- Kernel - mailweb.openeuler.org

[PATCH openEuler-1.0-LTS] crypto: hisilicon/qm - alloc reserve buffer to set and get xqc
by wangyuan 06 Nov '23

06 Nov '23

From: Yu'an Wang <wangyuan46(a)huawei.com> driver inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I8E6ED CVE: NA -------------------------------- If the temporarily applied memory is used to set or get the xqc information, the driver will release the memory after the mailbox timeout. However, the hardware still performs the operation. Therefore, the released memory may be written by hardware. So when load driver, reserve memory is applied for xqc configuration. Signed-off-by: Yu'an Wang <wangyuan46(a)huawei.com> --- drivers/crypto/hisilicon/qm.c | 554 +++++++++++++++------------------- drivers/crypto/hisilicon/qm.h | 13 + 2 files changed, 255 insertions(+), 312 deletions(-) diff --git a/drivers/crypto/hisilicon/qm.c b/drivers/crypto/hisilicon/qm.c index 64e8ef1bf8e7..b4fd97e1e03e 100644 --- a/drivers/crypto/hisilicon/qm.c +++ b/drivers/crypto/hisilicon/qm.c @@ -50,7 +50,7 @@ #define QM_SQ_TYPE_MASK 0xf -#define QM_SQ_TAIL_IDX(sqc) ((le16_to_cpu((sqc)->w11) >> 6) & 0x1) +#define QM_SQ_TAIL_IDX(sqc) ((le16_to_cpu((sqc).w11) >> 6) & 0x1) /* cqc shift */ #define QM_CQ_HOP_NUM_SHIFT 0 @@ -64,7 +64,7 @@ #define QM_QC_CQE_SIZE 4 -#define QM_CQ_TAIL_IDX(cqc) ((le16_to_cpu((cqc)->w11) >> 6) & 0x1) +#define QM_CQ_TAIL_IDX(cqc) ((le16_to_cpu((cqc).w11) >> 6) & 0x1) /* eqc shift */ #define QM_EQE_AEQE_SIZE (2UL << 12) @@ -217,19 +217,6 @@ #define QM_MK_SQC_DW3_V2(sqe_sz) \ ((QM_Q_DEPTH - 1) | ((u32)ilog2(sqe_sz) << QM_SQ_SQE_SIZE_SHIFT)) -#define INIT_QC_COMMON(qc, base, pasid) do { \ - (qc)->head = 0; \ - (qc)->tail = 0; \ - (qc)->base_l = cpu_to_le32(lower_32_bits(base)); \ - (qc)->base_h = cpu_to_le32(upper_32_bits(base)); \ - (qc)->dw3 = 0; \ - (qc)->w8 = 0; \ - (qc)->rsvd0 = 0; \ - (qc)->pasid = cpu_to_le16(pasid); \ - (qc)->w11 = 0; \ - (qc)->rsvd1 = 0; \ -} while (0) - #define QMC_ALIGN(sz) ALIGN(sz, 32) static int __hisi_qm_start(struct hisi_qm *qm); @@ -537,6 +524,61 @@ static int hisi_qm_mb_read(struct hisi_qm *qm, u64 *base, u8 cmd, u16 queue) return 0; } +/* op 0: set xqc info to hardware, 1: get xqc info from hardware. */ +static int qm_set_and_get_xqc(struct hisi_qm *qm, u8 cmd, void *xqc, u32 qp_id, + bool op) +{ + struct hisi_qm *pf_qm = pci_get_drvdata(pci_physfn(qm->pdev)); + struct qm_mailbox mailbox; + dma_addr_t xqc_dma; + void *tmp_xqc; + size_t size; + int ret; + + switch (cmd) { + case QM_MB_CMD_SQC: + size = sizeof(struct qm_sqc); + tmp_xqc = qm->xqc_buf.sqc; + xqc_dma = qm->xqc_buf.sqc_dma; + break; + case QM_MB_CMD_CQC: + size = sizeof(struct qm_cqc); + tmp_xqc = qm->xqc_buf.cqc; + xqc_dma = qm->xqc_buf.cqc_dma; + break; + case QM_MB_CMD_EQC: + size = sizeof(struct qm_eqc); + tmp_xqc = qm->xqc_buf.eqc; + xqc_dma = qm->xqc_buf.eqc_dma; + break; + case QM_MB_CMD_AEQC: + size = sizeof(struct qm_aeqc); + tmp_xqc = qm->xqc_buf.aeqc; + xqc_dma = qm->xqc_buf.aeqc_dma; + break; + } + + /* No need to judge if master OOO is blocked. */ + if (qm_check_dev_error(pf_qm)) { + dev_err(&qm->pdev->dev, + "QM mailbox operation failed since qm is stop!\n"); + return -EIO; + } + + mutex_lock(&qm->mailbox_lock); + if (!op) + memcpy(tmp_xqc, xqc, size); + + qm_mb_pre_init(&mailbox, cmd, xqc_dma, qp_id, op); + ret = qm_mb_nolock(qm, &mailbox); + if (!ret && op) + memcpy(xqc, tmp_xqc, size); + + mutex_unlock(&qm->mailbox_lock); + + return ret; +} + static void qm_db_v1(struct hisi_qm *qm, u16 qn, u8 cmd, u16 index, u8 priority) { u64 doorbell; @@ -1166,35 +1208,6 @@ static ssize_t qm_cmd_read(struct file *filp, char __user *buffer, return (*pos = len); } -static void *qm_ctx_alloc(struct hisi_qm *qm, size_t ctx_size, - dma_addr_t *dma_addr) -{ - struct device *dev = &qm->pdev->dev; - void *ctx_addr; - - ctx_addr = kzalloc(ctx_size, GFP_KERNEL); - if (!ctx_addr) - return ERR_PTR(-ENOMEM); - - *dma_addr = dma_map_single(dev, ctx_addr, ctx_size, DMA_FROM_DEVICE); - if (dma_mapping_error(dev, *dma_addr)) { - dev_err(dev, "DMA mapping error!