[PATCH 4/8] drivers-edac-add Cell MC driver - Kernel

This is a discussion on [PATCH 4/8] drivers-edac-add Cell MC driver - Kernel ; From: Benjamin Herrenschmidt Adds driver for the Cell memory controller when used without a Hypervisor such as on the IBM Cell blades. There might still be some improvements to do to this such as finding if it's possible to properly ...

+ Reply to Thread
Results 1 to 2 of 2

Thread: [PATCH 4/8] drivers-edac-add Cell MC driver

  1. [PATCH 4/8] drivers-edac-add Cell MC driver

    From: Benjamin Herrenschmidt

    Adds driver for the Cell memory controller when used without a
    Hypervisor such as on the IBM Cell blades. There might still
    be some improvements to do to this such as finding if it's
    possible to properly obtain more details about the address
    of the error but it's good enough already to report CE counts
    which is our main priority at the moment.

    Signed-off-by: Benjamin Herrenschmidt
    CC: Alan Cox Signed-off-by: Doug Thompson
    ---

    device_initcall(cell_publish_devices);
    drivers/edac/Kconfig | 7 +
    drivers/edac/Makefile | 1
    drivers/edac/cell_edac.c | 258 +++++++++++++++++++++++++++++++++++++++++++++++
    3 files changed, 266 insertions(+)

    Index: linux-2.6.23/drivers/edac/Kconfig
    ================================================== =================
    --- linux-2.6.23.orig/drivers/edac/Kconfig
    +++ linux-2.6.23/drivers/edac/Kconfig
    @@ -131,5 +131,12 @@ config EDAC_PASEMI
    Support for error detection and correction on PA Semi
    PWRficient.

    +config EDAC_CELL
    + tristate "Cell Broadband Engine memory controller"
    + depends on EDAC_MM_EDAC && PPC_CELL_NATIVE
    + help
    + Support for error detection and correction on the
    + Cell Broadband Engine internal memory controller
    + on platform without a hypervisor

    endif # EDAC
    Index: linux-2.6.23/drivers/edac/Makefile
    ================================================== =================
    --- linux-2.6.23.orig/drivers/edac/Makefile
    +++ linux-2.6.23/drivers/edac/Makefile
    @@ -28,4 +28,5 @@ obj-$(CONFIG_EDAC_I3000) += i3000_edac.
    obj-$(CONFIG_EDAC_I82860) += i82860_edac.o
    obj-$(CONFIG_EDAC_R82600) += r82600_edac.o
    obj-$(CONFIG_EDAC_PASEMI) += pasemi_edac.o
    +obj-$(CONFIG_EDAC_CELL) += cell_edac.o

