parisc: Add Page Deallocation Table (PDT) support

The firmare in most parisc machines maintains a Page Deallocation Table (PDT)
which holds a list of physical memory addresses where hardware detected memory
errors (single bit and double bit errors).

This patch adds the missing PDC firmware calls and the logic to read the PDT
from firmware, report all current PDT entries and exclude the reported bad
memory from being used by Linux.

Signed-off-by: Helge Deller <deller@gmx.de>

4 files changed, 261 insertions, 1 deletions

diff --git a/arch/parisc/kernel/Makefile b/arch/parisc/kernel/Makefile
index 69a11183d48d..c4294df69fb6 100644
--- a/arch/parisc/kernel/Makefile
+++ b/arch/parisc/kernel/Makefile
@@ -4,7 +4,7 @@
 
 extra-y			:= head.o vmlinux.lds
 
-obj-y	     	:= cache.o pacache.o setup.o traps.o time.o irq.o \
+obj-y	     	:= cache.o pacache.o setup.o pdt.o traps.o time.o irq.o \
 		   pa7300lc.o syscall.o entry.o sys_parisc.o firmware.o \
 		   ptrace.o hardware.o inventory.o drivers.o \
 		   signal.o hpmc.o real2.o parisc_ksyms.o unaligned.o \
diff --git a/arch/parisc/kernel/firmware.c b/arch/parisc/kernel/firmware.c
index 9d797ae4fa22..98190252c12f 100644
--- a/arch/parisc/kernel/firmware.c
+++ b/arch/parisc/kernel/firmware.c
@@ -957,6 +957,41 @@ int pdc_tod_read(struct pdc_tod *tod)
 }
 EXPORT_SYMBOL(pdc_tod_read);
 
+int pdc_mem_pdt_info(struct pdc_mem_retinfo *rinfo)
+{
+	int retval;
+	unsigned long flags;
+
+	spin_lock_irqsave(&pdc_lock, flags);
+	retval = mem_pdc_call(PDC_MEM, PDC_MEM_MEMINFO, __pa(pdc_result), 0);
+	convert_to_wide(pdc_result);
+	memcpy(rinfo, pdc_result, sizeof(*rinfo));
+	spin_unlock_irqrestore(&pdc_lock, flags);
+
+	return retval;
+}
+
+int pdc_mem_pdt_read_entries(struct pdc_mem_read_pdt *pret,
+		unsigned long *pdt_entries_ptr)
+{
+	int retval;
+	unsigned long flags;
+
+	spin_lock_irqsave(&pdc_lock, flags);
+	retval = mem_pdc_call(PDC_MEM, PDC_MEM_READ_PDT, __pa(pdc_result),
+			__pa(pdc_result2));
+	if (retval == PDC_OK) {
+		convert_to_wide(pdc_result);
+		memcpy(pret, pdc_result, sizeof(*pret));
+		convert_to_wide(pdc_result2);
+		memcpy(pdt_entries_ptr, pdc_result2,
+			pret->pdt_entries * sizeof(*pdt_entries_ptr));
+	}
+	spin_unlock_irqrestore(&pdc_lock, flags);
+
+	return retval;
+}
+
 /**
  * pdc_tod_set - Set the Time-Of-Day clock.
  * @sec: The number of seconds since epoch.
@@ -1383,6 +1418,79 @@ int pdc_pat_io_pci_cfg_write(unsigned long pci_addr, int pci_size, u32 val)
 