\n"); - kfree(ctx_addr); - return ERR_PTR(-ENOMEM); - } - - return ctx_addr; -} - -static void qm_ctx_free(struct hisi_qm *qm, size_t ctx_size, - const void *ctx_addr, dma_addr_t *dma_addr) -{ - struct device *dev = &qm->pdev->dev; - - dma_unmap_single(dev, *dma_addr, ctx_size, DMA_FROM_DEVICE); - kfree(ctx_addr); -} - static int dump_show(struct hisi_qm *qm, void *info, unsigned int info_size, char *info_name) { @@ -1205,8 +1218,10 @@ static int dump_show(struct hisi_qm *qm, void *info, #define BYTE_PER_DW 4 info_buf = kzalloc(info_size, GFP_KERNEL); - if (!info_buf) + if (!info_buf) { + dev_err(dev, "Fail to alloc dump info buf\n"); return -ENOMEM; + } for (i = 0; i < info_size; i++, info_curr++) { if (i % BYTE_PER_DW == 0) @@ -1230,21 +1245,11 @@ static int dump_show(struct hisi_qm *qm, void *info, return 0; } -static int qm_dump_sqc_raw(struct hisi_qm *qm, dma_addr_t dma_addr, u16 qp_id) -{ - return hisi_qm_mb_write(qm, QM_MB_CMD_SQC, dma_addr, qp_id, 1); -} - -static int qm_dump_cqc_raw(struct hisi_qm *qm, dma_addr_t dma_addr, u16 qp_id) -{ - return hisi_qm_mb_write(qm, QM_MB_CMD_CQC, dma_addr, qp_id, 1); -} - static int qm_sqc_dump(struct hisi_qm *qm, const char *s) { struct device *dev = &qm->pdev->dev; - struct qm_sqc *sqc, *sqc_curr; - dma_addr_t sqc_dma; + struct qm_sqc *sqc_curr; + struct qm_sqc sqc; u32 qp_id; int ret; @@ -1257,40 +1262,28 @@ static int qm_sqc_dump(struct hisi_qm *qm, const char *s) return -EINVAL; } - sqc = qm_ctx_alloc(qm, sizeof(struct qm_sqc), &sqc_dma); - if (IS_ERR(sqc)) - return PTR_ERR(sqc); - - ret = qm_dump_sqc_raw(qm, sqc_dma, qp_id); - if (ret) { - down_read(&qm->qps_lock); - if (qm->sqc) { - sqc_curr = qm->sqc + qp_id; - - ret = dump_show(qm, sqc_curr, sizeof(struct qm_sqc), - "SOFT SQC"); - if (ret) - dev_info(dev, "Show soft sqc failed!\n"); - } - up_read(&qm->qps_lock); + ret = qm_set_and_get_xqc(qm, QM_MB_CMD_SQC, &sqc, qp_id, 1); + if (!ret) + return dump_show(qm, &sqc, sizeof(struct qm_sqc), "SQC"); - goto mailbox_fail; + down_read(&qm->qps_lock); + if (qm->sqc) { + sqc_curr = qm->sqc + qp_id; + ret = dump_show(qm, sqc_curr, sizeof(struct qm_sqc), + "SOFT SQC"); + if (ret) + dev_info(dev, "Show soft sqc failed!\n"); } + up_read(&qm->qps_lock); - ret = dump_show(qm, sqc, sizeof(struct qm_sqc), "SQC"); - if (ret) - dev_info(dev, "Show hw sqc failed!\n"); - -mailbox_fail: - qm_ctx_free(qm, sizeof(struct qm_sqc), sqc, &sqc_dma); return ret; } static int qm_cqc_dump(struct hisi_qm *qm, const char *s) { struct device *dev = &qm->pdev->dev; - struct qm_cqc *cqc, *cqc_curr; - dma_addr_t cqc_dma; + struct qm_cqc *cqc_curr; + struct qm_cqc cqc; u32 qp_id; int ret; @@ -1303,40 +1296,28 @@ static int qm_cqc_dump(struct hisi_qm *qm, const char *s) return -EINVAL; } - cqc = qm_ctx_alloc(qm, sizeof(struct qm_cqc), &cqc_dma); - if (IS_ERR(cqc)) - return PTR_ERR(cqc); - - ret = qm_dump_cqc_raw(qm, cqc_dma, qp_id); - if (ret) { - down_read(&qm->qps_lock); - if (qm->cqc) { - cqc_curr = qm->cqc + qp_id; + ret = qm_set_and_get_xqc(qm, QM_MB_CMD_CQC, &cqc, qp_id, 1); + if (!ret) + return dump_show(qm, &cqc, sizeof(struct qm_cqc), "CQC"); - ret = dump_show(qm, cqc_curr, sizeof(struct qm_cqc), - "SOFT CQC"); - if (ret) - dev_info(dev, "Show soft cqc failed!\n"); - } - up_read(&qm->qps_lock); + down_read(&qm->qps_lock); + if (qm->cqc) { + cqc_curr = qm->cqc + qp_id; - goto mailbox_fail; + ret = dump_show(qm, cqc_curr, sizeof(struct qm_cqc), + "SOFT CQC"); + if (ret) + dev_info(dev, "Show soft cqc failed!\n"); } + up_read(&qm->qps_lock); - ret = dump_show(qm, cqc, sizeof(struct qm_cqc), "CQC"); - if (ret) - dev_info(dev, "Show hw cqc failed!\n"); - -mailbox_fail: - qm_ctx_free(qm, sizeof(struct qm_cqc), cqc, &cqc_dma); return ret; } static int qm_eqc_dump(struct hisi_qm *qm, char *s) { struct device *dev = &qm->pdev->dev; - struct qm_eqc *eqc; - dma_addr_t eqc_dma; + struct qm_eqc eqc; int ret; if (strsep(&s, " ")) { @@ -1344,28 +1325,17 @@ static int qm_eqc_dump(struct hisi_qm *qm, char *s) return -EINVAL; } - eqc = qm_ctx_alloc(qm, sizeof(struct qm_eqc), &eqc_dma); - if (IS_ERR(eqc)) - return PTR_ERR(eqc); - - ret = hisi_qm_mb_write(qm, QM_MB_CMD_EQC, eqc_dma, 0, 1); + ret = qm_set_and_get_xqc(qm, QM_MB_CMD_EQC, &eqc, 0, 1); if (ret) - goto mailbox_fail; - - ret = dump_show(qm, eqc, sizeof(struct qm_eqc), "EQC"); - if (ret) - dev_info(dev, "Show eqc failed!\n"); + return ret; -mailbox_fail: - qm_ctx_free(qm, sizeof(struct qm_eqc), eqc, &eqc_dma); - return ret; + return dump_show(qm, &eqc, sizeof(struct qm_eqc), "EQC"); } static int qm_aeqc_dump(struct hisi_qm *qm, char *s) { struct device *dev = &qm->pdev->dev; - struct qm_aeqc *aeqc; - dma_addr_t aeqc_dma; + struct qm_aeqc aeqc; int ret; if (strsep(&s, " ")) { @@ -1373,21 +1343,11 @@ static int qm_aeqc_dump(struct hisi_qm *qm, char *s) return -EINVAL; } - aeqc = qm_ctx_alloc(qm, sizeof(struct qm_aeqc), &aeqc_dma); - if (IS_ERR(aeqc)) - return PTR_ERR(aeqc); - - ret = hisi_qm_mb_write(qm, QM_MB_CMD_AEQC, aeqc_dma, 0, 1); - if (ret) - goto mailbox_fail; - - ret = dump_show(qm, aeqc, sizeof(struct qm_aeqc), "AEQC"); + ret = qm_set_and_get_xqc(qm, QM_MB_CMD_AEQC, &aeqc, 0, 1); if (ret) - dev_info(dev, "Show hw aeqc failed!