s390/pci: performance statistics and debug infrastructure

Add support for reading the PCI function measurement block counters
provided by the hypervisor. Add two s390 debug features, one for
critical errors and one for tracing and provide wrappers to log data.

Signed-off-by: Jan Glauber <jang@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/pci.h

@@ -9,6 +9,7 @@
 #include <asm-generic/pci.h>
 #include <asm-generic/pci-dma-compat.h>
 #include <asm/pci_clp.h>
+#include <asm/pci_debug.h>
 
 #define PCIBIOS_MIN_IO		0x1000
 #define PCIBIOS_MIN_MEM		0x10000000
@@ -33,6 +34,25 @@ int pci_proc_domain(struct pci_bus *);
 #define ZPCI_FC_BLOCKED			0x20
 #define ZPCI_FC_DMA_ENABLED		0x10
 
+struct zpci_fmb {
+	u32 format	:  8;
+	u32 dma_valid	:  1;
+	u32		: 23;
+	u32 samples;
+	u64 last_update;
+	/* hardware counters */
+	u64 ld_ops;
+	u64 st_ops;
+	u64 stb_ops;
+	u64 rpcit_ops;
+	u64 dma_rbytes;
+	u64 dma_wbytes;
+	/* software counters */
+	atomic64_t allocated_pages;
+	atomic64_t mapped_pages;
+	atomic64_t unmapped_pages;
+} __packed __aligned(16);
+
 struct msi_map {
 	unsigned long irq;
 	struct msi_desc *msi;
@@ -92,7 +112,15 @@ struct zpci_dev {
 	u64		end_dma;	/* End of available DMA addresses */
 	u64		dma_mask;	/* DMA address space mask */
 
+	/* Function measurement block */
+	struct zpci_fmb *fmb;
+	u16		fmb_update;	/* update interval */
+
 	enum pci_bus_speed max_bus_speed;
+
+	struct dentry	*debugfs_dev;
+	struct dentry	*debugfs_perf;
+	struct dentry	*debugfs_debug;
 };
 
 struct pci_hp_callback_ops {
@@ -155,4 +183,15 @@ extern struct list_head zpci_list;
 extern struct pci_hp_callback_ops hotplug_ops;
 extern unsigned int pci_probe;
 
+/* FMB */
+int zpci_fmb_enable_device(struct zpci_dev *);
+int zpci_fmb_disable_device(struct zpci_dev *);
+
+/* Debug */
+int zpci_debug_init(void);
+void zpci_debug_exit(void);
+void zpci_debug_init_device(struct zpci_dev *);
+void zpci_debug_exit_device(struct zpci_dev *);
+void zpci_debug_info(struct zpci_dev *, struct seq_file *);
+
 #endif

arch/s390/include/asm/pci_debug.h

@@ -0,0 +1,36 @@
+#ifndef _S390_ASM_PCI_DEBUG_H
+#define _S390_ASM_PCI_DEBUG_H
+
+#include <asm/debug.h>
+
+extern debug_info_t *pci_debug_msg_id;
+extern debug_info_t *pci_debug_err_id;
+
+#ifdef CONFIG_PCI_DEBUG
+#define zpci_dbg(fmt, args...)							\
+	do {									\
+		if (pci_debug_msg_id->level >= 2)				\
+			debug_sprintf_event(pci_debug_msg_id, 2, fmt , ## args);\
+	} while (0)
+
+#else /* !CONFIG_PCI_DEBUG */
+#define zpci_dbg(fmt, args...) do { } while (0)
+#endif
+
+#define zpci_err(text...)							\
+	do {									\
+		char debug_buffer[16];						\
+		snprintf(debug_buffer, 16, text);				\
+		debug_text_event(pci_debug_err_id, 0, debug_buffer);		\
+	} while (0)
+
+static inline void zpci_err_hex(void *addr, int len)
+{
+	while (len > 0) {
+		debug_event(pci_debug_err_id, 0, (void *) addr, len);
+		len -= pci_debug_err_id->buf_size;
+		addr += pci_debug_err_id->buf_size;
+	}
+}
+
+#endif

arch/s390/pci/Makefile

@@ -3,4 +3,4 @@
 #
 
 obj-$(CONFIG_PCI)	+= pci.o pci_dma.o pci_clp.o pci_msi.o \
-			   pci_sysfs.o pci_event.o
+			   pci_sysfs.o pci_event.o pci_debug.o

