pci_schizo.c

linux-2.6.15.6

		p->pre_handler_arg2 = (void *) pbm->sync_reg;
	}

	return __irq(bucket);
}

/* SCHIZO error handling support. */
enum schizo_error_type {
	UE_ERR, CE_ERR, PCI_ERR, SAFARI_ERR
};

static DEFINE_SPINLOCK(stc_buf_lock);
static unsigned long stc_error_buf[128];
static unsigned long stc_tag_buf[16];
static unsigned long stc_line_buf[16];

#define SCHIZO_UE_INO		0x30 /* Uncorrectable ECC error */
#define SCHIZO_CE_INO		0x31 /* Correctable ECC error */
#define SCHIZO_PCIERR_A_INO	0x32 /* PBM A PCI bus error */
#define SCHIZO_PCIERR_B_INO	0x33 /* PBM B PCI bus error */
#define SCHIZO_SERR_INO		0x34 /* Safari interface error */

struct pci_pbm_info *pbm_for_ino(struct pci_controller_info *p, u32 ino)
{
	ino &= IMAP_INO;
	if (p->pbm_A.ino_bitmap & (1UL << ino))
		return &p->pbm_A;
	if (p->pbm_B.ino_bitmap & (1UL << ino))
		return &p->pbm_B;

	printk("PCI%d: No ino_bitmap entry for ino[%x], bitmaps "
	       "PBM_A[%016lx] PBM_B[%016lx]",
	       p->index, ino,
	       p->pbm_A.ino_bitmap,
	       p->pbm_B.ino_bitmap);
	printk("PCI%d: Using PBM_A, report this problem immediately.\n",
	       p->index);

	return &p->pbm_A;
}

static void schizo_clear_other_err_intr(struct pci_controller_info *p, int irq)
{
	struct pci_pbm_info *pbm;
	struct ino_bucket *bucket;
	unsigned long iclr;

	/* Do not clear the interrupt for the other PCI bus.
	 *
	 * This "ACK both PBM IRQs" only needs to be performed
	 * for chip-wide error interrupts.
	 */
	if ((irq & IMAP_INO) == SCHIZO_PCIERR_A_INO ||
	    (irq & IMAP_INO) == SCHIZO_PCIERR_B_INO)
		return;

	pbm = pbm_for_ino(p, irq);
	if (pbm == &p->pbm_A)
		pbm = &p->pbm_B;
	else
		pbm = &p->pbm_A;

	irq = schizo_irq_build(pbm, NULL,
			       (pbm->portid << 6) | (irq & IMAP_INO));
	bucket = __bucket(irq);
	iclr = bucket->iclr;

	upa_writel(ICLR_IDLE, iclr);
}

#define SCHIZO_STC_ERR	0xb800UL /* --> 0xba00 */
#define SCHIZO_STC_TAG	0xba00UL /* --> 0xba80 */
#define SCHIZO_STC_LINE	0xbb00UL /* --> 0xbb80 */

#define SCHIZO_STCERR_WRITE	0x2UL
#define SCHIZO_STCERR_READ	0x1UL

#define SCHIZO_STCTAG_PPN	0x3fffffff00000000UL
#define SCHIZO_STCTAG_VPN	0x00000000ffffe000UL
#define SCHIZO_STCTAG_VALID	0x8000000000000000UL
#define SCHIZO_STCTAG_READ	0x4000000000000000UL

#define SCHIZO_STCLINE_LINDX	0x0000000007800000UL
#define SCHIZO_STCLINE_SPTR	0x000000000007e000UL
#define SCHIZO_STCLINE_LADDR	0x0000000000001fc0UL
#define SCHIZO_STCLINE_EPTR	0x000000000000003fUL
#define SCHIZO_STCLINE_VALID	0x0000000000600000UL
#define SCHIZO_STCLINE_FOFN	0x0000000000180000UL

static void __schizo_check_stc_error_pbm(struct pci_pbm_info *pbm,
					 enum schizo_error_type type)
{
	struct pci_strbuf *strbuf = &pbm->stc;
	unsigned long regbase = pbm->pbm_regs;
	unsigned long err_base, tag_base, line_base;
	u64 control;
	int i;

	err_base = regbase + SCHIZO_STC_ERR;
	tag_base = regbase + SCHIZO_STC_TAG;
	line_base = regbase + SCHIZO_STC_LINE;

	spin_lock(&stc_buf_lock);