\n"); + return ret; -mailbox_fail: - qm_ctx_free(qm, sizeof(struct qm_aeqc), aeqc, &aeqc_dma); - return ret; + return dump_show(qm, &aeqc, sizeof(struct qm_aeqc), "AEQC"); } static int q_dump_param_parse(struct hisi_qm *qm, char *s, @@ -1432,7 +1392,6 @@ static int q_dump_param_parse(struct hisi_qm *qm, char *s, static int qm_sq_dump(struct hisi_qm *qm, char *s) { - struct device *dev = &qm->pdev->dev; struct hisi_qp *qp; u32 qp_id, sqe_id; void *sqe_curr; @@ -1447,16 +1406,11 @@ static int qm_sq_dump(struct hisi_qm *qm, char *s) memset(sqe_curr + qm->debug.sqe_mask_offset, SQE_ADDR_MASK, qm->debug.sqe_mask_len); - ret = dump_show(qm, sqe_curr, qm->sqe_size, "SQE"); - if (ret) - dev_info(dev, "Show sqe failed!\n"); - - return ret; + return dump_show(qm, sqe_curr, qm->sqe_size, "SQE"); } static int qm_cq_dump(struct hisi_qm *qm, char *s) { - struct device *dev = &qm->pdev->dev; struct qm_cqe *cqe_curr; struct hisi_qp *qp; u32 qp_id, cqe_id; @@ -1468,11 +1422,8 @@ static int qm_cq_dump(struct hisi_qm *qm, char *s) qp = &qm->qp_array[qp_id]; cqe_curr = qp->cqe + cqe_id; - ret = dump_show(qm, cqe_curr, sizeof(struct qm_cqe), "CQE"); - if (ret) - dev_info(dev, "Show cqe failed!\n"); - return ret; + return dump_show(qm, cqe_curr, sizeof(struct qm_cqe), "CQE"); } static int qm_eq_dump(struct hisi_qm *qm, const char *s) @@ -1895,79 +1846,46 @@ EXPORT_SYMBOL_GPL(hisi_qm_release_qp); static int qm_sq_ctx_cfg(struct hisi_qp *qp, int qp_id, int pasid) { struct hisi_qm *qm = qp->qm; - struct device *dev = &qm->pdev->dev; enum qm_hw_ver ver = qm->ver; - struct qm_sqc *sqc; - dma_addr_t sqc_dma; - int ret; + struct qm_sqc sqc = {0}; - sqc = kzalloc(sizeof(struct qm_sqc), GFP_KERNEL); - if (!sqc) - return -ENOMEM; - sqc_dma = dma_map_single(dev, sqc, sizeof(struct qm_sqc), - DMA_TO_DEVICE); - if (dma_mapping_error(dev, sqc_dma)) { - kfree(sqc); - return -ENOMEM; - } - - INIT_QC_COMMON(sqc, qp->sqe_dma, pasid); if (ver == QM_HW_V1) { - sqc->dw3 = cpu_to_le32(QM_MK_SQC_DW3_V1(0, 0, 0, qm->sqe_size)); - sqc->w8 = cpu_to_le16(QM_Q_DEPTH - 1); + sqc.dw3 = cpu_to_le32(QM_MK_SQC_DW3_V1(0, 0, 0, qm->sqe_size)); + sqc.w8 = cpu_to_le16(QM_Q_DEPTH - 1); } else if (ver == QM_HW_V2) { - sqc->dw3 = cpu_to_le32(QM_MK_SQC_DW3_V2(qm->sqe_size)); - sqc->w8 = 0; /* rand_qc */ + sqc.dw3 = cpu_to_le32(QM_MK_SQC_DW3_V2(qm->sqe_size)); + sqc.w8 = 0; /* rand_qc */ } - sqc->cq_num = cpu_to_le16(qp_id); - sqc->w13 = cpu_to_le16(QM_MK_SQC_W13(0, 1, qp->alg_type)); - - ret = hisi_qm_mb_write(qm, QM_MB_CMD_SQC, sqc_dma, qp_id, 0); + sqc.cq_num = cpu_to_le16(qp_id); + sqc.w13 = cpu_to_le16(QM_MK_SQC_W13(0, 1, qp->alg_type)); + sqc.base_l = cpu_to_le32(lower_32_bits(qp->sqe_dma)); + sqc.base_h = cpu_to_le32(upper_32_bits(qp->sqe_dma)); + sqc.pasid = cpu_to_le16(pasid); - dma_unmap_single(dev, sqc_dma, sizeof(struct qm_sqc), DMA_TO_DEVICE); - kfree(sqc); - - return ret; + return qm_set_and_get_xqc(qm, QM_MB_CMD_SQC, &sqc, qp_id, 0); } static int qm_cq_ctx_cfg(struct hisi_qp *qp, int qp_id, int pasid) { struct hisi_qm *qm = qp->qm; - struct device *dev = &qm->pdev->dev; enum qm_hw_ver ver = qm->ver; - struct qm_cqc *cqc; - dma_addr_t cqc_dma; - int ret; - - cqc = kzalloc(sizeof(struct qm_cqc), GFP_KERNEL); - if (!cqc) - return -ENOMEM; - - cqc_dma = dma_map_single(dev, cqc, sizeof(struct qm_cqc), - DMA_TO_DEVICE); - if (dma_mapping_error(dev, cqc_dma)) { - kfree(cqc); - return -ENOMEM; - } + struct qm_cqc cqc = {0}; - INIT_QC_COMMON(cqc, qp->cqe_dma, pasid); if (ver == QM_HW_V1) { - cqc->dw3 = cpu_to_le32(QM_MK_CQC_DW3_V1(0, 0, 0, + cqc.dw3 = cpu_to_le32(QM_MK_CQC_DW3_V1(0, 0, 0, QM_QC_CQE_SIZE)); - cqc->w8 = cpu_to_le16(QM_Q_DEPTH - 1); + cqc.w8 = cpu_to_le16(QM_Q_DEPTH - 1); } else if (ver == QM_HW_V2) { - cqc->dw3 = cpu_to_le32(QM_MK_CQC_DW3_V2(QM_QC_CQE_SIZE)); - cqc->w8 = 0; /* rand_qc */ + cqc.dw3 = cpu_to_le32(QM_MK_CQC_DW3_V2(QM_QC_CQE_SIZE)); + cqc.w8 = 0; /* rand_qc */ } - cqc->dw6 = cpu_to_le32(1 << QM_CQ_PHASE_SHIFT | - qp->c_flag << QM_CQ_FLAG_SHIFT); + cqc.dw6 = cpu_to_le32(1 << QM_CQ_PHASE_SHIFT | + qp->c_flag << QM_CQ_FLAG_SHIFT); + cqc.base_l = cpu_to_le32(lower_32_bits(qp->cqe_dma)); + cqc.base_h = cpu_to_le32(upper_32_bits(qp->cqe_dma)); + cqc.pasid = cpu_to_le16(pasid); - ret = hisi_qm_mb_write(qm, QM_MB_CMD_CQC, cqc_dma, qp_id, 0); - - dma_unmap_single(dev, cqc_dma, sizeof(struct qm_cqc), DMA_TO_DEVICE); - kfree(cqc); - - return ret; + return qm_set_and_get_xqc(qm, QM_MB_CMD_CQC, &cqc, qp_id, 0); } static int qm_qp_ctx_cfg(struct hisi_qp *qp, int qp_id, int pasid) @@ -2043,54 +1961,40 @@ static void qp_stop_fail_cb(struct hisi_qp *qp) static void qm_qp_has_no_task(struct hisi_qp *qp) { - size_t size = sizeof(struct qm_sqc) + sizeof(struct qm_cqc); - struct device *dev = &qp->qm->pdev->dev; - struct qm_sqc *sqc; - struct qm_cqc *cqc; - dma_addr_t dma_addr; - void *addr; - int i = 0; - int ret; + struct hisi_qm *qm = qp->qm; + struct device *dev = &qm->pdev->dev; + struct qm_sqc sqc; + struct qm_cqc cqc; + int ret, i = 0; if (qp->qm->err_ini.err_info.is_qm_ecc_mbit || qp->qm->err_ini.err_info.is_dev_ecc_mbit) return; - addr = qm_ctx_alloc(qp->qm, size, &dma_addr); - if (IS_ERR(addr)) { - dev_err(dev, "alloc ctx for sqc and cqc failed!