s390/pci: provide support for MIO instructions

Provide support for PCI I/O instructions that work on mapped IO addresses.
Signed-off-by: Sebastian Ott <sebott@linux.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/io.h

@@ -30,14 +30,8 @@ void unxlate_dev_mem_ptr(phys_addr_t phys, void *addr);
 #define ioremap_wc			ioremap_nocache
 #define ioremap_wt			ioremap_nocache
 
-static inline void __iomem *ioremap(unsigned long offset, unsigned long size)
-{
-	return (void __iomem *) offset;
-}
-
-static inline void iounmap(volatile void __iomem *addr)
-{
-}
+void __iomem *ioremap(unsigned long offset, unsigned long size);
+void iounmap(volatile void __iomem *addr);
 
 static inline void __iomem *ioport_map(unsigned long port, unsigned int nr)
 {
@@ -57,14 +51,17 @@ static inline void ioport_unmap(void __iomem *p)
  * the corresponding device and create the mapping cookie.
  */
 #define pci_iomap pci_iomap
+#define pci_iomap_range pci_iomap_range
 #define pci_iounmap pci_iounmap
-#define pci_iomap_wc pci_iomap
-#define pci_iomap_wc_range pci_iomap_range
+#define pci_iomap_wc pci_iomap_wc
+#define pci_iomap_wc_range pci_iomap_wc_range
 
 #define memcpy_fromio(dst, src, count)	zpci_memcpy_fromio(dst, src, count)
 #define memcpy_toio(dst, src, count)	zpci_memcpy_toio(dst, src, count)
 #define memset_io(dst, val, count)	zpci_memset_io(dst, val, count)
 
+#define mmiowb()	zpci_barrier()
+
 #define __raw_readb	zpci_read_u8
 #define __raw_readw	zpci_read_u16
 #define __raw_readl	zpci_read_u32

arch/s390/include/asm/pci.h

@@ -86,6 +86,8 @@ enum zpci_state {
 
 struct zpci_bar_struct {
 	struct resource *res;		/* bus resource */
+	void __iomem	*mio_wb;
+	void __iomem	*mio_wt;
 	u32		val;		/* bar start & 3 flag bits */
 	u16		map_idx;	/* index into bar mapping array */
 	u8		size;		/* order 2 exponent */
@@ -135,6 +137,7 @@ struct zpci_dev {
 	struct iommu_device iommu_dev;  /* IOMMU core handle */
 
 	char res_name[16];
+	bool mio_capable;
 	struct zpci_bar_struct bars[PCI_BAR_COUNT];
 
 	u64		start_dma;	/* Start of available DMA addresses */

arch/s390/include/asm/pci_clp.h

@@ -43,6 +43,8 @@ struct clp_fh_list_entry {
 
 #define CLP_SET_ENABLE_PCI_FN	0	/* Yes, 0 enables it */
 #define CLP_SET_DISABLE_PCI_FN	1	/* Yes, 1 disables it */
+#define CLP_SET_ENABLE_MIO	2
+#define CLP_SET_DISABLE_MIO	3
 
 #define CLP_UTIL_STR_LEN	64
 #define CLP_PFIP_NR_SEGMENTS	4
@@ -80,7 +82,8 @@ struct clp_req_query_pci {
 struct clp_rsp_query_pci {
 	struct clp_rsp_hdr hdr;
 	u16 vfn;			/* virtual fn number */
-	u16			:  7;
+	u16			:  6;
+	u16 mio_addr_avail	:  1;
 	u16 util_str_avail	:  1;	/* utility string available? */
 	u16 pfgid		:  8;	/* pci function group id */
 	u32 fid;			/* pci function id */
@@ -96,6 +99,15 @@ struct clp_rsp_query_pci {
 	u32 reserved[11];
 	u32 uid;			/* user defined id */
 	u8 util_str[CLP_UTIL_STR_LEN];	/* utility string */
+	u32 reserved2[16];
+	u32 mio_valid : 6;
+	u32 : 26;
+	u32 : 32;
+	struct {
+		u64 wb;
+		u64 wt;
+	} addr[PCI_BAR_COUNT];
+	u32 reserved3[6];
 } __packed;
 
 /* Query PCI function group request */

arch/s390/include/asm/pci_insn.h

@@ -2,6 +2,8 @@
 #ifndef _ASM_S390_PCI_INSN_H
 #define _ASM_S390_PCI_INSN_H
 
+#include <linux/jump_label.h>
+
 /* Load/Store status codes */
 #define ZPCI_PCI_ST_FUNC_NOT_ENABLED		4
 #define ZPCI_PCI_ST_FUNC_IN_ERR			8
@@ -122,6 +124,8 @@ union zpci_sic_iib {
 	struct zpci_cdiib cdiib;
 };
 
