[Skiboot] [RFC PATCH 7/9] Runtime virtual memory for OPAL V4

[Nicholas Piggin]

This patch removes the real-mode switch from the V4 entry, and allows
skiboot to run in virtual mode.

When vm_map is used to access addresses outside the global mappings,
MSR[DR] is disabled, although MSR[IR] remains on (is this wise? It's
not a common configuration for Linux).

Signed-off-by: Nicholas Piggin <npiggin at gmail.com>
---
 core/mem_region.c | 35 +++++++++++++++++++++++++----------
 core/opal.c       | 20 +++++++++++++++-----
 core/vm.c         | 24 +++++++++++++++++++-----
 include/cpu.h     |  1 +
 4 files changed, 60 insertions(+), 20 deletions(-)

diff --git a/core/mem_region.c b/core/mem_region.c
index 69f24d630..c7a689255 100644
--- a/core/mem_region.c
+++ b/core/mem_region.c
@@ -168,16 +168,20 @@ static void mem_poison(struct free_hdr *f)
 	unsigned long start = (unsigned long)(f + 1);
 	unsigned long *t = tailer(f);
 	size_t poison_size = (unsigned long)t - start;
-	void *mem;
 
 	/* We only poison up to a limit, as otherwise boot is
 	 * kinda slow */
 	if (poison_size > POISON_MEM_REGION_LIMIT)
 		poison_size = POISON_MEM_REGION_LIMIT;
 
-	mem = vm_map(start, poison_size, true);
-	memset(mem, POISON_MEM_REGION_WITH, poison_size);
-	vm_unmap(start, poison_size);
+	if (cpu_in_os()) {
+		memset((void *)start, POISON_MEM_REGION_WITH, poison_size);
+	} else {
+		void *mem;
+		mem = vm_map(start, poison_size, true);
+		memset(mem, POISON_MEM_REGION_WITH, poison_size);
+		vm_unmap(start, poison_size);
+	}
 }
 #endif
 
@@ -211,9 +215,14 @@ static void init_allocatable_region(struct mem_region *region)
 	f->hdr.prev_free = false;
 	list_add(&region->free_list, &f->list);
 
-	t = vm_map((unsigned long)tailer(f), sizeof(long), true);
-	*t = num_longs;
-	vm_unmap((unsigned long)tailer(f), sizeof(long));
+	if (cpu_in_os()) {
+		t = tailer(f);
+		*t = num_longs;
+	} else {
+		t = vm_map((unsigned long)tailer(f), sizeof(long), true);
+		*t = num_longs;
+		vm_unmap((unsigned long)tailer(f), sizeof(long));
+	}
 
 #if POISON_MEM_REGION == 1
 	mem_poison(f);
@@ -277,9 +286,15 @@ static void make_free(struct mem_region *region, struct free_hdr *f,
 
 no_unmap:
 	/* Fix up tailer. */
-	t = vm_map((unsigned long)tailer(f), sizeof(long), true);
-	*t = f->hdr.num_longs;
-	vm_unmap((unsigned long)tailer(f), sizeof(long));
+	if (cpu_in_os()) {
+		t = tailer(f);
+		*t = f->hdr.num_longs;
+	} else {
+		t = vm_map((unsigned long)tailer(f), sizeof(long), true);
+		*t = f->hdr.num_longs;
+		vm_unmap((unsigned long)tailer(f), sizeof(long));
+	}
+
 }
 
 /* Can we fit this many longs with this alignment in this free block? */
diff --git a/core/opal.c b/core/opal.c
index 27ce7a375..7fb4fae86 100644
--- a/core/opal.c
+++ b/core/opal.c
@@ -398,6 +398,8 @@ static int64_t opal_v4_le_entry(uint64_t token, uint64_t r4, uint64_t r5,
 {
 	struct cpu_thread *cpu;
 	opal_call_fn *fn;
+	bool was_vm_setup;
+	uint64_t old_opal_call_msr;
 	uint64_t msr;
 	uint32_t pir;
 	uint64_t r16;
@@ -406,14 +408,20 @@ static int64_t opal_v4_le_entry(uint64_t token, uint64_t r4, uint64_t r5,
 	msr = mfmsr();
 	assert(!(msr & MSR_EE));
 