    Index: linux-2.6.23/drivers/edac/cell_edac.c
    ================================================== =================
    --- /dev/null
    +++ linux-2.6.23/drivers/edac/cell_edac.c
    @@ -0,0 +1,258 @@
    +/*
    + * Cell MIC driver for ECC counting
    + *
    + * Copyright 2007 Benjamin Herrenschmidt, IBM Corp.
    + *
    + *
    + * This file may be distributed under the terms of the
    + * GNU General Public License.
    + */
    +#undef DEBUG
    +
    +#include
    +#include
    +#include
    +#include
    +#include
    +#include
    +#include
    +
    +#include "edac_core.h"
    +
    +struct cell_edac_priv
    +{
    + struct cbe_mic_tm_regs __iomem *regs;
    + int node;
    + int chanmask;
    +#ifdef DEBUG
    + u64 prev_fir;
    +#endif
    +};
    +
    +static void cell_edac_count_ce(struct mem_ctl_info *mci, int chan, u64 ar)
    +{
    + struct cell_edac_priv *priv = mci->pvt_info;
    + struct csrow_info *csrow = &mci->csrows[0];
    + unsigned long address, pfn, offset;
    +
    + dev_dbg(mci->dev, "ECC CE err on node %d, channel %d, ar = 0x%016lx\n",
    + priv->node, chan, ar);
    +
    + /* Address decoding is likely a bit bogus, to dbl check */
    + address = (ar & 0xffffffffe0000000ul) >> 29;
    + if (priv->chanmask == 0x3)
    + address = (address << 1) | chan;
    + pfn = address >> PAGE_SHIFT;
    + offset = address & ~PAGE_MASK;
    +
    + /* TODO: Decoding of the error addresss */
    + edac_mc_handle_ce(mci, csrow->first_page + pfn, offset,
    + 0, 0, chan, "");
    +}
    +
    +static void cell_edac_count_ue(struct mem_ctl_info *mci, int chan, u64 ar)
    +{
    + struct cell_edac_priv *priv = mci->pvt_info;
    + struct csrow_info *csrow = &mci->csrows[0];
    + unsigned long address, pfn, offset;
    +
    + dev_dbg(mci->dev, "ECC UE err on node %d, channel %d, ar = 0x%016lx\n",
    + priv->node, chan, ar);
    +
    + /* Address decoding is likely a bit bogus, to dbl check */
    + address = (ar & 0xffffffffe0000000ul) >> 29;
    + if (priv->chanmask == 0x3)
    + address = (address << 1) | chan;
    + pfn = address >> PAGE_SHIFT;
    + offset = address & ~PAGE_MASK;
    +
    + /* TODO: Decoding of the error addresss */
    + edac_mc_handle_ue(mci, csrow->first_page + pfn, offset, 0, "");
    +}
    +
    +static void cell_edac_check(struct mem_ctl_info *mci)
    +{
    + struct cell_edac_priv *priv = mci->pvt_info;
    + u64 fir, addreg, clear = 0;
    +
    + fir = in_be64(&priv->regs->mic_fir);
    +#ifdef DEBUG
    + if (fir != priv->prev_fir) {
    + dev_dbg(mci->dev, "fir change : 0x%016lx\n", fir);
    + priv->prev_fir = fir;
    + }
    +#endif
    + if ((priv->chanmask & 0x1) && (fir & CBE_MIC_FIR_ECC_SINGLE_0_ERR)) {
    + addreg = in_be64(&priv->regs->mic_df_ecc_address_0);
    + clear |= CBE_MIC_FIR_ECC_SINGLE_0_RESET;
    + cell_edac_count_ce(mci, 0, addreg);
    + }
    + if ((priv->chanmask & 0x2) && (fir & CBE_MIC_FIR_ECC_SINGLE_1_ERR)) {
    + addreg = in_be64(&priv->regs->mic_df_ecc_address_1);
    + clear |= CBE_MIC_FIR_ECC_SINGLE_1_RESET;
    + cell_edac_count_ce(mci, 1, addreg);
    + }
    + if ((priv->chanmask & 0x1) && (fir & CBE_MIC_FIR_ECC_MULTI_0_ERR)) {
    + addreg = in_be64(&priv->regs->mic_df_ecc_address_0);
    + clear |= CBE_MIC_FIR_ECC_MULTI_0_RESET;
    + cell_edac_count_ue(mci, 0, addreg);
    + }
    + if ((priv->chanmask & 0x2) && (fir & CBE_MIC_FIR_ECC_MULTI_1_ERR)) {
    + addreg = in_be64(&priv->regs->mic_df_ecc_address_1);
    + clear |= CBE_MIC_FIR_ECC_MULTI_1_RESET;
    + cell_edac_count_ue(mci, 1, addreg);
    + }
    +
    + /* The procedure for clearing FIR bits is a bit ... weird */
    + if (clear) {
    + fir &= ~(CBE_MIC_FIR_ECC_ERR_MASK | CBE_MIC_FIR_ECC_SET_MASK);
    + fir |= CBE_MIC_FIR_ECC_RESET_MASK;
    + fir &= ~clear;
    + out_be64(&priv->regs->mic_fir, fir);
    + (void)in_be64(&priv->regs->mic_fir);
    +
    + mb(); /* sync up */
    +#ifdef DEBUG
    + fir = in_be64(&priv->regs->mic_fir);
    + dev_dbg(mci->dev, "fir clear : 0x%016lx\n", fir);
    +#endif
    + }
    +}
    +
    +static void __devinit cell_edac_init_csrows(struct mem_ctl_info *mci)
    +{
    + struct csrow_info *csrow = &mci->csrows[0];
    + struct cell_edac_priv *priv = mci->pvt_info;
    + struct device_node *np;
    +
    + for (np = NULL;
    + (np = of_find_node_by_name(np, "memory")) != NULL {
    + struct resource r;
    +
    + /* We "know" that the Cell firmware only creates one entry
    + * in the "memory" nodes. If that changes, this code will
    + * need to be adapted.
    + */
    + if (of_address_to_resource(np, 0, &r))
    + continue;
    + if (of_node_to_nid(np) != priv->node)
    + continue;
    + csrow->first_page = r.start >> PAGE_SHIFT;
    + csrow->nr_pages = (r.end - r.start + 1) >> PAGE_SHIFT;
    + csrow->last_page = csrow->first_page + csrow->nr_pages - 1;
    + csrow->mtype = MEM_XDR;
    + csrow->edac_mode = EDAC_FLAG_EC | EDAC_FLAG_SECDED;
    + dev_dbg(mci->dev,
    + "Initialized on node %d, chanmask=0x%x,"
    + " first_page=0x%lx, nr_pages=0x%x\n",
    + priv->node, priv->chanmask,
    + csrow->first_page, csrow->nr_pages);
    + break;
    + }
    +}
    +
    +static int __devinit cell_edac_probe(struct platform_device *pdev)
    +{
    + struct cbe_mic_tm_regs __iomem *regs;
    + struct mem_ctl_info *mci;
    + struct cell_edac_priv *priv;
    + u64 reg;
    + int rc, chanmask;
    +
    + regs = cbe_get_cpu_mic_tm_regs(cbe_node_to_cpu(pdev->id));
    + if (regs == NULL)
    + return -ENODEV;
    +
    + /* Get channel population */
    + reg = in_be64(&regs->mic_mnt_cfg);
    + dev_dbg(&pdev->dev, "MIC_MNT_CFG = 0x%016lx\n", reg);
    + chanmask = 0;
    + if (reg & CBE_MIC_MNT_CFG_CHAN_0_POP)
    + chanmask |= 0x1;
    + if (reg & CBE_MIC_MNT_CFG_CHAN_1_POP)
    + chanmask |= 0x2;
    + if (chanmask == 0) {
    + dev_warn(&pdev->dev,
    + "Yuck ! No channel populated ? Aborting !\n");
    + return -ENODEV;
    + }
    + dev_dbg(&pdev->dev, "Initial FIR = 0x%016lx\n",
    + in_be64(&regs->mic_fir));
    +
    + /* Allocate & init EDAC MC data structure */
    + mci = edac_mc_alloc(sizeof(struct cell_edac_priv), 1,
    + chanmask == 3 ? 2 : 1, pdev->id);
    + if (mci == NULL)
    + return -ENOMEM;
    + priv = mci->pvt_info;
    + priv->regs = regs;
    + priv->node = pdev->id;
    + priv->chanmask = chanmask;
    + mci->dev = &pdev->dev;
    + mci->mtype_cap = MEM_FLAG_XDR;
    + mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_EC | EDAC_FLAG_SECDED;
    + mci->edac_cap = EDAC_FLAG_EC | EDAC_FLAG_SECDED;
    + mci->mod_name = "cell_edac";
    + mci->ctl_name = "MIC";
    + mci->dev_name = pdev->dev.bus_id;
    + mci->edac_check = cell_edac_check;
    + cell_edac_init_csrows(mci);
    +
    + /* Register with EDAC core */
    + rc = edac_mc_add_mc(mci);
    + if (rc) {
    + dev_err(&pdev->dev, "failed to register with EDAC core\n");
    + edac_mc_free(mci);
    + return rc;
    + }
    +
    + return 0;
    +}
    +
    +static int __devexit cell_edac_remove(struct platform_device *pdev)
    +{
    + struct mem_ctl_info *mci = edac_mc_del_mc(&pdev->dev);
    + if (mci)
    + edac_mc_free(mci);
    + return 0;
    +}
    +
    +static struct platform_driver cell_edac_driver = {
    + .driver = {
    + .name = "cbe-mic",
    + .owner = THIS_MODULE,
    + },
    + .probe = cell_edac_probe,
    + .remove = cell_edac_remove,
    +};
    +
    +static int __init cell_edac_init(void)
    +{
    + /* Sanity check registers data structure */
    + BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
    + mic_df_ecc_address_0) != 0xf8);
    + BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
    + mic_df_ecc_address_1) != 0x1b8);
    + BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
    + mic_df_config) != 0x218);
    + BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
    + mic_fir) != 0x230);
    + BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
    + mic_mnt_cfg) != 0x210);
    + BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
    + mic_exc) != 0x208);
    +
    + return platform_driver_register(&cell_edac_driver);
    +}
    +
    +static void __exit cell_edac_exit(void)
    +{
    + platform_driver_unregister(&cell_edac_driver);
    +}
    +
    +module_init(cell_edac_init);
    +module_exit(cell_edac_exit);
    +
    +MODULE_LICENSE("GPL");
    +MODULE_AUTHOR("Benjamin Herrenschmidt ");
    +MODULE_DESCRIPTION("ECC counting for Cell MIC");
    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