+DECLARE_STATIC_KEY_FALSE(have_mio);
+
 u8 zpci_mod_fc(u64 req, struct zpci_fib *fib, u8 *status);
 int zpci_refresh_trans(u64 fn, u64 addr, u64 range);
 int __zpci_load(u64 *data, u64 req, u64 offset);
@@ -129,6 +133,7 @@ int zpci_load(u64 *data, const volatile void __iomem *addr, unsigned long len);
 int __zpci_store(u64 data, u64 req, u64 offset);
 int zpci_store(const volatile void __iomem *addr, u64 data, unsigned long len);
 int __zpci_store_block(const u64 *data, u64 req, u64 offset);
+void zpci_barrier(void);
 int __zpci_set_irq_ctrl(u16 ctl, u8 isc, union zpci_sic_iib *iib);
 
 static inline int zpci_set_irq_ctrl(u16 ctl, u8 isc)

arch/s390/pci/pci.c

@@ -25,6 +25,7 @@
 #include <linux/export.h>
 #include <linux/delay.h>
 #include <linux/seq_file.h>
+#include <linux/jump_label.h>
 #include <linux/pci.h>
 
 #include <asm/isc.h>
@@ -50,6 +51,8 @@ static unsigned long *zpci_iomap_bitmap;
 struct zpci_iomap_entry *zpci_iomap_start;
 EXPORT_SYMBOL_GPL(zpci_iomap_start);
 
+DEFINE_STATIC_KEY_FALSE(have_mio);
+
 static struct kmem_cache *zdev_fmb_cache;
 
 struct zpci_dev *get_zdev_by_fid(u32 fid)
@@ -223,18 +226,48 @@ void __iowrite64_copy(void __iomem *to, const void *from, size_t count)
        zpci_memcpy_toio(to, from, count);
 }
 
+void __iomem *ioremap(unsigned long ioaddr, unsigned long size)
+{
+	struct vm_struct *area;
+	unsigned long offset;
+
+	if (!size)
+		return NULL;
+
+	if (!static_branch_unlikely(&have_mio))
+		return (void __iomem *) ioaddr;
+
+	offset = ioaddr & ~PAGE_MASK;
+	ioaddr &= PAGE_MASK;
+	size = PAGE_ALIGN(size + offset);
+	area = get_vm_area(size, VM_IOREMAP);
+	if (!area)
+		return NULL;
+
+	if (ioremap_page_range((unsigned long) area->addr,
+			       (unsigned long) area->addr + size,
+			       ioaddr, PAGE_KERNEL)) {
+		vunmap(area->addr);
+		return NULL;
+	}
+	return (void __iomem *) ((unsigned long) area->addr + offset);
+}
+EXPORT_SYMBOL(ioremap);
+
+void iounmap(volatile void __iomem *addr)
+{
+	if (static_branch_likely(&have_mio))
+		vunmap((__force void *) ((unsigned long) addr & PAGE_MASK));
+}
+EXPORT_SYMBOL(iounmap);
+
 /* Create a virtual mapping cookie for a PCI BAR */
-void __iomem *pci_iomap_range(struct pci_dev *pdev,
-			      int bar,
-			      unsigned long offset,
-			      unsigned long max)
+static void __iomem *pci_iomap_range_fh(struct pci_dev *pdev, int bar,
+					unsigned long offset, unsigned long max)
 {
 	struct zpci_dev *zdev =	to_zpci(pdev);
 	int idx;
 
-	if (!pci_resource_len(pdev, bar) || bar >= PCI_BAR_COUNT)
-		return NULL;
-
 	idx = zdev->bars[bar].map_idx;
 	spin_lock(&zpci_iomap_lock);
 	/* Detect overrun */
@@ -245,6 +278,30 @@ void __iomem *pci_iomap_range(struct pci_dev *pdev,
 
 	return (void __iomem *) ZPCI_ADDR(idx) + offset;
 }
+
+static void __iomem *pci_iomap_range_mio(struct pci_dev *pdev, int bar,
+					 unsigned long offset,
+					 unsigned long max)
+{
+	unsigned long barsize = pci_resource_len(pdev, bar);
+	struct zpci_dev *zdev = to_zpci(pdev);
+	void __iomem *iova;
+
+	iova = ioremap((unsigned long) zdev->bars[bar].mio_wt, barsize);
+	return iova ? iova + offset : iova;
+}
+
+void __iomem *pci_iomap_range(struct pci_dev *pdev, int bar,
+			      unsigned long offset, unsigned long max)
+{
+	if (!pci_resource_len(pdev, bar) || bar >= PCI_BAR_COUNT)
+		return NULL;
+
+	if (static_branch_likely(&have_mio))
+		return pci_iomap_range_mio(pdev, bar, offset, max);
+	else
+		return pci_iomap_range_fh(pdev, bar, offset, max);
+}
 EXPORT_SYMBOL(pci_iomap_range);
 