 	return retval;
 }
+
+/**
+ * pdc_pat_mem_pdc_info - Retrieve information about page deallocation table
+ * @rinfo: memory pdt information
+ *
+ */
+int pdc_pat_mem_pdt_info(struct pdc_pat_mem_retinfo *rinfo)
+{
+	int retval;
+	unsigned long flags;
+
+	spin_lock_irqsave(&pdc_lock, flags);
+	retval = mem_pdc_call(PDC_PAT_MEM, PDC_PAT_MEM_PD_INFO,
+			__pa(&pdc_result));
+	if (retval == PDC_OK)
+		memcpy(rinfo, &pdc_result, sizeof(*rinfo));
+	spin_unlock_irqrestore(&pdc_lock, flags);
+
+	return retval;
+}
+
+/**
+ * pdc_pat_mem_read_cell_pdt - Read PDT entries from (old) PAT firmware
+ * @pret: array of PDT entries
+ * @pdt_entries_ptr: ptr to hold number of PDT entries
+ * @max_entries: maximum number of entries to be read
+ *
+ */
+int pdc_pat_mem_read_cell_pdt(struct pdc_pat_mem_read_pd_retinfo *pret,
+		unsigned long *pdt_entries_ptr, unsigned long max_entries)
+{
+	int retval;
+	unsigned long flags, entries;
+
+	spin_lock_irqsave(&pdc_lock, flags);
+	/* PDC_PAT_MEM_CELL_READ is available on early PAT machines only */
+	retval = mem_pdc_call(PDC_PAT_MEM, PDC_PAT_MEM_CELL_READ,
+			__pa(&pdc_result), parisc_cell_num, __pa(&pdc_result2));
+
+	if (retval == PDC_OK) {
+		/* build up return value as for PDC_PAT_MEM_PD_READ */
+		entries = min(pdc_result[0], max_entries);
+		pret->pdt_entries = entries;
+		pret->actual_count_bytes = entries * sizeof(unsigned long);
+		memcpy(pdt_entries_ptr, &pdc_result2, pret->actual_count_bytes);
+	}
+
+	spin_unlock_irqrestore(&pdc_lock, flags);
+	WARN_ON(retval == PDC_OK && pdc_result[0] > max_entries);
+
+	return retval;
+}
+/**
+ * pdc_pat_mem_read_pd_pdt - Read PDT entries from (newer) PAT firmware
+ * @pret: array of PDT entries
+ * @pdt_entries_ptr: ptr to hold number of PDT entries
+ *
+ */
+int pdc_pat_mem_read_pd_pdt(struct pdc_pat_mem_read_pd_retinfo *pret,
+		unsigned long *pdt_entries_ptr, unsigned long count,
+		unsigned long offset)
+{
+	int retval;
+	unsigned long flags;
+
+	spin_lock_irqsave(&pdc_lock, flags);
+	retval = mem_pdc_call(PDC_PAT_MEM, PDC_PAT_MEM_PD_READ,
+		__pa(&pret), __pa(pdt_entries_ptr),
+		count, offset);
+	spin_unlock_irqrestore(&pdc_lock, flags);
+
+	return retval;
+}
 #endif /* CONFIG_64BIT */
 
 
diff --git a/arch/parisc/kernel/inventory.c b/arch/parisc/kernel/inventory.c
index c9789d9c73b4..b0fe19ac4d78 100644
--- a/arch/parisc/kernel/inventory.c
+++ b/arch/parisc/kernel/inventory.c
@@ -40,6 +40,11 @@
 
 int pdc_type __read_mostly = PDC_TYPE_ILLEGAL;
 
