LoongArch: Use TLB for ioremap()

We can support more cache attributes (e.g., CC, SUC and WUC) and page
protection when we use TLB for ioremap(). The implementation is based
on GENERIC_IOREMAP.

The existing simple ioremap() implementation has better performance so
we keep it and introduce ARCH_IOREMAP to control the selection.

We move pagetable_init() earlier to make early ioremap() works, and we
modify the PCI ecam mapping because the TLB-based version of ioremap()
will actually take the size into account.
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>

arch/loongarch/Kconfig

@@ -62,6 +62,7 @@ config LOONGARCH
 	select GENERIC_CPU_AUTOPROBE
 	select GENERIC_ENTRY
 	select GENERIC_GETTIMEOFDAY
+	select GENERIC_IOREMAP if !ARCH_IOREMAP
 	select GENERIC_IRQ_MULTI_HANDLER
 	select GENERIC_IRQ_PROBE
 	select GENERIC_IRQ_SHOW
@@ -168,6 +169,9 @@ config MACH_LOONGSON32
 config MACH_LOONGSON64
 	def_bool 64BIT
 
+config FIX_EARLYCON_MEM
+	def_bool y
+
 config PAGE_SIZE_4KB
 	bool
 
@@ -404,6 +408,13 @@ config FORCE_MAX_ZONEORDER
 	  The page size is not necessarily 4KB.  Keep this in mind
 	  when choosing a value for this option.
 
+config ARCH_IOREMAP
+	bool "Enable LoongArch DMW-based ioremap()"
+	help
+	  We use generic TLB-based ioremap() by default since it has page
+	  protection support. However, you can enable LoongArch DMW-based
+	  ioremap() for better performance.
+
 config SECCOMP
 	bool "Enable seccomp to safely compute untrusted bytecode"
 	depends on PROC_FS

arch/loongarch/include/asm/fixmap.h

@@ -10,4 +10,19 @@
 
 #define NR_FIX_BTMAPS 64
 
+enum fixed_addresses {
+	FIX_HOLE,
+	FIX_EARLYCON_MEM_BASE,
+	__end_of_fixed_addresses
+};
+
+#define FIXADDR_SIZE	(__end_of_fixed_addresses << PAGE_SHIFT)
+#define FIXADDR_START	(FIXADDR_TOP - FIXADDR_SIZE)
+#define FIXMAP_PAGE_IO	PAGE_KERNEL_SUC
+
+extern void __set_fixmap(enum fixed_addresses idx,
+			 phys_addr_t phys, pgprot_t flags);
+
+#include <asm-generic/fixmap.h>
+
 #endif

arch/loongarch/include/asm/io.h

@@ -27,71 +27,38 @@ extern void __init early_iounmap(void __iomem *addr, unsigned long size);
 #define early_memremap early_ioremap
 #define early_memunmap early_iounmap
 
+#ifdef CONFIG_ARCH_IOREMAP
+
 static inline void __iomem *ioremap_prot(phys_addr_t offset, unsigned long size,
 					 unsigned long prot_val)
 {
-	if (prot_val == _CACHE_CC)
+	if (prot_val & _CACHE_CC)
 		return (void __iomem *)(unsigned long)(CACHE_BASE + offset);
 	else
 		return (void __iomem *)(unsigned long)(UNCACHE_BASE + offset);
 }
 
-/*
- * ioremap -   map bus memory into CPU space
- * @offset:    bus address of the memory
- * @size:      size of the resource to map
- *
- * ioremap performs a platform specific sequence of operations to
- * make bus memory CPU accessible via the readb/readw/readl/writeb/
- * writew/writel functions and the other mmio helpers. The returned
- * address is not guaranteed to be usable directly as a virtual
- * address.
- */
 #define ioremap(offset, size)		\
-	ioremap_prot((offset), (size), _CACHE_SUC)
+	ioremap_prot((offset), (size), pgprot_val(PAGE_KERNEL_SUC))
+
+#define iounmap(addr) 			((void)(addr))
+
+#endif
 