arch/s390/pci/pci.c

@@ -98,6 +98,10 @@ EXPORT_SYMBOL_GPL(zpci_iomap_start);
 static int __read_mostly aisb_max;
 
 static struct kmem_cache *zdev_irq_cache;
+static struct kmem_cache *zdev_fmb_cache;
+
+debug_info_t *pci_debug_msg_id;
+debug_info_t *pci_debug_err_id;
 
 static inline int irq_to_msi_nr(unsigned int irq)
 {
@@ -216,6 +220,7 @@ struct mod_pci_args {
 	u64 base;
 	u64 limit;
 	u64 iota;
+	u64 fmb_addr;
 };
 
 static int mod_pci(struct zpci_dev *zdev, int fn, u8 dmaas, struct mod_pci_args *args)
@@ -232,6 +237,7 @@ static int mod_pci(struct zpci_dev *zdev, int fn, u8 dmaas, struct mod_pci_args
 	fib->pba = args->base;
 	fib->pal = args->limit;
 	fib->iota = args->iota;
+	fib->fmb_addr = args->fmb_addr;
 
 	rc = mpcifc_instr(req, fib);
 	free_page((unsigned long) fib);
@@ -242,7 +248,7 @@ static int mod_pci(struct zpci_dev *zdev, int fn, u8 dmaas, struct mod_pci_args
 int zpci_register_ioat(struct zpci_dev *zdev, u8 dmaas,
 		       u64 base, u64 limit, u64 iota)
 {
-	struct mod_pci_args args = { base, limit, iota };
+	struct mod_pci_args args = { base, limit, iota, 0 };
 
 	WARN_ON_ONCE(iota & 0x3fff);
 	args.iota |= ZPCI_IOTA_RTTO_FLAG;
@@ -252,7 +258,7 @@ int zpci_register_ioat(struct zpci_dev *zdev, u8 dmaas,
 /* Modify PCI: Unregister I/O address translation parameters */
 int zpci_unregister_ioat(struct zpci_dev *zdev, u8 dmaas)
 {
-	struct mod_pci_args args = { 0, 0, 0 };
+	struct mod_pci_args args = { 0, 0, 0, 0 };
 
 	return mod_pci(zdev, ZPCI_MOD_FC_DEREG_IOAT, dmaas, &args);
 }
@@ -260,11 +266,46 @@ int zpci_unregister_ioat(struct zpci_dev *zdev, u8 dmaas)
 /* Modify PCI: Unregister adapter interruptions */
 static int zpci_unregister_airq(struct zpci_dev *zdev)
 {
-	struct mod_pci_args args = { 0, 0, 0 };
+	struct mod_pci_args args = { 0, 0, 0, 0 };
 
 	return mod_pci(zdev, ZPCI_MOD_FC_DEREG_INT, 0, &args);
 }
 
+/* Modify PCI: Set PCI function measurement parameters */
+int zpci_fmb_enable_device(struct zpci_dev *zdev)
+{
+	struct mod_pci_args args = { 0, 0, 0, 0 };
+
+	if (zdev->fmb)
+		return -EINVAL;
+
+	zdev->fmb = kmem_cache_alloc(zdev_fmb_cache, GFP_KERNEL);
+	if (!zdev->fmb)
+		return -ENOMEM;
+	memset(zdev->fmb, 0, sizeof(*zdev->fmb));
+	WARN_ON((u64) zdev->fmb & 0xf);
+
+	args.fmb_addr = virt_to_phys(zdev->fmb);
+	return mod_pci(zdev, ZPCI_MOD_FC_SET_MEASURE, 0, &args);
+}
+
+/* Modify PCI: Disable PCI function measurement */
+int zpci_fmb_disable_device(struct zpci_dev *zdev)
+{
+	struct mod_pci_args args = { 0, 0, 0, 0 };
+	int rc;
+
+	if (!zdev->fmb)
+		return -EINVAL;
+
+	/* Function measurement is disabled if fmb address is zero */
+	rc = mod_pci(zdev, ZPCI_MOD_FC_SET_MEASURE, 0, &args);
+
+	kmem_cache_free(zdev_fmb_cache, zdev->fmb);
+	zdev->fmb = NULL;
+	return rc;
+}
+
 #define ZPCI_PCIAS_CFGSPC	15
 
 static int zpci_cfg_load(struct zpci_dev *zdev, int offset, u32 *val, u8 len)
@@ -633,6 +674,7 @@ static void zpci_remove_device(struct pci_dev *pdev)
 	dev_info(&pdev->dev, "Removing device %u\n", zdev->domain);
 	zdev->state = ZPCI_FN_STATE_CONFIGURED;
 	zpci_dma_exit_device(zdev);
+	zpci_fmb_disable_device(zdev);
 	zpci_sysfs_remove_device(&pdev->dev);
 	zpci_unmap_resources(pdev);
 	list_del(&zdev->entry);		/* can be called from init */
@@ -799,6 +841,16 @@ static void zpci_irq_exit(void)
 	kfree(bucket);
 }
 