+/* cell number and location (PAT firmware only) */
+unsigned long parisc_cell_num __read_mostly;
+unsigned long parisc_cell_loc __read_mostly;
+
+
 void __init setup_pdc(void)
 {
 	long status;
@@ -78,6 +83,10 @@ void __init setup_pdc(void)
 	if (status == PDC_OK) {
 		pdc_type = PDC_TYPE_PAT;
 		pr_cont("64 bit PAT.\n");
+		parisc_cell_num = cell_info.cell_num;
+		parisc_cell_loc = cell_info.cell_loc;
+		pr_info("PAT: Running on cell %lu and location %lu.\n",
+			parisc_cell_num, parisc_cell_loc);
 		return;
 	}
 #endif
diff --git a/arch/parisc/kernel/pdt.c b/arch/parisc/kernel/pdt.c
new file mode 100644
index 000000000000..f3a797e670b0
--- /dev/null
+++ b/arch/parisc/kernel/pdt.c
@@ -0,0 +1,143 @@
+/*
+ *    Page Deallocation Table (PDT) support
+ *
+ *    The Page Deallocation Table (PDT) holds a table with pointers to bad
+ *    memory (broken RAM modules) which is maintained by firmware.
+ *
+ *    Copyright 2017 by Helge Deller <deller@gmx.de>
+ *
+ *    TODO:
+ *    - check regularily for new bad memory
+ *    - add userspace interface with procfs or sysfs
+ *    - increase number of PDT entries dynamically
+ */
+
+#include <linux/memblock.h>
+#include <linux/seq_file.h>
+
+#include <asm/pdc.h>
+#include <asm/pdcpat.h>
+#include <asm/sections.h>
+#include <asm/pgtable.h>
+
+enum pdt_access_type {
+	PDT_NONE,
+	PDT_PDC,
+	PDT_PAT_NEW,
+	PDT_PAT_OLD
+};
+
+static enum pdt_access_type pdt_type;
+
+/* global PDT status information */
+static struct pdc_mem_retinfo pdt_status;
+
+#define MAX_PDT_TABLE_SIZE	PAGE_SIZE
+#define MAX_PDT_ENTRIES		(MAX_PDT_TABLE_SIZE / sizeof(unsigned long))
+static unsigned long pdt_entry[MAX_PDT_ENTRIES] __page_aligned_bss;
+
+
+/* report PDT entries via /proc/meminfo */
+void arch_report_meminfo(struct seq_file *m)
+{
+	if (pdt_type == PDT_NONE)
+		return;
+
+	seq_printf(m, "PDT_max_entries: %7lu\n",
+			pdt_status.pdt_size);
+	seq_printf(m, "PDT_cur_entries: %7lu\n",
+			pdt_status.pdt_entries);
+}
+
+/*
+ * pdc_pdt_init()
+ *
+ * Initialize kernel PDT structures, read initial PDT table from firmware,
+ * report all current PDT entries and mark bad memory with memblock_reserve()
+ * to avoid that the kernel will use broken memory areas.
+ *
+ */
+void __init pdc_pdt_init(void)
+{
+	int ret, i;
+	unsigned long entries;
+	struct pdc_mem_read_pdt pdt_read_ret;
+
+	if (is_pdc_pat()) {
+		struct pdc_pat_mem_retinfo pat_rinfo;
+
+		pdt_type = PDT_PAT_NEW;
+		ret = pdc_pat_mem_pdt_info(&pat_rinfo);
+		pdt_status.pdt_size = pat_rinfo.max_pdt_entries;
+		pdt_status.pdt_entries = pat_rinfo.current_pdt_entries;
+		pdt_status.pdt_status = 0;
+		pdt_status.first_dbe_loc = pat_rinfo.first_dbe_loc;
+		pdt_status.good_mem = pat_rinfo.good_mem;
+	} else {
+		pdt_type = PDT_PDC;
+		ret = pdc_mem_pdt_info(&pdt_status);
+	}
+
+	if (ret != PDC_OK) {
+		pdt_type = PDT_NONE;
+		pr_info("PDT: Firmware does not provide any page deallocation"
+			" information.\n");
+		return;
+	}
+
+	entries = pdt_status.pdt_entries;
+	WARN_ON(entries > MAX_PDT_ENTRIES);
+
+	pr_info("PDT: size %lu, entries %lu, status %lu, dbe_loc 0x%lx,"
+		" good_mem %lu\n",
+			pdt_status.pdt_size, pdt_status.pdt_entries,
+			pdt_status.pdt_status, pdt_status.first_dbe_loc,
+			pdt_status.good_mem);
+
+	if (entries == 0) {
+		pr_info("PDT: Firmware reports all memory OK.\n");
+		return;
+	}
+
+	if (pdt_status.first_dbe_loc &&
+		pdt_status.first_dbe_loc <= __pa((unsigned long)&_end))
+		pr_crit("CRITICAL: Bad memory inside kernel image memory area!\n");
+
+	pr_warn("PDT: Firmware reports %lu entries of faulty memory:\n",
+		entries);
+
+	if (pdt_type == PDT_PDC)
+		ret = pdc_mem_pdt_read_entries(&pdt_read_ret, pdt_entry);
+	else {
+#ifdef CONFIG_64BIT
+		struct pdc_pat_mem_read_pd_retinfo pat_pret;
+
+		ret = pdc_pat_mem_read_cell_pdt(&pat_pret, pdt_entry,
+			MAX_PDT_ENTRIES);
+		if (ret != PDC_OK) {
+			pdt_type = PDT_PAT_OLD;
+			ret = pdc_pat_mem_read_pd_pdt(&pat_pret, pdt_entry,
+				MAX_PDT_TABLE_SIZE, 0);
+		}
+#else
+		ret = PDC_BAD_PROC;
+#endif
+	}
+
+	if (ret != PDC_OK) {
+		pdt_type = PDT_NONE;
+		pr_debug("PDT type %d, retval = %d\n", pdt_type, ret);
+		return;
+	}
+
+	for (i = 0; i < pdt_status.pdt_entries; i++) {
+		if (i < 20)
+			pr_warn("PDT: BAD PAGE #%d at 0x%08lx (error_type = %lu)\n",
+				i,
+				pdt_entry[i] & PAGE_MASK,
+				pdt_entry[i] & 1);
+
+		/* mark memory page bad */
+		memblock_reserve(pdt_entry[i] & PAGE_MASK, PAGE_SIZE);
+	}
+}