\n"); - return; - } - while (++i) { - ret = qm_dump_sqc_raw(qp->qm, dma_addr, qp->qp_id); + ret = qm_set_and_get_xqc(qm, QM_MB_CMD_SQC, &sqc, qp->qp_id, 1); if (ret) { - dev_err(dev, "Failed to dump sqc!\n"); - break; + dev_err_ratelimited(dev, "Fail to dump sqc!\n"); + return; } - sqc = addr; - ret = qm_dump_cqc_raw(qp->qm, - (dma_addr + sizeof(struct qm_sqc)), qp->qp_id); + ret = qm_set_and_get_xqc(qm, QM_MB_CMD_CQC, &cqc, qp->qp_id, 1); if (ret) { - dev_err(dev, "Failed to dump cqc!\n"); - break; + dev_err_ratelimited(dev, "Fail to dump cqc!\n"); + return; } - cqc = addr + sizeof(struct qm_sqc); - if ((sqc->tail == cqc->tail) && - (QM_SQ_TAIL_IDX(sqc) == QM_CQ_TAIL_IDX(cqc))) - break; + if ((QM_SQ_TAIL_IDX(sqc) == QM_CQ_TAIL_IDX(cqc)) && + (sqc.tail == cqc.tail)) + return; if (i == MAX_WAIT_COUNTS) { - dev_err(dev, "Fail to wait for device stop!\n"); - break; + dev_err(dev, "Fail to empty queue %u!\n", qp->qp_id); + return; } usleep_range(WAIT_PERIOD_US_MIN, WAIT_PERIOD_US_MAX); } - - qm_ctx_free(qp->qm, size, addr, &dma_addr); } static int hisi_qm_stop_qp_nolock(struct hisi_qp *qp) @@ -2597,10 +2501,12 @@ static int hisi_qp_memory_init(struct hisi_qm *qm, size_t dma_size, int id) struct hisi_qp *qp; qp = &qm->qp_array[id]; - qp->qdma.va = dma_alloc_coherent(dev, dma_size, - &qp->qdma.dma, GFP_KERNEL); - if (!qp->qdma.va) + qp->qdma.va = dma_alloc_coherent(dev, dma_size, &qp->qdma.dma, + GFP_KERNEL); + if (!qp->qdma.va) { + dev_err(dev, "Fail to alloc qp dma buf size=%zx\n", dma_size); return -ENOMEM; + } qp->sqe = qp->qdma.va; qp->sqe_dma = qp->qdma.dma; @@ -2613,13 +2519,79 @@ static int hisi_qp_memory_init(struct hisi_qm *qm, size_t dma_size, int id) return 0; } +static int hisi_qp_alloc_memory(struct hisi_qm *qm) +{ + size_t qp_dma_size; + int i, ret; + + qm->qp_array = kcalloc(qm->qp_num, sizeof(struct hisi_qp), GFP_KERNEL); + if (!qm->qp_array) + return -ENOMEM; + + /* one more page for device or qp statuses */ + qp_dma_size = qm->sqe_size * QM_Q_DEPTH + + sizeof(struct cqe) * QM_Q_DEPTH; + qp_dma_size = PAGE_ALIGN(qp_dma_size) + PAGE_SIZE; + for (i = 0; i < qm->qp_num; i++) { + ret = hisi_qp_memory_init(qm, qp_dma_size, i); + if (ret) + goto err_init_qp_mem; + } + + return 0; + +err_init_qp_mem: + hisi_qp_memory_uninit(qm, i); + + return ret; +} + +static void hisi_qm_free_rsv_buf(struct hisi_qm *qm) +{ + struct qm_dma *xqc_dma = &qm->xqc_buf.qcdma; + struct device *dev = &qm->pdev->dev; + + dma_free_coherent(dev, xqc_dma->size, xqc_dma->va, xqc_dma->dma); +} + +static int hisi_qm_alloc_rsv_buf(struct hisi_qm *qm) +{ + struct qm_rsv_buf *xqc_buf = &qm->xqc_buf; + struct qm_dma *xqc_dma = &xqc_buf->qcdma; + struct device *dev = &qm->pdev->dev; + size_t off = 0; + +#define QM_XQC_BUF_INIT(xqc_buf, type) do { \ + (xqc_buf)->type = ((xqc_buf)->qcdma.va + (off)); \ + (xqc_buf)->type##_dma = (xqc_buf)->qcdma.dma + (off); \ + off += QMC_ALIGN(sizeof(struct qm_##type)); \ +} while (0) + + xqc_dma->size = QMC_ALIGN(sizeof(struct qm_eqc)) + + QMC_ALIGN(sizeof(struct qm_aeqc)) + + QMC_ALIGN(sizeof(struct qm_sqc)) + + QMC_ALIGN(sizeof(struct qm_cqc)); + + xqc_dma->va = dma_alloc_coherent(dev, xqc_dma->size, &xqc_dma->dma, + GFP_ATOMIC); + if (!xqc_dma->va) { + dev_err(dev, "Fail to alloc qcdma size=%zx\n", xqc_dma->size); + return -ENOMEM; + } + + QM_XQC_BUF_INIT(xqc_buf, eqc); + QM_XQC_BUF_INIT(xqc_buf, aeqc); + QM_XQC_BUF_INIT(xqc_buf, sqc); + QM_XQC_BUF_INIT(xqc_buf, cqc); + + return 0; +} + static int hisi_qm_memory_init(struct hisi_qm *qm) { struct device *dev = &qm->pdev->dev; - size_t qp_dma_size; + int ret = -ENOMEM; size_t off = 0; - int ret = 0; - int i; #define QM_INIT_BUF(qm, type, num) do { \ (qm)->type = ((qm)->qdma.va + (off)); \ @@ -2635,41 +2607,35 @@ static int hisi_qm_memory_init(struct hisi_qm *qm) QMC_ALIGN(sizeof(struct qm_aeqe) * QM_Q_DEPTH) + QMC_ALIGN(sizeof(struct qm_sqc) * qm->qp_num) + QMC_ALIGN(sizeof(struct qm_cqc) * qm->qp_num); - qm->qdma.va = dma_alloc_coherent(dev, qm->qdma.size, - &qm->qdma.dma, GFP_ATOMIC | __GFP_ZERO); - dev_dbg(dev, "allocate qm dma buf size=%zx)\n", qm->qdma.size); - if (!qm->qdma.va) - return -ENOMEM; + qm->qdma.va = dma_alloc_coherent(dev, qm->qdma.size, &qm->qdma.dma, + GFP_ATOMIC | __GFP_ZERO); + if (!qm->qdma.va) { + dev_err(dev, "Fail to alloc qdma size=%zx\n", qm->qdma.size); + goto err_destroy_idr; + } QM_INIT_BUF(qm, eqe, QM_EQ_DEPTH); QM_INIT_BUF(qm, aeqe, QM_Q_DEPTH); QM_INIT_BUF(qm, sqc, qm->qp_num); QM_INIT_BUF(qm, cqc, qm->qp_num); - qm->qp_array = kcalloc(qm->qp_num, sizeof(struct hisi_qp), GFP_KERNEL); - if (!qm->qp_array) { - ret = -ENOMEM; - goto err_alloc_qp_array; - } + ret = hisi_qm_alloc_rsv_buf(qm); + if (ret) + goto err_free_qdma; - /* one more page for device or qp statuses */ - qp_dma_size = qm->sqe_size * QM_Q_DEPTH + - sizeof(struct cqe) * QM_Q_DEPTH; - qp_dma_size = PAGE_ALIGN(qp_dma_size) + PAGE_SIZE; - for (i = 0; i < qm->qp_num; i++) { - ret = hisi_qp_memory_init(qm, qp_dma_size, i); - if (ret) - goto err_init_qp_mem; + ret = hisi_qp_alloc_memory(qm); + if (ret) + goto err_free_reserve_buf; - dev_dbg(dev, "allocate qp dma buf size=%zx)\n", qp_dma_size); - } + return 0; + +err_free_reserve_buf: + hisi_qm_free_rsv_buf(qm); +err_free_qdma: + dma_free_coherent(dev, qm->qdma.size, qm->qdma.va, qm->qdma.dma); +err_destroy_idr: + idr_destroy(&qm->qp_idr); - return ret; -err_init_qp_mem: - hisi_qp_memory_uninit(qm, i); -err_alloc_qp_array: - dma_free_coherent(dev, qm->qdma.size, - qm->qdma.va, qm->qdma.dma); return ret; } @@ -2861,8 +2827,7 @@ void hisi_qm_uninit(struct hisi_qm *qm) hisi_qp_memory_uninit(qm, qm->qp_num); idr_destroy(&qm->qp_idr); - - /* qm hardware buffer free on put_queue if no dma api */ + hisi_qm_free_rsv_buf(qm); if (qm->qdma.va) { hisi_qm_cache_wb(qm); dma_free_coherent(dev, qm->qdma.size, @@ -2962,59 +2927,26 @@ static void qm_init_eq_aeq_status(struct hisi_qm *qm) static int qm_eq_ctx_cfg(struct hisi_qm *qm) { - struct device *dev = &qm->pdev->dev; - struct qm_eqc *eqc; - dma_addr_t eqc_dma; - int ret; + struct qm_eqc eqc = {0}; - eqc = kzalloc(sizeof(struct qm_eqc), GFP_KERNEL); - if (!eqc) - return -ENOMEM; - eqc_dma = dma_map_single(dev, eqc, sizeof(struct qm_eqc), - DMA_TO_DEVICE); - if (dma_mapping_error(dev, eqc_dma)) { - kfree(eqc); - return -ENOMEM; - } - - eqc->base_l = cpu_to_le32(lower_32_bits(qm->eqe_dma)); - eqc->base_h = cpu_to_le32(upper_32_bits(qm->eqe_dma)); + eqc.base_l = cpu_to_le32(lower_32_bits(qm->eqe_dma)); + eqc.base_h = cpu_to_le32(upper_32_bits(qm->eqe_dma)); if (qm->ver == QM_HW_V1) - eqc->dw3 = cpu_to_le32(QM_EQE_AEQE_SIZE); - eqc->dw6 = cpu_to_le32((QM_EQ_DEPTH - 1) | (1 << QM_EQC_PHASE_SHIFT)); - ret = hisi_qm_mb_write(qm, QM_MB_CMD_EQC, eqc_dma, 0, 0); - dma_unmap_single(dev, eqc_dma, sizeof(struct qm_eqc), DMA_TO_DEVICE); - kfree(eqc); + eqc.dw3 = cpu_to_le32(QM_EQE_AEQE_SIZE); + eqc.dw6 = cpu_to_le32((QM_EQ_DEPTH - 1) | (1 << QM_EQC_PHASE_SHIFT)); - return ret; + return qm_set_and_get_xqc(qm, QM_MB_CMD_EQC, &eqc, 0, 0); } static int qm_aeq_ctx_cfg(struct hisi_qm *qm) { - struct device *dev = &qm->pdev->dev; - struct qm_aeqc *aeqc; - dma_addr_t aeqc_dma; - int ret; + struct qm_aeqc aeqc = {0}; - aeqc = kzalloc(sizeof(struct qm_aeqc), GFP_KERNEL); - if (!aeqc) - return -ENOMEM; - aeqc_dma = dma_map_single(dev, aeqc, sizeof(struct qm_aeqc), - DMA_TO_DEVICE); - if (dma_mapping_error(dev, aeqc_dma)) { - kfree(aeqc); - return -ENOMEM; - } + aeqc.base_l = cpu_to_le32(lower_32_bits(qm->aeqe_dma)); + aeqc.base_h = cpu_to_le32(upper_32_bits(qm->aeqe_dma)); + aeqc.dw6 = cpu_to_le32((QM_Q_DEPTH - 1) | (1 << QM_EQC_PHASE_SHIFT)); - aeqc->base_l = cpu_to_le32(lower_32_bits(qm->aeqe_dma)); - aeqc->base_h = cpu_to_le32(upper_32_bits(qm->aeqe_dma)); - aeqc->dw6 = cpu_to_le32((QM_Q_DEPTH - 1) | (1 << QM_EQC_PHASE_SHIFT)); - ret = hisi_qm_mb_write(qm, QM_MB_CMD_AEQC, aeqc_dma, 0, 0); - - dma_unmap_single(dev, aeqc_dma, sizeof(struct qm_aeqc), DMA_TO_DEVICE); - kfree(aeqc); - - return ret; + return qm_set_and_get_xqc(qm, QM_MB_CMD_AEQC, &aeqc, 0, 0); } static int qm_eq_aeq_ctx_cfg(struct hisi_qm *qm) @@ -3841,10 +3773,8 @@ static int qm_vf_reset_prepare(struct pci_dev *pdev, ret = hisi_qm_stop(qm, stop_reason); if (ret) { - hisi_qm_set_hw_reset(qm, - QM_RESET_STOP_TX_OFFSET); - hisi_qm_set_hw_reset(qm, - QM_RESET_STOP_RX_OFFSET); + hisi_qm_set_hw_reset(qm, QM_RESET_STOP_TX_OFFSET); + hisi_qm_set_hw_reset(qm, QM_RESET_STOP_RX_OFFSET); atomic_set(&qm->status.flags, QM_STOP); } } diff --git a/drivers/crypto/hisilicon/qm.h b/drivers/crypto/hisilicon/qm.h index d7d23d1ec34c..9f5e440d7396 100644 --- a/drivers/crypto/hisilicon/qm.h +++ b/drivers/crypto/hisilicon/qm.h @@ -286,6 +286,18 @@ struct hisi_qm_list { bool (*check)(struct hisi_qm *qm); }; +struct qm_rsv_buf { + struct qm_sqc *sqc; + struct qm_cqc *cqc; + struct qm_eqc *eqc; + struct qm_aeqc *aeqc; + dma_addr_t sqc_dma; + dma_addr_t cqc_dma; + dma_addr_t eqc_dma; + dma_addr_t aeqc_dma; + struct qm_dma qcdma; +}; + struct hisi_qm { enum qm_hw_ver ver; enum qm_fun_type fun_type; @@ -309,6 +321,7 @@ struct hisi_qm { dma_addr_t cqc_dma; dma_addr_t eqe_dma; dma_addr_t aeqe_dma; + struct qm_rsv_buf xqc_buf; struct hisi_qm_status status; struct hisi_qm_err_ini err_ini; -- 2.30.0