+void zpci_debug_info(struct zpci_dev *zdev, struct seq_file *m)
+{
+	if (!zdev)
+		return;
+
+	seq_printf(m, "global irq retries: %u\n", atomic_read(&irq_retries));
+	seq_printf(m, "aibv[0]:%016lx  aibv[1]:%016lx  aisb:%016lx\n",
+		   get_imap(0)->aibv, get_imap(1)->aibv, *bucket->aisb);
+}
+
 static struct resource *zpci_alloc_bus_resource(unsigned long start, unsigned long size,
 						unsigned long flags, int domain)
 {
@@ -994,6 +1046,8 @@ int zpci_scan_device(struct zpci_dev *zdev)
 		goto out;
 	}
 
+	zpci_debug_init_device(zdev);
+	zpci_fmb_enable_device(zdev);
 	zpci_map_resources(zdev);
 	pci_bus_add_devices(zdev->bus);
 
@@ -1020,6 +1074,11 @@ static int zpci_mem_init(void)
 	if (!zdev_irq_cache)
 		goto error_zdev;
 
+	zdev_fmb_cache = kmem_cache_create("PCI_FMB_cache", sizeof(struct zpci_fmb),
+				16, 0, NULL);
+	if (!zdev_fmb_cache)
+		goto error_fmb;
+
 	/* TODO: use realloc */
 	zpci_iomap_start = kzalloc(ZPCI_IOMAP_MAX_ENTRIES * sizeof(*zpci_iomap_start),
 				   GFP_KERNEL);
@@ -1028,6 +1087,8 @@ static int zpci_mem_init(void)
 	return 0;
 
 error_iomap:
+	kmem_cache_destroy(zdev_fmb_cache);
+error_fmb:
 	kmem_cache_destroy(zdev_irq_cache);
 error_zdev:
 	return -ENOMEM;
@@ -1037,6 +1098,7 @@ static void zpci_mem_exit(void)
 {
 	kfree(zpci_iomap_start);
 	kmem_cache_destroy(zdev_irq_cache);
+	kmem_cache_destroy(zdev_fmb_cache);
 }
 
 unsigned int pci_probe = 1;
@@ -1066,6 +1128,10 @@ static int __init pci_base_init(void)
 		test_facility(69), test_facility(70),
 		test_facility(71));
 
+	rc = zpci_debug_init();
+	if (rc)
+		return rc;
+
 	rc = zpci_mem_init();
 	if (rc)
 		goto out_mem;
@@ -1098,6 +1164,7 @@ out_irq:
 out_hash:
 	zpci_mem_exit();
 out_mem:
+	zpci_debug_exit();
 	return rc;
 }
 subsys_initcall(pci_base_init);

arch/s390/pci/pci_clp.c

@@ -51,6 +51,7 @@ static void clp_store_query_pci_fngrp(struct zpci_dev *zdev,
 	zdev->tlb_refresh = response->refresh;
 	zdev->dma_mask = response->dasm;
 	zdev->msi_addr = response->msia;
+	zdev->fmb_update = response->mui;
 
 	pr_debug("Supported number of MSI vectors: %u\n", response->noi);
 	switch (response->version) {

