[PATCH 6/6] x86/intel_rdt: Support dcache and icache mask for Code data prioritization

[Vikas Shivappa <vikas.shivappa@linux.intel.com>]

Add support to read and write the per cgroup data and instruction cache
masks.
 - When a new cgroup directory is created it inherits the parents dcache
 and icache mask. So a mkdir never fails.
 - When user changes the mask, we look to see if we can reuse an
 existing CLOSid or else allocate a new one. When we run of CLOSids , we
 simply return -ENOSPC.
---
 arch/x86/kernel/cpu/intel_rdt.c | 177 ++++++++++++++++++++++++++++++++++++----
 1 file changed, 163 insertions(+), 14 deletions(-)

diff --git a/arch/x86/kernel/cpu/intel_rdt.c b/arch/x86/kernel/cpu/intel_rdt.c
index 285db1e..d0eaf04 100644
--- a/arch/x86/kernel/cpu/intel_rdt.c
+++ b/arch/x86/kernel/cpu/intel_rdt.c
@@ -428,6 +428,22 @@ static int intel_cache_alloc_cbm_read(struct seq_file *m, void *v)
 	return 0;
 }
 
+static int cdp_dcache_read(struct seq_file *m, void *v)
+{
+	struct intel_rdt *ir = css_rdt(seq_css(m));
+
+	seq_printf(m, "%08lx\n", cdp_cm_map[ir->closid].dcache_mask);
+	return 0;
+}
+
+static int cdp_icache_read(struct seq_file *m, void *v)
+{
+	struct intel_rdt *ir = css_rdt(seq_css(m));
+
+	seq_printf(m, "%08lx\n", cdp_cm_map[ir->closid].icache_mask);
+	return 0;
+}
+
 static inline bool cbm_is_contiguous(unsigned long var)
 {
 	unsigned long maxcbm = MAX_CBM_LENGTH;
@@ -445,13 +461,24 @@ static inline bool cbm_is_contiguous(unsigned long var)
 	return true;
 }
 
-static int cbm_validate(struct intel_rdt *ir, unsigned long cbmvalue)
+static int cache_mask_validate(struct intel_rdt *ir, unsigned long cbmvalue)
 {
+	u32 max_cbm = boot_cpu_data.x86_cache_max_cbm_len;
 	struct cgroup_subsys_state *css;
 	struct intel_rdt *par, *c;
 	unsigned long *cbm_tmp;
+	u64 max_mask;
 	int err = 0;
 
+	if (ir == &rdt_root_group)
+		return -EPERM;
+
+	max_mask = (1ULL << max_cbm) - 1;
+	if (cbmvalue & ~max_mask) {
+		err = -EINVAL;
+		goto out_err;
+	}
+
 	if (!cbm_is_contiguous(cbmvalue)) {
 		pr_err("bitmask should have >=%d bits and be contiguous\n",
 			 min_bitmask_len);
@@ -486,10 +513,12 @@ out_err:
 static bool cbm_search(unsigned long cbm, u32 *closid)
 {
 	u32 maxid = boot_cpu_data.x86_cache_max_closid;
+	unsigned long cache_mask;
 	u32 i;
 
 	for (i = 0; i < maxid; i++) {
-		if (bitmap_equal(&cbm, &cat_cm_map[i].cache_mask, MAX_CBM_LENGTH)) {
+		cache_mask = cat_cm_map[i].cache_mask;
+		if (bitmap_equal(&cbm, &cache_mask, MAX_CBM_LENGTH)) {
 			*closid = i;
 			return true;
 		}
@@ -511,30 +540,19 @@ static bool cbm_search(unsigned long cbm, u32 *closid)
 static int intel_cache_alloc_cbm_write(struct cgroup_subsys_state *css,
 				 struct cftype *cft, u64 cbmvalue)
 {
-	u32 max_cbm = boot_cpu_data.x86_cache_max_cbm_len;
 	struct intel_rdt *ir = css_rdt(css);
 	ssize_t err = 0;
-	u64 max_mask;
 	u32 closid;
 
-	if (ir == &rdt_root_group)
-		return -EPERM;
-
 	/*
 	 * Need global mutex as cbm write may allocate a closid.
 	 */
 	mutex_lock(&rdt_group_mutex);
 
-	max_mask = (1ULL << max_cbm) - 1;
-	if (cbmvalue & ~max_mask) {
-		err = -EINVAL;
-		goto out;
-	}
-
 	if (cbmvalue == cat_cm_map[ir->closid].cache_mask)
 		goto out;
 
-	err = cbm_validate(ir, cbmvalue);
+	err = cache_mask_validate(ir, cbmvalue);
 	if (err)
 		goto out;
 