	/* This is __REALLY__ dangerous.  When we put the
	 * streaming buffer into diagnostic mode to probe
	 * it's tags and error status, we _must_ clear all
	 * of the line tag valid bits before re-enabling
	 * the streaming buffer.  If any dirty data lives
	 * in the STC when we do this, we will end up
	 * invalidating it before it has a chance to reach
	 * main memory.
	 */
	control = schizo_read(strbuf->strbuf_control);
	schizo_write(strbuf->strbuf_control,
		     (control | SCHIZO_STRBUF_CTRL_DENAB));
	for (i = 0; i < 128; i++) {
		unsigned long val;

		val = schizo_read(err_base + (i * 8UL));
		schizo_write(err_base + (i * 8UL), 0UL);
		stc_error_buf[i] = val;
	}
	for (i = 0; i < 16; i++) {
		stc_tag_buf[i] = schizo_read(tag_base + (i * 8UL));
		stc_line_buf[i] = schizo_read(line_base + (i * 8UL));
		schizo_write(tag_base + (i * 8UL), 0UL);
		schizo_write(line_base + (i * 8UL), 0UL);
	}

	/* OK, state is logged, exit diagnostic mode. */
	schizo_write(strbuf->strbuf_control, control);

	for (i = 0; i < 16; i++) {
		int j, saw_error, first, last;

		saw_error = 0;
		first = i * 8;
		last = first + 8;
		for (j = first; j < last; j++) {
			unsigned long errval = stc_error_buf[j];
			if (errval != 0) {
				saw_error++;
				printk("%s: STC_ERR(%d)[wr(%d)rd(%d)]\n",
				       pbm->name,
				       j,
				       (errval & SCHIZO_STCERR_WRITE) ? 1 : 0,
				       (errval & SCHIZO_STCERR_READ) ? 1 : 0);
			}
		}
		if (saw_error != 0) {
			unsigned long tagval = stc_tag_buf[i];
			unsigned long lineval = stc_line_buf[i];
			printk("%s: STC_TAG(%d)[PA(%016lx)VA(%08lx)V(%d)R(%d)]\n",
			       pbm->name,
			       i,
			       ((tagval & SCHIZO_STCTAG_PPN) >> 19UL),
			       (tagval & SCHIZO_STCTAG_VPN),
			       ((tagval & SCHIZO_STCTAG_VALID) ? 1 : 0),
			       ((tagval & SCHIZO_STCTAG_READ) ? 1 : 0));

			/* XXX Should spit out per-bank error information... -DaveM */
			printk("%s: STC_LINE(%d)[LIDX(%lx)SP(%lx)LADDR(%lx)EP(%lx)"
			       "V(%d)FOFN(%d)]\n",
			       pbm->name,
			       i,
			       ((lineval & SCHIZO_STCLINE_LINDX) >> 23UL),
			       ((lineval & SCHIZO_STCLINE_SPTR) >> 13UL),
			       ((lineval & SCHIZO_STCLINE_LADDR) >> 6UL),
			       ((lineval & SCHIZO_STCLINE_EPTR) >> 0UL),
			       ((lineval & SCHIZO_STCLINE_VALID) ? 1 : 0),
			       ((lineval & SCHIZO_STCLINE_FOFN) ? 1 : 0));
		}
	}

	spin_unlock(&stc_buf_lock);
}

/* IOMMU is per-PBM in Schizo, so interrogate both for anonymous
 * controller level errors.
 */

#define SCHIZO_IOMMU_TAG	0xa580UL
#define SCHIZO_IOMMU_DATA	0xa600UL

#define SCHIZO_IOMMU_TAG_CTXT	0x0000001ffe000000UL
#define SCHIZO_IOMMU_TAG_ERRSTS	0x0000000001800000UL
#define SCHIZO_IOMMU_TAG_ERR	0x0000000000400000UL
#define SCHIZO_IOMMU_TAG_WRITE	0x0000000000200000UL
#define SCHIZO_IOMMU_TAG_STREAM	0x0000000000100000UL
#define SCHIZO_IOMMU_TAG_SIZE	0x0000000000080000UL
#define SCHIZO_IOMMU_TAG_VPAGE	0x000000000007ffffUL

#define SCHIZO_IOMMU_DATA_VALID	0x0000000100000000UL
#define SCHIZO_IOMMU_DATA_CACHE	0x0000000040000000UL
#define SCHIZO_IOMMU_DATA_PPAGE	0x000000003fffffffUL

static void schizo_check_iommu_error_pbm(struct pci_pbm_info *pbm,
					 enum schizo_error_type type)
{
	struct pci_iommu *iommu = pbm->iommu;
	unsigned long iommu_tag[16];
	unsigned long iommu_data[16];
	unsigned long flags;
	u64 control;
	int i;

	spin_lock_irqsave(&iommu->lock, flags);
	control = schizo_read(iommu->iommu_control);
	if (control & SCHIZO_IOMMU_CTRL_XLTEERR) {
		unsigned long base;
		char *type_string;