 void __iomem *pci_iomap(struct pci_dev *dev, int bar, unsigned long maxlen)
@@ -253,7 +310,37 @@ void __iomem *pci_iomap(struct pci_dev *dev, int bar, unsigned long maxlen)
 }
 EXPORT_SYMBOL(pci_iomap);
 
-void pci_iounmap(struct pci_dev *pdev, void __iomem *addr)
+static void __iomem *pci_iomap_wc_range_mio(struct pci_dev *pdev, int bar,
+					    unsigned long offset, unsigned long max)
+{
+	unsigned long barsize = pci_resource_len(pdev, bar);
+	struct zpci_dev *zdev = to_zpci(pdev);
+	void __iomem *iova;
+
+	iova = ioremap((unsigned long) zdev->bars[bar].mio_wb, barsize);
+	return iova ? iova + offset : iova;
+}
+
+void __iomem *pci_iomap_wc_range(struct pci_dev *pdev, int bar,
+				 unsigned long offset, unsigned long max)
+{
+	if (!pci_resource_len(pdev, bar) || bar >= PCI_BAR_COUNT)
+		return NULL;
+
+	if (static_branch_likely(&have_mio))
+		return pci_iomap_wc_range_mio(pdev, bar, offset, max);
+	else
+		return pci_iomap_range_fh(pdev, bar, offset, max);
+}
+EXPORT_SYMBOL(pci_iomap_wc_range);
+
+void __iomem *pci_iomap_wc(struct pci_dev *dev, int bar, unsigned long maxlen)
+{
+	return pci_iomap_wc_range(dev, bar, 0, maxlen);
+}
+EXPORT_SYMBOL(pci_iomap_wc);
+
+static void pci_iounmap_fh(struct pci_dev *pdev, void __iomem *addr)
 {
 	unsigned int idx = ZPCI_IDX(addr);
 
@@ -266,6 +353,19 @@ void pci_iounmap(struct pci_dev *pdev, void __iomem *addr)
 	}
 	spin_unlock(&zpci_iomap_lock);
 }
+
+static void pci_iounmap_mio(struct pci_dev *pdev, void __iomem *addr)
+{
+	iounmap(addr);
+}
+
+void pci_iounmap(struct pci_dev *pdev, void __iomem *addr)
+{
+	if (static_branch_likely(&have_mio))
+		pci_iounmap_mio(pdev, addr);
+	else
+		pci_iounmap_fh(pdev, addr);
+}
 EXPORT_SYMBOL(pci_iounmap);
 
 static int pci_read(struct pci_bus *bus, unsigned int devfn, int where,
@@ -312,6 +412,7 @@ static struct resource iov_res = {
 
 static void zpci_map_resources(struct pci_dev *pdev)
 {
+	struct zpci_dev *zdev = to_zpci(pdev);
 	resource_size_t len;
 	int i;
 
@@ -319,8 +420,13 @@ static void zpci_map_resources(struct pci_dev *pdev)
 		len = pci_resource_len(pdev, i);
 		if (!len)
 			continue;
-		pdev->resource[i].start =
-			(resource_size_t __force) pci_iomap(pdev, i, 0);
+
+		if (static_branch_likely(&have_mio))
+			pdev->resource[i].start =
+				(resource_size_t __force) zdev->bars[i].mio_wb;
+		else
+			pdev->resource[i].start =
+				(resource_size_t __force) pci_iomap(pdev, i, 0);
 		pdev->resource[i].end = pdev->resource[i].start + len - 1;
 	}
 
@@ -341,6 +447,9 @@ static void zpci_unmap_resources(struct pci_dev *pdev)
 	resource_size_t len;
 	int i;
 
+	if (static_branch_likely(&have_mio))
+		return;
+
 	for (i = 0; i < PCI_BAR_COUNT; i++) {
 		len = pci_resource_len(pdev, i);
 		if (!len)
@@ -772,6 +881,9 @@ static int __init pci_base_init(void)
 	if (!test_facility(69) || !test_facility(71))
 		return 0;
 