 /*
- * ioremap_wc - map bus memory into CPU space
+ * On LoongArch, ioremap() has two variants, ioremap_wc() and ioremap_cache().
+ * They map bus memory into CPU space, the mapped memory is marked uncachable
+ * (_CACHE_SUC), uncachable but accelerated by write-combine (_CACHE_WUC) and
+ * cachable (_CACHE_CC) respectively for CPU access.
+ *
  * @offset:    bus address of the memory
  * @size:      size of the resource to map
- *
- * ioremap_wc performs a platform specific sequence of operations to
- * make bus memory CPU accessible via the readb/readw/readl/writeb/
- * writew/writel functions and the other mmio helpers. The returned
- * address is not guaranteed to be usable directly as a virtual
- * address.
- *
- * This version of ioremap ensures that the memory is marked uncachable
- * but accelerated by means of write-combining feature. It is specifically
- * useful for PCIe prefetchable windows, which may vastly improve a
- * communications performance. If it was determined on boot stage, what
- * CPU CCA doesn't support WUC, the method shall fall-back to the
- * _CACHE_SUC option (see cpu_probe() method).
  */
 #define ioremap_wc(offset, size)	\
-	ioremap_prot((offset), (size), _CACHE_WUC)
+	ioremap_prot((offset), (size), pgprot_val(PAGE_KERNEL_WUC))
 
-/*
- * ioremap_cache -  map bus memory into CPU space
- * @offset:	    bus address of the memory
- * @size:	    size of the resource to map
- *
- * ioremap_cache performs a platform specific sequence of operations to
- * make bus memory CPU accessible via the readb/readw/readl/writeb/
- * writew/writel functions and the other mmio helpers. The returned
- * address is not guaranteed to be usable directly as a virtual
- * address.
- *
- * This version of ioremap ensures that the memory is marked cachable by
- * the CPU.  Also enables full write-combining.	 Useful for some
- * memory-like regions on I/O busses.
- */
 #define ioremap_cache(offset, size)	\
-	ioremap_prot((offset), (size), _CACHE_CC)
-
-static inline void iounmap(const volatile void __iomem *addr)
-{
-}
+	ioremap_prot((offset), (size), pgprot_val(PAGE_KERNEL))
 
 #define mmiowb() asm volatile ("dbar 0" ::: "memory")
 

arch/loongarch/include/asm/pgtable-bits.h

@@ -83,8 +83,11 @@
 				 _PAGE_GLOBAL | _PAGE_KERN |  _CACHE_SUC)
 #define PAGE_KERNEL_WUC __pgprot(_PAGE_PRESENT | __READABLE | __WRITEABLE | \
 				 _PAGE_GLOBAL | _PAGE_KERN |  _CACHE_WUC)
+
 #ifndef __ASSEMBLY__
 
+#define _PAGE_IOREMAP		pgprot_val(PAGE_KERNEL_SUC)
+
 #define pgprot_noncached pgprot_noncached
 
 static inline pgprot_t pgprot_noncached(pgprot_t _prot)

arch/loongarch/kernel/setup.c

@@ -348,10 +348,10 @@ void __init setup_arch(char **cmdline_p)
 	init_environ();
 	efi_init();
 	memblock_init();
+	pagetable_init();
 	parse_early_param();
 
 	platform_init();
-	pagetable_init();
 	arch_mem_init(cmdline_p);
 
 	resource_init();

arch/loongarch/mm/init.c

@@ -152,6 +152,70 @@ EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 #endif
 #endif
 
+static pte_t *fixmap_pte(unsigned long addr)
+{
+	pgd_t *pgd;
+	p4d_t *p4d;
+	pud_t *pud;
+	pmd_t *pmd;
+
+	pgd = pgd_offset_k(addr);
+	p4d = p4d_offset(pgd, addr);
+
+	if (pgd_none(*pgd)) {
+		pud_t *new __maybe_unused;
+
+		new = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
+		pgd_populate(&init_mm, pgd, new);
+#ifndef __PAGETABLE_PUD_FOLDED
+		pud_init((unsigned long)new, (unsigned long)invalid_pmd_table);
+#endif
+	}
+
+	pud = pud_offset(p4d, addr);
+	if (pud_none(*pud)) {
+		pmd_t *new __maybe_unused;
+
+		new = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
+		pud_populate(&init_mm, pud, new);
+#ifndef __PAGETABLE_PMD_FOLDED
+		pmd_init((unsigned long)new, (unsigned long)invalid_pte_table);
+#endif
+	}
+
+	pmd = pmd_offset(pud, addr);
+	if (pmd_none(*pmd)) {
+		pte_t *new __maybe_unused;
+
+		new = memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
+		pmd_populate_kernel(&init_mm, pmd, new);
+	}
+
+	return pte_offset_kernel(pmd, addr);
+}
+
+void __init __set_fixmap(enum fixed_addresses idx,
+			       phys_addr_t phys, pgprot_t flags)
+{
+	unsigned long addr = __fix_to_virt(idx);
+	pte_t *ptep;
+
+	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
+
+	ptep = fixmap_pte(addr);
+	if (!pte_none(*ptep)) {
+		pte_ERROR(*ptep);
+		return;
+	}
+
+	if (pgprot_val(flags))
+		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
+	else {
+		pte_clear(&init_mm, addr, ptep);
+		flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
+	}
+}
+
 /*
  * Align swapper_pg_dir in to 64K, allows its address to be loaded
  * with a single LUI instruction in the TLB handlers.  If we used