		/* Clear the error encountered bit. */
		control &= ~SCHIZO_IOMMU_CTRL_XLTEERR;
		schizo_write(iommu->iommu_control, control);

		switch((control & SCHIZO_IOMMU_CTRL_XLTESTAT) >> 25UL) {
		case 0:
			type_string = "Protection Error";
			break;
		case 1:
			type_string = "Invalid Error";
			break;
		case 2:
			type_string = "TimeOut Error";
			break;
		case 3:
		default:
			type_string = "ECC Error";
			break;
		};
		printk("%s: IOMMU Error, type[%s]\n",
		       pbm->name, type_string);

		/* Put the IOMMU into diagnostic mode and probe
		 * it's TLB for entries with error status.
		 *
		 * It is very possible for another DVMA to occur
		 * while we do this probe, and corrupt the system
		 * further.  But we are so screwed at this point
		 * that we are likely to crash hard anyways, so
		 * get as much diagnostic information to the
		 * console as we can.
		 */
		schizo_write(iommu->iommu_control,
			     control | SCHIZO_IOMMU_CTRL_DENAB);

		base = pbm->pbm_regs;

		for (i = 0; i < 16; i++) {
			iommu_tag[i] =
				schizo_read(base + SCHIZO_IOMMU_TAG + (i * 8UL));
			iommu_data[i] =
				schizo_read(base + SCHIZO_IOMMU_DATA + (i * 8UL));

			/* Now clear out the entry. */
			schizo_write(base + SCHIZO_IOMMU_TAG + (i * 8UL), 0);
			schizo_write(base + SCHIZO_IOMMU_DATA + (i * 8UL), 0);
		}

		/* Leave diagnostic mode. */
		schizo_write(iommu->iommu_control, control);

		for (i = 0; i < 16; i++) {
			unsigned long tag, data;

			tag = iommu_tag[i];
			if (!(tag & SCHIZO_IOMMU_TAG_ERR))
				continue;

			data = iommu_data[i];
			switch((tag & SCHIZO_IOMMU_TAG_ERRSTS) >> 23UL) {
			case 0:
				type_string = "Protection Error";
				break;
			case 1:
				type_string = "Invalid Error";
				break;
			case 2:
				type_string = "TimeOut Error";
				break;
			case 3:
			default:
				type_string = "ECC Error";
				break;
			};
			printk("%s: IOMMU TAG(%d)[error(%s) ctx(%x) wr(%d) str(%d) "
			       "sz(%dK) vpg(%08lx)]\n",
			       pbm->name, i, type_string,
			       (int)((tag & SCHIZO_IOMMU_TAG_CTXT) >> 25UL),
			       ((tag & SCHIZO_IOMMU_TAG_WRITE) ? 1 : 0),
			       ((tag & SCHIZO_IOMMU_TAG_STREAM) ? 1 : 0),
			       ((tag & SCHIZO_IOMMU_TAG_SIZE) ? 64 : 8),
			       (tag & SCHIZO_IOMMU_TAG_VPAGE) << IOMMU_PAGE_SHIFT);
			printk("%s: IOMMU DATA(%d)[valid(%d) cache(%d) ppg(%016lx)]\n",
			       pbm->name, i,
			       ((data & SCHIZO_IOMMU_DATA_VALID) ? 1 : 0),
			       ((data & SCHIZO_IOMMU_DATA_CACHE) ? 1 : 0),
			       (data & SCHIZO_IOMMU_DATA_PPAGE) << IOMMU_PAGE_SHIFT);
		}
	}
	if (pbm->stc.strbuf_enabled)
		__schizo_check_stc_error_pbm(pbm, type);
	spin_unlock_irqrestore(&iommu->lock, flags);
}

static void schizo_check_iommu_error(struct pci_controller_info *p,
				     enum schizo_error_type type)
{
	schizo_check_iommu_error_pbm(&p->pbm_A, type);
	schizo_check_iommu_error_pbm(&p->pbm_B, type);
}

/* Uncorrectable ECC error status gathering. */
#define SCHIZO_UE_AFSR	0x10030UL
#define SCHIZO_UE_AFAR	0x10038UL

#define SCHIZO_UEAFSR_PPIO	0x8000000000000000UL /* Safari */
#define SCHIZO_UEAFSR_PDRD	0x4000000000000000UL /* Safari/Tomatillo */
#define SCHIZO_UEAFSR_PDWR	0x2000000000000000UL /* Safari */
#define SCHIZO_UEAFSR_SPIO	0x1000000000000000UL /* Safari */
#define SCHIZO_UEAFSR_SDMA	0x0800000000000000UL /* Safari/Tomatillo */
#define SCHIZO_UEAFSR_ERRPNDG	0x0300000000000000UL /* Safari */
#define SCHIZO_UEAFSR_BMSK	0x000003ff00000000UL /* Safari */
#define SCHIZO_UEAFSR_QOFF	0x00000000c0000000UL /* Safari/Tomatillo */
#define SCHIZO_UEAFSR_AID	0x000000001f000000UL /* Safari/Tomatillo */
#define SCHIZO_UEAFSR_PARTIAL	0x0000000000800000UL /* Safari */
#define SCHIZO_UEAFSR_OWNEDIN	0x0000000000400000UL /* Safari */
#define SCHIZO_UEAFSR_MTAGSYND	0x00000000000f0000UL /* Safari */
#define SCHIZO_UEAFSR_MTAG	0x000000000000e000UL /* Safari */
#define SCHIZO_UEAFSR_ECCSYND	0x00000000000001ffUL /* Safari */