2 1

[PATCH OLK-5.10 v4 0/7] AGDI: Add driver for Arm Generic Diagnostic Dump and Reset device
by Qian Zou 06 Nov '23

06 Nov '23

From: “z00447436” <zouqian4(a)huawei.com> Some use-cases, such as system management, require the ability to generate a non-maskable event to the OS to request the OS kernel to perform a diagnostic dump and reset the system. Arm Generic Diagnostic Dump and Reset device enables a maintainer to request OS to perform a diagnostic dump and reset a system via SDEI event or an interrupt. This patch implements SDEI path and discards the interrupted context before proceeding to the crash kernel. D Scott Phillips (1): arm64: sdei: abort running SDEI handlers during crash Ilkka Koskinen (4): ACPI: tables: Add AGDI to the list of known table signatures ACPI: AGDI: Add driver for Arm Generic Diagnostic Dump and Reset device ACPICA: iASL: Add suppport for AGDI table ACPI: AGDI: Fix missing prototype warning for acpi_agdi_init() Jia He (1): EDAC/ghes: Make ghes_edac a proper module Shuai Xue (1): ACPI: APEI: explicit init of HEST and GHES in apci_init() arch/arm64/include/asm/sdei.h | 6 ++ arch/arm64/kernel/entry.S | 27 +++++++- arch/arm64/kernel/sdei.c | 3 + arch/arm64/kernel/smp.c | 8 +-- drivers/acpi/apei/ghes.c | 23 +++---- drivers/acpi/arm64/Kconfig | 10 +++ drivers/acpi/arm64/Makefile | 1 + drivers/acpi/arm64/agdi.c | 117 ++++++++++++++++++++++++++++++++++ drivers/acpi/bus.c | 4 ++ drivers/acpi/pci_root.c | 3 - drivers/acpi/tables.c | 2 +- drivers/edac/Kconfig | 4 +- drivers/edac/ghes_edac.c | 40 +++++++++++- drivers/firmware/Kconfig | 1 + drivers/firmware/arm_sdei.c | 32 ++++++---- include/acpi/actbl2.h | 20 ++++++ include/acpi/apei.h | 4 +- include/acpi/ghes.h | 20 +----- include/linux/acpi_agdi.h | 13 ++++ include/linux/arm_sdei.h | 4 ++ 20 files changed, 283 insertions(+), 59 deletions(-) create mode 100644 drivers/acpi/arm64/agdi.c create mode 100644 include/linux/acpi_agdi.h -- 2.33.0

2 8

[PATCH OLK-5.10] blk-mq: avoid housekeeping CPUs scheduling a worker on a non-housekeeping CPU
by Xiongfeng Wang 06 Nov '23