+	if (test_facility(153))
+		static_branch_enable(&have_mio);
+
 	rc = zpci_debug_init();
 	if (rc)
 		goto out;

arch/s390/pci/pci_clp.c

@@ -163,7 +163,14 @@ static int clp_store_query_pci_fn(struct zpci_dev *zdev,
 		memcpy(zdev->util_str, response->util_str,
 		       sizeof(zdev->util_str));
 	}
+	zdev->mio_capable = response->mio_addr_avail;
+	for (i = 0; i < PCI_BAR_COUNT; i++) {
+		if (!(response->mio_valid & (1 << (PCI_BAR_COUNT - i - 1))))
+			continue;
 
+		zdev->bars[i].mio_wb = (void __iomem *) response->addr[i].wb;
+		zdev->bars[i].mio_wt = (void __iomem *) response->addr[i].wt;
+	}
 	return 0;
 }
 
@@ -279,11 +286,18 @@ int clp_enable_fh(struct zpci_dev *zdev, u8 nr_dma_as)
 	int rc;
 
 	rc = clp_set_pci_fn(&fh, nr_dma_as, CLP_SET_ENABLE_PCI_FN);
-	if (!rc)
-		/* Success -> store enabled handle in zdev */
-		zdev->fh = fh;
+	zpci_dbg(3, "ena fid:%x, fh:%x, rc:%d\n", zdev->fid, fh, rc);
+	if (rc)
+		goto out;
 
-	zpci_dbg(3, "ena fid:%x, fh:%x, rc:%d\n", zdev->fid, zdev->fh, rc);
+	zdev->fh = fh;
+	if (zdev->mio_capable) {
+		rc = clp_set_pci_fn(&fh, nr_dma_as, CLP_SET_ENABLE_MIO);
+		zpci_dbg(3, "ena mio fid:%x, fh:%x, rc:%d\n", zdev->fid, fh, rc);
+		if (rc)
+			clp_disable_fh(zdev);
+	}
+out:
 	return rc;
 }
 
@@ -296,11 +310,10 @@ int clp_disable_fh(struct zpci_dev *zdev)
 		return 0;
 
 	rc = clp_set_pci_fn(&fh, 0, CLP_SET_DISABLE_PCI_FN);
+	zpci_dbg(3, "dis fid:%x, fh:%x, rc:%d\n", zdev->fid, fh, rc);
 	if (!rc)
-		/* Success -> store disabled handle in zdev */
 		zdev->fh = fh;
 
-	zpci_dbg(3, "dis fid:%x, fh:%x, rc:%d\n", zdev->fid, zdev->fh, rc);
 	return rc;
 }
 

arch/s390/pci/pci_insn.c

@@ -8,6 +8,7 @@
 #include <linux/export.h>
 #include <linux/errno.h>
 #include <linux/delay.h>
+#include <linux/jump_label.h>
 #include <asm/facility.h>
 #include <asm/pci_insn.h>
 #include <asm/pci_debug.h>
@@ -161,13 +162,50 @@ int __zpci_load(u64 *data, u64 req, u64 offset)
 }
 EXPORT_SYMBOL_GPL(__zpci_load);
 
-int zpci_load(u64 *data, const volatile void __iomem *addr, unsigned long len)
+static inline int zpci_load_fh(u64 *data, const volatile void __iomem *addr,
+			       unsigned long len)
 {
 	struct zpci_iomap_entry *entry = &zpci_iomap_start[ZPCI_IDX(addr)];
 	u64 req = ZPCI_CREATE_REQ(entry->fh, entry->bar, len);
 
 	return __zpci_load(data, req, ZPCI_OFFSET(addr));
 }
+
+static inline int __pcilg_mio(u64 *data, u64 ioaddr, u64 len, u8 *status)
+{
+	register u64 addr asm("2") = ioaddr;
+	register u64 r3 asm("3") = len;
+	int cc = -ENXIO;
+	u64 __data;
+
+	asm volatile (
+		"       .insn   rre,0xb9d60000,%[data],%[ioaddr]\n"
+		"0:     ipm     %[cc]\n"
+		"       srl     %[cc],28\n"
+		"1:\n"
+		EX_TABLE(0b, 1b)
+		: [cc] "+d" (cc), [data] "=d" (__data), "+d" (r3)
+		: [ioaddr] "d" (addr)
+		: "cc");
+	*status = r3 >> 24 & 0xff;
+	*data = __data;
+	return cc;
+}
+
+int zpci_load(u64 *data, const volatile void __iomem *addr, unsigned long len)
+{
+	u8 status;
+	int cc;
+
+	if (!static_branch_unlikely(&have_mio))
+		return zpci_load_fh(data, addr, len);
+
+	cc = __pcilg_mio(data, (__force u64) addr, len, &status);
+	if (cc)
+		zpci_err_insn(cc, status, 0, (__force u64) addr);
+
+	return (cc > 0) ? -EIO : cc;
+}
 EXPORT_SYMBOL_GPL(zpci_load);
 