static irqreturn_t schizo_ue_intr(int irq, void *dev_id, struct pt_regs *regs)
{
	struct pci_controller_info *p = dev_id;
	unsigned long afsr_reg = p->pbm_B.controller_regs + SCHIZO_UE_AFSR;
	unsigned long afar_reg = p->pbm_B.controller_regs + SCHIZO_UE_AFAR;
	unsigned long afsr, afar, error_bits;
	int reported, limit;

	/* Latch uncorrectable error status. */
	afar = schizo_read(afar_reg);

	/* If either of the error pending bits are set in the
	 * AFSR, the error status is being actively updated by
	 * the hardware and we must re-read to get a clean value.
	 */
	limit = 1000;
	do {
		afsr = schizo_read(afsr_reg);
	} while ((afsr & SCHIZO_UEAFSR_ERRPNDG) != 0 && --limit);

	/* Clear the primary/secondary error status bits. */
	error_bits = afsr &
		(SCHIZO_UEAFSR_PPIO | SCHIZO_UEAFSR_PDRD | SCHIZO_UEAFSR_PDWR |
		 SCHIZO_UEAFSR_SPIO | SCHIZO_UEAFSR_SDMA);
	if (!error_bits)
		return IRQ_NONE;
	schizo_write(afsr_reg, error_bits);

	/* Log the error. */
	printk("PCI%d: Uncorrectable Error, primary error type[%s]\n",
	       p->index,
	       (((error_bits & SCHIZO_UEAFSR_PPIO) ?
		 "PIO" :
		 ((error_bits & SCHIZO_UEAFSR_PDRD) ?
		  "DMA Read" :
		  ((error_bits & SCHIZO_UEAFSR_PDWR) ?
		   "DMA Write" : "???")))));
	printk("PCI%d: bytemask[%04lx] qword_offset[%lx] SAFARI_AID[%02lx]\n",
	       p->index,
	       (afsr & SCHIZO_UEAFSR_BMSK) >> 32UL,
	       (afsr & SCHIZO_UEAFSR_QOFF) >> 30UL,
	       (afsr & SCHIZO_UEAFSR_AID) >> 24UL);
	printk("PCI%d: partial[%d] owned_in[%d] mtag[%lx] mtag_synd[%lx] ecc_sync[%lx]\n",
	       p->index,
	       (afsr & SCHIZO_UEAFSR_PARTIAL) ? 1 : 0,
	       (afsr & SCHIZO_UEAFSR_OWNEDIN) ? 1 : 0,
	       (afsr & SCHIZO_UEAFSR_MTAG) >> 13UL,
	       (afsr & SCHIZO_UEAFSR_MTAGSYND) >> 16UL,
	       (afsr & SCHIZO_UEAFSR_ECCSYND) >> 0UL);
	printk("PCI%d: UE AFAR [%016lx]\n", p->index, afar);
	printk("PCI%d: UE Secondary errors [", p->index);
	reported = 0;
	if (afsr & SCHIZO_UEAFSR_SPIO) {
		reported++;
		printk("(PIO)");
	}
	if (afsr & SCHIZO_UEAFSR_SDMA) {
		reported++;
		printk("(DMA)");
	}
	if (!reported)
		printk("(none)");
	printk("]\n");

	/* Interrogate IOMMU for error status. */
	schizo_check_iommu_error(p, UE_ERR);

	schizo_clear_other_err_intr(p, irq);

	return IRQ_HANDLED;
}

#define SCHIZO_CE_AFSR	0x10040UL
#define SCHIZO_CE_AFAR	0x10048UL

#define SCHIZO_CEAFSR_PPIO	0x8000000000000000UL
#define SCHIZO_CEAFSR_PDRD	0x4000000000000000UL
#define SCHIZO_CEAFSR_PDWR	0x2000000000000000UL
#define SCHIZO_CEAFSR_SPIO	0x1000000000000000UL
#define SCHIZO_CEAFSR_SDMA	0x0800000000000000UL
#define SCHIZO_CEAFSR_ERRPNDG	0x0300000000000000UL
#define SCHIZO_CEAFSR_BMSK	0x000003ff00000000UL
#define SCHIZO_CEAFSR_QOFF	0x00000000c0000000UL