  2. Re: [PATCH 4/8] drivers-edac-add Cell MC driver

    On Fri, 19 Oct 2007 13:17:43 -0600
    dougthompson@xmission.com wrote:

    > + /* The procedure for clearing FIR bits is a bit ... weird */
    > + if (clear) {
    > + fir &= ~(CBE_MIC_FIR_ECC_ERR_MASK | CBE_MIC_FIR_ECC_SET_MASK);
    > + fir |= CBE_MIC_FIR_ECC_RESET_MASK;
    > + fir &= ~clear;
    > + out_be64(&priv->regs->mic_fir, fir);
    > + (void)in_be64(&priv->regs->mic_fir);


    The (void) cast isn't particularly popular practice. Did you find that it
    actually does anything useful?

    > +
    > + mb(); /* sync up */


    What's this here for? It could do with a more usful comment.

    If it's trying to perform some synchronisation of device register access
    then I suspect it didn't work. Or maybe it happens to work because of how
    ppc implements mb(), in which case a direct use of the appropriate ppc
    primitive might be clearer.

    > +#ifdef DEBUG
    > + fir = in_be64(&priv->regs->mic_fir);
    > + dev_dbg(mci->dev, "fir clear : 0x%016lx\n", fir);
    > +#endif
    > + }

    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

+ Reply to Thread