-	if (msr & (MSR_IR|MSR_DR))
-		mtmsrd(msr & ~(MSR_IR|MSR_DR), 0);
-
 	pir = mfspr(SPR_PIR);
 	r16 = (uint64_t)__this_cpu;
 	__this_cpu = &cpu_stacks[pir].cpu;
 	cpu = this_cpu();
 
+	old_opal_call_msr = cpu->opal_call_msr;
+	cpu->opal_call_msr = msr;
+
+	was_vm_setup = this_cpu()->vm_setup;
+	if (msr & (MSR_IR|MSR_DR))
+		this_cpu()->vm_setup = true;
+	else
+		this_cpu()->vm_setup = false;
+
 	cpu->in_opal_call++;
 	if (cpu->in_opal_call == 1) {
 		cpu->current_token = token;
@@ -453,11 +461,13 @@ out:
 	assert(cpu->in_opal_call > 0);
 	cpu->in_opal_call--;
 
-	if (msr != mfmsr())
-		mtmsrd(msr, 0);
+	this_cpu()->vm_setup = was_vm_setup;
+	cpu->opal_call_msr = old_opal_call_msr;
 
 	__this_cpu = (struct cpu_thread *)r16;
 
+	assert(mfmsr() == msr);
+
 	return ret;
 }
 
diff --git a/core/vm.c b/core/vm.c
index 236a0e70b..b2830fccf 100644
--- a/core/vm.c
+++ b/core/vm.c
@@ -491,9 +491,18 @@ void *vm_map(unsigned long addr, unsigned long len, bool rw)
 	assert(!c->vm_local_map_inuse);
 	c->vm_local_map_inuse = true;
 
-	if (c->vm_setup) {
+	if (cpu_in_os() && c->vm_setup) {
+		assert(c->opal_call_msr & (MSR_IR|MSR_DR));
+		newaddr = addr;
+		mtmsr(c->opal_call_msr & ~MSR_DR);
+		c->vm_setup = false;
+	} else if (!c->vm_setup) {
+		newaddr = addr;
+	} else {
 		struct vm_map *new = &c->vm_local_map;
 
+		assert(!cpu_in_os());
+
 		newaddr = LOCAL_EA_BEGIN + LOCAL_EA_PERCPU * c->pir;
 
 		new->name = "local";
@@ -504,9 +513,6 @@ void *vm_map(unsigned long addr, unsigned long len, bool rw)
 		new->writeable = rw;
 		new->executable = false;
 		new->ci = false;
-
-	} else {
-		newaddr = addr;
 	}
 
 	return (void *)newaddr + offset;
@@ -527,10 +533,18 @@ void vm_unmap(unsigned long addr, unsigned long len)
 	assert(c->vm_local_map_inuse);
 	c->vm_local_map_inuse = false;
 
-	if (c->vm_setup) {
+	if (cpu_in_os() && (c->opal_call_msr & (MSR_IR|MSR_DR))) {
+		assert(!c->vm_setup);
+		c->vm_setup = true;
+		mtmsr(c->opal_call_msr);
+	} else if (!c->vm_setup) {
+		return;
+	} else {
 		struct vm_map *vmm;
 		unsigned long ea;
 
+		assert(!cpu_in_os());
+
 		newaddr = LOCAL_EA_BEGIN + LOCAL_EA_PERCPU * c->pir;
 
 		vmm = &c->vm_local_map;
diff --git a/include/cpu.h b/include/cpu.h
index b769024d1..bbfbf7ca9 100644
--- a/include/cpu.h
+++ b/include/cpu.h
@@ -59,6 +59,7 @@ struct cpu_thread {
 	enum cpu_thread_state		state;
 	struct dt_node			*node;
 	struct trace_info		*trace;
+	uint64_t			opal_call_msr;
 	uint64_t			save_r1;
 	void				*icp_regs;
 	uint32_t			in_opal_call;
-- 
2.23.0