06 Nov '23

hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I7ZBTV CVE: N/A ---------------------------------------------------- When NOHZ_FULL is enabled, such as in HPC situation, CPUs are divided into housekeeping CPUs and non-housekeeping CPUs. Non-housekeeping CPUs are NOHZ_FULL CPUs and are often monopolized by the userspace process, such HPC application process. Any sort of interruption is not expected. blk_mq_hctx_next_cpu() selects each cpu in 'hctx->cpumask' alternately to schedule the work thread blk_mq_run_work_fn(). When 'hctx->cpumask' contains housekeeping CPU and non-housekeeping CPU at the same time, a housekeeping CPU, which want to request a IO, may schedule a worker on a non-housekeeping CPU. This may affect the performance of the userspace application running on non-housekeeping CPUs. So let's just schedule the worker thread on the current CPU when the current CPU is housekeeping CPU. Signed-off-by: Xiongfeng Wang <wangxiongfeng2(a)huawei.com> --- block/blk-mq.c | 11 ++++++++++- include/linux/sched/isolation.h | 2 ++ kernel/sched/isolation.c | 8 ++++++++ 3 files changed, 20 insertions(+), 1 deletion(-) diff --git a/block/blk-mq.c b/block/blk-mq.c index 1d1200afb771..d0dce9ef9499 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -23,6 +23,7 @@ #include <linux/sched/sysctl.h> #include <linux/sched/topology.h> #include <linux/sched/signal.h> +#include <linux/sched/isolation.h> #include <linux/delay.h> #include <linux/crash_dump.h> #include <linux/prefetch.h> @@ -1676,6 +1677,8 @@ static int blk_mq_hctx_next_cpu(struct blk_mq_hw_ctx *hctx) static void __blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async, unsigned long msecs) { + int work_cpu; + if (unlikely(blk_mq_hctx_stopped(hctx))) return; @@ -1697,7 +1700,13 @@ static void __blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async, if (!percpu_ref_tryget(&hctx->queue->q_usage_counter)) return; - kblockd_mod_delayed_work_on(blk_mq_hctx_next_cpu(hctx), &hctx->run_work, + if (enhanced_isolcpus && tick_nohz_full_enabled() && + housekeeping_cpu(smp_processor_id(), HK_FLAG_WQ)) + work_cpu = smp_processor_id(); + else + work_cpu = blk_mq_hctx_next_cpu(hctx); + + kblockd_mod_delayed_work_on(work_cpu, &hctx->run_work, msecs_to_jiffies(msecs)); percpu_ref_put(&hctx->queue->q_usage_counter); } diff --git a/include/linux/sched/isolation.h b/include/linux/sched/isolation.h index cc9f393e2a70..2f93081ad7a0 100644 --- a/include/linux/sched/isolation.h +++ b/include/linux/sched/isolation.h @@ -18,6 +18,7 @@ enum hk_flags { }; #ifdef CONFIG_CPU_ISOLATION +extern bool enhanced_isolcpus; DECLARE_STATIC_KEY_FALSE(housekeeping_overridden); extern int housekeeping_any_cpu(enum hk_flags flags); extern const struct cpumask *housekeeping_cpumask(enum hk_flags flags); @@ -28,6 +29,7 @@ extern void __init housekeeping_init(void); #else +#define enhanced_isolcpus 0 static inline int housekeeping_any_cpu(enum hk_flags flags) { return smp_processor_id(); diff --git a/kernel/sched/isolation.c b/kernel/sched/isolation.c index 5a6ea03f9882..785ef5201116 100644 --- a/kernel/sched/isolation.c +++ b/kernel/sched/isolation.c @@ -198,3 +198,11 @@ static int __init housekeeping_isolcpus_setup(char *str) return housekeeping_setup(str, flags); } __setup("isolcpus=", housekeeping_isolcpus_setup); + +bool enhanced_isolcpus; +static int __init enhanced_isolcpus_setup(char *str) +{ + enhanced_isolcpus = true; + return 0; +} +__setup("enhanced_isolcpus", enhanced_isolcpus_setup); -- 2.20.1

3 3

[PATCH OLK-5.10 v3 0/7] AGDI: Add driver for Arm Generic Diagnostic Dump and Reset device
by Qian Zou 06 Nov '23

06 Nov '23

From: “z00447436” <zouqian4(a)huawei.com> Some use-cases, such as system management, require the ability to generate a non-maskable event to the OS to request the OS kernel to perform a diagnostic dump and reset the system. Arm Generic Diagnostic Dump and Reset device enables a maintainer to request OS to perform a diagnostic dump and reset a system via SDEI event or an interrupt. This patch implements SDEI path and discards the interrupted context before proceeding to the crash kernel. D Scott Phillips (1): arm64: sdei: abort running SDEI handlers during crash Ilkka Koskinen (4): ACPI: tables: Add AGDI to the list of known table signatures ACPI: AGDI: Add driver for Arm Generic Diagnostic Dump and Reset device ACPICA: iASL: Add suppport for AGDI table ACPI: AGDI: Fix missing prototype warning for acpi_agdi_init() Jia He (1): EDAC/ghes: Make ghes_edac a proper module Shuai Xue (1): ACPI: APEI: explicit init of HEST and GHES in apci_init() arch/arm64/include/asm/sdei.h | 6 ++ arch/arm64/kernel/entry.S | 27 +++++++- arch/arm64/kernel/sdei.c | 3 + arch/arm64/kernel/smp.c | 8 +-- drivers/acpi/apei/ghes.c | 23 +++---- drivers/acpi/arm64/Kconfig | 10 +++ drivers/acpi/arm64/Makefile | 1 + drivers/acpi/arm64/agdi.c | 117 ++++++++++++++++++++++++++++++++++ drivers/acpi/bus.c | 4 ++ drivers/acpi/pci_root.c | 3 - drivers/acpi/tables.c | 2 +- drivers/edac/Kconfig | 4 +- drivers/edac/ghes_edac.c | 40 +++++++++++- drivers/firmware/Kconfig | 1 + drivers/firmware/arm_sdei.c | 32 ++++++---- include/acpi/actbl2.h | 20 ++++++ include/acpi/apei.h | 4 +- include/acpi/ghes.h | 25 -------- include/linux/acpi_agdi.h | 13 ++++ include/linux/arm_sdei.h | 4 ++ 20 files changed, 281 insertions(+), 66 deletions(-) create mode 100644 drivers/acpi/arm64/agdi.c create mode 100644 include/linux/acpi_agdi.h -- 2.33.0

2 8

[PATCH OLK-5.10 v2 0/7] AGDI: Add driver for Arm Generic Diagnostic Dump and Reset device
by Qian Zou 06 Nov '23

06 Nov '23

From: “z00447436” <zouqian4(a)huawei.com> Some use-cases, such as system management, require the ability to generate a non-maskable event to the OS to request the OS kernel to perform a diagnostic dump and reset the system. Arm Generic Diagnostic Dump and Reset device enables a maintainer to request OS to perform a diagnostic dump and reset a system via SDEI event or an interrupt. This patch implements SDEI path and discards the interrupted context before proceeding to the crash kernel. D Scott Phillips (1): arm64: sdei: abort running SDEI handlers during crash Ilkka Koskinen (4): ACPI: tables: Add AGDI to the list of known table signatures ACPI: AGDI: Add driver for Arm Generic Diagnostic Dump and Reset device ACPICA: iASL: Add suppport for AGDI table ACPI: AGDI: Fix missing prototype warning for acpi_agdi_init() Jia He (1): EDAC/ghes: Make ghes_edac a proper module Shuai Xue (1): ACPI: APEI: explicit init of HEST and GHES in apci_init() arch/arm64/include/asm/sdei.h | 6 ++ arch/arm64/kernel/entry.S | 27 +++++++- arch/arm64/kernel/sdei.c | 3 + arch/arm64/kernel/smp.c | 8 +-- drivers/acpi/apei/ghes.c | 23 +++---- drivers/acpi/arm64/Kconfig | 10 +++ drivers/acpi/arm64/Makefile | 1 + drivers/acpi/arm64/agdi.c | 117 ++++++++++++++++++++++++++++++++++ drivers/acpi/bus.c | 4 ++ drivers/acpi/pci_root.c | 3 - drivers/acpi/tables.c | 2 +- drivers/edac/Kconfig | 4 +- drivers/edac/ghes_edac.c | 40 +++++++++++- drivers/firmware/Kconfig | 1 + drivers/firmware/arm_sdei.c | 32 ++++++---- include/acpi/actbl2.h | 20 ++++++ include/acpi/apei.h | 4 +- include/acpi/ghes.h | 25 -------- include/linux/acpi_agdi.h | 13 ++++ include/linux/arm_sdei.h | 4 ++ 20 files changed, 281 insertions(+), 66 deletions(-) create mode 100644 drivers/acpi/arm64/agdi.c create mode 100644 include/linux/acpi_agdi.h -- 2.33.0