 /* PCI Store */
@@ -208,13 +246,48 @@ int __zpci_store(u64 data, u64 req, u64 offset)
 }
 EXPORT_SYMBOL_GPL(__zpci_store);
 
-int zpci_store(const volatile void __iomem *addr, u64 data, unsigned long len)
+static inline int zpci_store_fh(const volatile void __iomem *addr, u64 data,
+				unsigned long len)
 {
 	struct zpci_iomap_entry *entry = &zpci_iomap_start[ZPCI_IDX(addr)];
 	u64 req = ZPCI_CREATE_REQ(entry->fh, entry->bar, len);
 
 	return __zpci_store(data, req, ZPCI_OFFSET(addr));
 }
+
+static inline int __pcistg_mio(u64 data, u64 ioaddr, u64 len, u8 *status)
+{
+	register u64 addr asm("2") = ioaddr;
+	register u64 r3 asm("3") = len;
+	int cc = -ENXIO;
+
+	asm volatile (
+		"       .insn   rre,0xb9d40000,%[data],%[ioaddr]\n"
+		"0:     ipm     %[cc]\n"
+		"       srl     %[cc],28\n"
+		"1:\n"
+		EX_TABLE(0b, 1b)
+		: [cc] "+d" (cc), "+d" (r3)
+		: [data] "d" (data), [ioaddr] "d" (addr)
+		: "cc");
+	*status = r3 >> 24 & 0xff;
+	return cc;
+}
+
+int zpci_store(const volatile void __iomem *addr, u64 data, unsigned long len)
+{
+	u8 status;
+	int cc;
+
+	if (!static_branch_unlikely(&have_mio))
+		return zpci_store_fh(addr, data, len);
+
+	cc = __pcistg_mio(data, (__force u64) addr, len, &status);
+	if (cc)
+		zpci_err_insn(cc, status, 0, (__force u64) addr);
+
+	return (cc > 0) ? -EIO : cc;
+}
 EXPORT_SYMBOL_GPL(zpci_store);
 
 /* PCI Store Block */
@@ -253,8 +326,8 @@ int __zpci_store_block(const u64 *data, u64 req, u64 offset)
 }
 EXPORT_SYMBOL_GPL(__zpci_store_block);
 
-int zpci_write_block(volatile void __iomem *dst,
-		     const void *src, unsigned long len)
+static inline int zpci_write_block_fh(volatile void __iomem *dst,
+				      const void *src, unsigned long len)
 {
 	struct zpci_iomap_entry *entry = &zpci_iomap_start[ZPCI_IDX(dst)];
 	u64 req = ZPCI_CREATE_REQ(entry->fh, entry->bar, len);
@@ -262,4 +335,52 @@ int zpci_write_block(volatile void __iomem *dst,
 
 	return __zpci_store_block(src, req, offset);
 }
+
+static inline int __pcistb_mio(const u64 *data, u64 ioaddr, u64 len, u8 *status)
+{
+	int cc = -ENXIO;
+
+	asm volatile (
+		"       .insn   rsy,0xeb00000000d4,%[len],%[ioaddr],%[data]\n"
+		"0:     ipm     %[cc]\n"
+		"       srl     %[cc],28\n"
+		"1:\n"
+		EX_TABLE(0b, 1b)
+		: [cc] "+d" (cc), [len] "+d" (len)
+		: [ioaddr] "d" (ioaddr), [data] "Q" (*data)
+		: "cc");
+	*status = len >> 24 & 0xff;
+	return cc;
+}
+
+int zpci_write_block(volatile void __iomem *dst,
+		     const void *src, unsigned long len)
+{
+	u8 status;
+	int cc;
+
+	if (!static_branch_unlikely(&have_mio))
+		return zpci_write_block_fh(dst, src, len);
+
+	cc = __pcistb_mio(src, (__force u64) dst, len, &status);
+	if (cc)
+		zpci_err_insn(cc, status, 0, (__force u64) dst);
+
+	return (cc > 0) ? -EIO : cc;
+}
 EXPORT_SYMBOL_GPL(zpci_write_block);
+
+static inline void __pciwb_mio(void)
+{
+	unsigned long unused = 0;
+
+	asm volatile (".insn    rre,0xb9d50000,%[op],%[op]\n"
+		      : [op] "+d" (unused));
+}
+
+void zpci_barrier(void)
+{
+	if (static_branch_likely(&have_mio))
+		__pciwb_mio();
+}
+EXPORT_SYMBOL_GPL(zpci_barrier);