@@ -567,6 +585,133 @@ out:
 	return err;
 }
 
+static void cdp_clos_map_dump(void)
+{
+	u32 i;
+
+	pr_debug("CBMMAP\n");
+	for (i = 0; i < boot_cpu_data.x86_cache_max_closid; i++) {
+		pr_debug("dcache_mask: 0x%x, icache_mask=0x%x,clos_refcnt:%u\n",
+			(unsigned int)cdp_cm_map[i].dcache_mask,
+			(unsigned int)cdp_cm_map[i].icache_mask,
+			cdp_cm_map[i].clos_refcnt);
+	}
+}
+
+static bool cdp_mask_search(unsigned long dcache_mask,
+			    unsigned long icache_mask, u32 *closid)
+{
+	u32 maxid = boot_cpu_data.x86_cache_max_closid;
+	unsigned long dcm, icm;
+	u32 i;
+
+	for (i = 0; i < maxid; i++) {
+		dcm = cdp_cm_map[i].dcache_mask;
+		icm = cdp_cm_map[i].icache_mask;
+		if (bitmap_equal(&dcache_mask, &dcm, MAX_CBM_LENGTH)
+		    && bitmap_equal(&icache_mask, &icm, MAX_CBM_LENGTH)) {
+			*closid = i;
+			return true;
+		}
+	}
+
+	return false;
+}
+
+static int cdp_mask_write(struct intel_rdt *ir, u64 dcache_mask,
+			  u64 icache_mask)
+{
+	int err = 0;
+	u32 closid;
+
+	/*
+	 * Try to get a reference for a different CLOSid and release the
+	 * reference to the current CLOSid.
+	 * Need to put down the reference here and get it back in case we
+	 * run out of closids. Otherwise we run into a problem when
+	 * we could be using the last closid that could have been available.
+	 */
+	closid_put(ir->closid);
+	if (cdp_mask_search(dcache_mask, icache_mask, &closid)) {
+		ir->closid = closid;
+		closid_get(closid);
+	} else {
+		err = closid_alloc(ir);
+		if (err) {
+			closid_get(ir->closid);
+			goto out;
+		}
+		closid = ir->closid;
+		cdp_cm_map[closid].dcache_mask = dcache_mask;
+		cdp_cm_map[closid].icache_mask = icache_mask;
+		msr_update_all(DCACHE_MASK_INDEX(closid), dcache_mask);
+		msr_update_all(ICACHE_MASK_INDEX(closid), icache_mask);
+	}
+out:
+
+	return err;
+}
+
+static int cdp_dcache_write(struct cgroup_subsys_state *css,
+				 struct cftype *cft, u64 new_dcache_mask)
+{
+	unsigned long curr_icache_mask, curr_dcache_mask;
+	struct intel_rdt *ir = css_rdt(css);
+	int err = 0;
+
+	/*
+	 * Need global mutex as cache mask write may allocate a closid.
+	 */
+	mutex_lock(&rdt_group_mutex);
+
+	curr_dcache_mask = cdp_cm_map[ir->closid].dcache_mask;
+	curr_icache_mask = cdp_cm_map[ir->closid].icache_mask;
+
+	if (new_dcache_mask == curr_dcache_mask)
+		goto out;
+
+	err = cache_mask_validate(ir, new_dcache_mask);
+	if (err)
+		goto out;
+
+	err = cdp_mask_write(ir, new_dcache_mask, curr_icache_mask);
+	cdp_clos_map_dump();
+out:
+	mutex_unlock(&rdt_group_mutex);
+
+	return err;
+}
+
+static int cdp_icache_write(struct cgroup_subsys_state *css,
+				 struct cftype *cft, u64 new_icache_mask)
+{
+	unsigned long curr_icache_mask, curr_dcache_mask;
+	struct intel_rdt *ir = css_rdt(css);
+	int err = 0;
+
+	/*
+	 * Need global mutex as cache mask write may allocate a closid.
+	 */
+	mutex_lock(&rdt_group_mutex);
+
+	curr_dcache_mask = cdp_cm_map[ir->closid].dcache_mask;
+	curr_icache_mask = cdp_cm_map[ir->closid].icache_mask;
+
+	if (new_icache_mask == curr_icache_mask)
+		goto out;
+
+	err = cache_mask_validate(ir, new_icache_mask);
+	if (err)
+		goto out;
+
+	err = cdp_mask_write(ir, curr_dcache_mask, new_icache_mask);
+	cdp_clos_map_dump();
+out:
+	mutex_unlock(&rdt_group_mutex);
+
+	return err;
+}
+
 static inline bool rdt_cpumask_update(int cpu)
 {
 	static cpumask_t tmp;
@@ -722,10 +867,14 @@ static struct cftype rdt_files[] = {
 	{
 		.name 		= "icache_mask",
 		.flags 		= CFTYPE_ONLY_ON_ROOT | CFTYPE_NOT_ON_ROOT,
+		.seq_show 	= cdp_icache_read,
+		.write_u64 	= cdp_icache_write,
 	},
 	{
 		.name		= "dcache_mask",
 		.flags		= CFTYPE_ONLY_ON_ROOT | CFTYPE_NOT_ON_ROOT,
+		.seq_show 	= cdp_dcache_read,
+		.write_u64 	= cdp_dcache_write,
 	},
 	{ }	/* terminate */
 };
-- 
1.9.1