arch/loongarch/pci/acpi.c

@@ -82,6 +82,69 @@ static int acpi_prepare_root_resources(struct acpi_pci_root_info *ci)
 	return 0;
 }
 
+/*
+ * Create a PCI config space window
+ *  - reserve mem region
+ *  - alloc struct pci_config_window with space for all mappings
+ *  - ioremap the config space
+ */
+static struct pci_config_window *arch_pci_ecam_create(struct device *dev,
+		struct resource *cfgres, struct resource *busr, const struct pci_ecam_ops *ops)
+{
+	int bsz, bus_range, err;
+	struct resource *conflict;
+	struct pci_config_window *cfg;
+
+	if (busr->start > busr->end)
+		return ERR_PTR(-EINVAL);
+
+	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
+	if (!cfg)
+		return ERR_PTR(-ENOMEM);
+
+	cfg->parent = dev;
+	cfg->ops = ops;
+	cfg->busr.start = busr->start;
+	cfg->busr.end = busr->end;
+	cfg->busr.flags = IORESOURCE_BUS;
+	bus_range = resource_size(cfgres) >> ops->bus_shift;
+
+	bsz = 1 << ops->bus_shift;
+
+	cfg->res.start = cfgres->start;
+	cfg->res.end = cfgres->end;
+	cfg->res.flags = IORESOURCE_MEM | IORESOURCE_BUSY;
+	cfg->res.name = "PCI ECAM";
+
+	conflict = request_resource_conflict(&iomem_resource, &cfg->res);
+	if (conflict) {
+		err = -EBUSY;
+		dev_err(dev, "can't claim ECAM area %pR: address conflict with %s %pR\n",
+			&cfg->res, conflict->name, conflict);
+		goto err_exit;
+	}
+
+	cfg->win = pci_remap_cfgspace(cfgres->start, bus_range * bsz);
+	if (!cfg->win)
+		goto err_exit_iomap;
+
+	if (ops->init) {
+		err = ops->init(cfg);
+		if (err)
+			goto err_exit;
+	}
+	dev_info(dev, "ECAM at %pR for %pR\n", &cfg->res, &cfg->busr);
+
+	return cfg;
+
+err_exit_iomap:
+	err = -ENOMEM;
+	dev_err(dev, "ECAM ioremap failed\n");
+err_exit:
+	pci_ecam_free(cfg);
+	return ERR_PTR(err);
+}
+
 /*
  * Lookup the bus range for the domain in MCFG, and set up config space
  * mapping.
@@ -106,11 +169,16 @@ pci_acpi_setup_ecam_mapping(struct acpi_pci_root *root)
 
 	bus_shift = ecam_ops->bus_shift ? : 20;
 
+	if (bus_shift == 20)
+		cfg = pci_ecam_create(dev, &cfgres, bus_res, ecam_ops);
+	else {
 		cfgres.start = root->mcfg_addr + (bus_res->start << bus_shift);
 		cfgres.end = cfgres.start + (resource_size(bus_res) << bus_shift) - 1;
+		cfgres.end |= BIT(28) + (((PCI_CFG_SPACE_EXP_SIZE - 1) & 0xf00) << 16);
 		cfgres.flags = IORESOURCE_MEM;
+		cfg = arch_pci_ecam_create(dev, &cfgres, bus_res, ecam_ops);
+	}
 
-	cfg = pci_ecam_create(dev, &cfgres, bus_res, ecam_ops);
 	if (IS_ERR(cfg)) {
 		dev_err(dev, "%04x:%pR error %ld mapping ECAM\n", seg, bus_res, PTR_ERR(cfg));
 		return NULL;