arch/s390/pci/pci_debug.c

@@ -0,0 +1,193 @@
+/*
+ *  Copyright IBM Corp. 2012
+ *
+ *  Author(s):
+ *    Jan Glauber <jang@linux.vnet.ibm.com>
+ */
+
+#define COMPONENT "zPCI"
+#define pr_fmt(fmt) COMPONENT ": " fmt
+
+#include <linux/kernel.h>
+#include <linux/seq_file.h>
+#include <linux/debugfs.h>
+#include <linux/pci.h>
+#include <asm/debug.h>
+
+#include <asm/pci_dma.h>
+
+static struct dentry *debugfs_root;
+
+static char *pci_perf_names[] = {
+	/* hardware counters */
+	"Load operations",
+	"Store operations",
+	"Store block operations",
+	"Refresh operations",
+	"DMA read bytes",
+	"DMA write bytes",
+	/* software counters */
+	"Allocated pages",
+	"Mapped pages",
+	"Unmapped pages",
+};
+
+static int pci_perf_show(struct seq_file *m, void *v)
+{
+	struct zpci_dev *zdev = m->private;
+	u64 *stat;
+	int i;
+
+	if (!zdev)
+		return 0;
+	if (!zdev->fmb)
+		return seq_printf(m, "FMB statistics disabled\n");
+
+	/* header */
+	seq_printf(m, "FMB @ %p\n", zdev->fmb);
+	seq_printf(m, "Update interval: %u ms\n", zdev->fmb_update);
+	seq_printf(m, "Samples: %u\n", zdev->fmb->samples);
+	seq_printf(m, "Last update TOD: %Lx\n", zdev->fmb->last_update);
+
+	/* hardware counters */
+	stat = (u64 *) &zdev->fmb->ld_ops;
+	for (i = 0; i < 4; i++)
+		seq_printf(m, "%26s:\t%llu\n",
+			   pci_perf_names[i], *(stat + i));
+	if (zdev->fmb->dma_valid)
+		for (i = 4; i < 6; i++)
+			seq_printf(m, "%26s:\t%llu\n",
+				   pci_perf_names[i], *(stat + i));
+	/* software counters */
+	for (i = 6; i < ARRAY_SIZE(pci_perf_names); i++)
+		seq_printf(m, "%26s:\t%llu\n",
+			   pci_perf_names[i],
+			   atomic64_read((atomic64_t *) (stat + i)));
+
+	return 0;
+}
+
+static ssize_t pci_perf_seq_write(struct file *file, const char __user *ubuf,
+				  size_t count, loff_t *off)
+{
+	struct zpci_dev *zdev = ((struct seq_file *) file->private_data)->private;
+	unsigned long val;
+	int rc;
+
+	if (!zdev)
+		return 0;
+
+	rc = kstrtoul_from_user(ubuf, count, 10, &val);
+	if (rc)
+		return rc;
+
+	switch (val) {
+	case 0:
+		rc = zpci_fmb_disable_device(zdev);
+		if (rc)
+			return rc;
+		break;
+	case 1:
+		rc = zpci_fmb_enable_device(zdev);
+		if (rc)
+			return rc;
+		break;
+	}
+	return count;
+}
+
+static int pci_perf_seq_open(struct inode *inode, struct file *filp)
+{
+	return single_open(filp, pci_perf_show,
+			   filp->f_path.dentry->d_inode->i_private);
+}
+
+static const struct file_operations debugfs_pci_perf_fops = {
+	.open	 = pci_perf_seq_open,
+	.read	 = seq_read,
+	.write	 = pci_perf_seq_write,
+	.llseek  = seq_lseek,
+	.release = single_release,
+};
+
+static int pci_debug_show(struct seq_file *m, void *v)
+{
+	struct zpci_dev *zdev = m->private;
+
+	zpci_debug_info(zdev, m);
+	return 0;
+}
+
+static int pci_debug_seq_open(struct inode *inode, struct file *filp)
+{
+	return single_open(filp, pci_debug_show,
+			   filp->f_path.dentry->d_inode->i_private);
+}
+
+static const struct file_operations debugfs_pci_debug_fops = {
+	.open	 = pci_debug_seq_open,
+	.read	 = seq_read,
+	.llseek  = seq_lseek,
+	.release = single_release,
+};
+
+void zpci_debug_init_device(struct zpci_dev *zdev)
+{
+	zdev->debugfs_dev = debugfs_create_dir(dev_name(&zdev->pdev->dev),
+					       debugfs_root);
+	if (IS_ERR(zdev->debugfs_dev))
+		zdev->debugfs_dev = NULL;
+
+	zdev->debugfs_perf = debugfs_create_file("statistics",
+				S_IFREG | S_IRUGO | S_IWUSR,
+				zdev->debugfs_dev, zdev,
+				&debugfs_pci_perf_fops);
+	if (IS_ERR(zdev->debugfs_perf))
+		zdev->debugfs_perf = NULL;
+
+	zdev->debugfs_debug = debugfs_create_file("debug",
+				S_IFREG | S_IRUGO | S_IWUSR,
+				zdev->debugfs_dev, zdev,
+				&debugfs_pci_debug_fops);
+	if (IS_ERR(zdev->debugfs_debug))
+		zdev->debugfs_debug = NULL;
+}
+
+void zpci_debug_exit_device(struct zpci_dev *zdev)
+{
+	debugfs_remove(zdev->debugfs_perf);
+	debugfs_remove(zdev->debugfs_debug);
+	debugfs_remove(zdev->debugfs_dev);
+}
+
+int __init zpci_debug_init(void)
+{
+	/* event trace buffer */
+	pci_debug_msg_id = debug_register("pci_msg", 16, 1, 16 * sizeof(long));
+	if (!pci_debug_msg_id)
+		return -EINVAL;
+	debug_register_view(pci_debug_msg_id, &debug_sprintf_view);
+	debug_set_level(pci_debug_msg_id, 3);
+	zpci_dbg("Debug view initialized\n");
+
+	/* error log */
+	pci_debug_err_id = debug_register("pci_error", 2, 1, 16);
+	if (!pci_debug_err_id)
+		return -EINVAL;
+	debug_register_view(pci_debug_err_id, &debug_hex_ascii_view);
+	debug_set_level(pci_debug_err_id, 6);
+	zpci_err("Debug view initialized\n");
+
+	debugfs_root = debugfs_create_dir("pci", NULL);
+	return 0;
+}
+
+void zpci_debug_exit(void)
+{
+	if (pci_debug_msg_id)
+		debug_unregister(pci_debug_msg_id);
+	if (pci_debug_err_id)
+		debug_unregister(pci_debug_err_id);
+
+	debugfs_remove(debugfs_root);
+}

