb_setup);
For working around hardware bugs we would want a command line parameter
or other boot time way to flip this. But I guess you're saying because
we haven't converted all uses of tlbie we can't really support that
anyway, and so a runtime switch is sufficient?

cheers
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help