arch/s390/pci/pci_dma.c

@@ -291,8 +291,10 @@ static dma_addr_t s390_dma_map_pages(struct device *dev, struct page *page,
 	if (direction == DMA_NONE || direction == DMA_TO_DEVICE)
 		flags |= ZPCI_TABLE_PROTECTED;
 
-	if (!dma_update_trans(zdev, pa, dma_addr, size, flags))
+	if (!dma_update_trans(zdev, pa, dma_addr, size, flags)) {
+		atomic64_add(nr_pages, (atomic64_t *) &zdev->fmb->mapped_pages);
 		return dma_addr + offset;
+	}
 
 out_free:
 	dma_free_iommu(zdev, iommu_page_index, nr_pages);
@@ -315,6 +317,7 @@ static void s390_dma_unmap_pages(struct device *dev, dma_addr_t dma_addr,
 			     ZPCI_TABLE_PROTECTED | ZPCI_PTE_INVALID))
 		dev_err(dev, "Failed to unmap addr: %Lx\n", dma_addr);
 
+	atomic64_add(npages, (atomic64_t *) &zdev->fmb->unmapped_pages);
 	iommu_page_index = (dma_addr - zdev->start_dma) >> PAGE_SHIFT;
 	dma_free_iommu(zdev, iommu_page_index, npages);
 }
@@ -323,6 +326,7 @@ static void *s390_dma_alloc(struct device *dev, size_t size,
 			    dma_addr_t *dma_handle, gfp_t flag,
 			    struct dma_attrs *attrs)
 {
+	struct zpci_dev *zdev = get_zdev(container_of(dev, struct pci_dev, dev));
 	struct page *page;
 	unsigned long pa;
 	dma_addr_t map;
@@ -331,6 +335,8 @@ static void *s390_dma_alloc(struct device *dev, size_t size,
 	page = alloc_pages(flag, get_order(size));
 	if (!page)
 		return NULL;
+
+	atomic64_add(size / PAGE_SIZE, (atomic64_t *) &zdev->fmb->allocated_pages);
 	pa = page_to_phys(page);
 	memset((void *) pa, 0, size);
 

arch/s390/pci/pci_event.c

@@ -45,6 +45,8 @@ static void zpci_event_log_err(struct zpci_ccdf_err *ccdf)
 {
 	struct zpci_dev *zdev = get_zdev_by_fid(ccdf->fid);
 
+	zpci_err("SEI error CCD:\n");
+	zpci_err_hex(ccdf, sizeof(*ccdf));
 	dev_err(&zdev->pdev->dev, "event code: 0x%x\n", ccdf->pec);
 }