2 8

[PATCH openEuler-1.0-LTS] fs: lockd: avoid possible wrong NULL parameter
by ZhaoLong Wang 06 Nov '23

06 Nov '23

From: Su Hui <suhui(a)nfschina.com> stable inclusion from stable-v4.19.295 commit 35f0749756b848ad4f4a165ad6b1dfa8d0e45a96 category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I8E5Q5 CVE: NA Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… -------------------------------- [ Upstream commit de8d38cf44bac43e83bad28357ba84784c412752 ] clang's static analysis warning: fs/lockd/mon.c: line 293, column 2: Null pointer passed as 2nd argument to memory copy function. Assuming 'hostname' is NULL and calling 'nsm_create_handle()', this will pass NULL as 2nd argument to memory copy function 'memcpy()'. So return NULL if 'hostname' is invalid. Fixes: 77a3ef33e2de ("NSM: More clean up of nsm_get_handle()") Signed-off-by: Su Hui <suhui(a)nfschina.com> Reviewed-by: Nick Desaulniers <ndesaulniers(a)google.com> Reviewed-by: Jeff Layton <jlayton(a)kernel.org> Signed-off-by: Chuck Lever <chuck.lever(a)oracle.com> Signed-off-by: Sasha Levin <sashal(a)kernel.org> Signed-off-by: ZhaoLong Wang <wangzhaolong1(a)huawei.com> --- fs/lockd/mon.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/fs/lockd/mon.c b/fs/lockd/mon.c index 654594ef4f94..68a2eac548c3 100644 --- a/fs/lockd/mon.c +++ b/fs/lockd/mon.c @@ -275,6 +275,9 @@ static struct nsm_handle *nsm_create_handle(const struct sockaddr *sap, { struct nsm_handle *new; + if (!hostname) + return NULL; + new = kzalloc(sizeof(*new) + hostname_len + 1, GFP_KERNEL); if (unlikely(new == NULL)) return NULL; -- 2.34.3

2 1

[PATCH openEuler-1.0-LTS] crypto: hisilicon - qm obtain the mailbox config at one time
by wangyuan 06 Nov '23

06 Nov '23

From: Yu'an Wang <wangyuan46(a)huawei.com> driver inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I8E0N4 CVE: NA -------------------------------- The mailbox configuration needs to be obtained from the hardware at one time. If the mailbox configuration is obtained for multiple times, the read value may be incorrect. Use the instruction to read mailbox data instead of readl. Signed-off-by: Yu'an Wang <wangyuan46(a)huawei.com> --- drivers/crypto/hisilicon/qm.c | 187 +++++++++++++++++++++++++--------- 1 file changed, 137 insertions(+), 50 deletions(-) diff --git a/drivers/crypto/hisilicon/qm.c b/drivers/crypto/hisilicon/qm.c index f2706dc0d55e..64e8ef1bf8e7 100644 --- a/drivers/crypto/hisilicon/qm.c +++ b/drivers/crypto/hisilicon/qm.c @@ -37,6 +37,7 @@ #define QM_MB_OP_SHIFT 14 #define QM_MB_CMD_DATA_ADDR_L 0x304 #define QM_MB_CMD_DATA_ADDR_H 0x308 +#define QM_MB_STATUS_MASK GENMASK(12, 9) /* sqc shift */ #define QM_SQ_HOP_NUM_SHIFT 0 @@ -113,7 +114,7 @@ #define QM_SQC_VFT_NUM_MASK_v2 0x3ff #define QM_DFX_CNT_CLR_CE 0x100118 -#define QM_IN_IDLE_ST_REG 0x1040e4 +#define QM_IN_IDLE_ST_REG 0x1040e4 #define QM_ABNORMAL_INT_SOURCE 0x100000 #define QM_ABNORMAL_INT_MASK 0x100004 @@ -169,6 +170,8 @@ #define TASK_TIMEOUT 10000 #define WAIT_PERIOD 20 +#define QM_MB_WAIT_READY_CNT 10 +#define QM_MB_MAX_WAIT_CNT 21000 #define WAIT_PERIOD_US_MAX 200 #define WAIT_PERIOD_US_MIN 100 #define REMOVE_WAIT_DELAY 10 @@ -382,16 +385,6 @@ static bool qm_qp_avail_state(struct hisi_qm *qm, struct hisi_qp *qp, return avail; } -/* return 0 mailbox ready, -ETIMEDOUT hardware timeout */ -static int qm_wait_mb_ready(struct hisi_qm *qm) -{ - u32 val; - - return readl_relaxed_poll_timeout(qm->io_base + QM_MB_CMD_SEND_BASE, - val, !((val >> QM_MB_BUSY_SHIFT) & - 0x1), POLL_PERIOD, POLL_TIMEOUT); -} - /* 128 bit should be wrote to hardware at one time to trigger a mailbox */ static void qm_mb_write(struct hisi_qm *qm, const void *src) { @@ -400,7 +393,7 @@ static void qm_mb_write(struct hisi_qm *qm, const void *src) asm volatile("ldp %0, %1, %3\n" "stp %0, %1, %2\n" - "dsb sy\n" + "dmb oshst\n" : "=&r" (tmp0), "=&r" (tmp1), "+Q" (*((char __iomem *)fun_base)) @@ -408,45 +401,142 @@ static void qm_mb_write(struct hisi_qm *qm, const void *src) : "memory"); } -static int qm_mb(struct hisi_qm *qm, u8 cmd, dma_addr_t dma_addr, u16 queue, - bool op) +/* 128 bit should be read from hardware at one time*/ +static void qm_mb_read(struct hisi_qm *qm, void *dst) +{ + const void __iomem *fun_base = qm->io_base + QM_MB_CMD_SEND_BASE; + unsigned long tmp0 = 0, tmp1 = 0; + + asm volatile("ldp %0, %1, %3\n" + "stp %0, %1, %2\n" + "dmb oshst\n" + : "=&r" (tmp0), + "=&r" (tmp1), + "+Q" (*((char *)dst)) + : "Q" (*((char __iomem *)fun_base)) + : "memory"); +} + +static int qm_wait_mb_ready(struct hisi_qm *qm) { struct qm_mailbox mailbox; - int ret = 0; + int i = 0; + + while (i++ < QM_MB_WAIT_READY_CNT) { + qm_mb_read(qm, &mailbox); + if (!((le16_to_cpu(mailbox.w0) >> QM_MB_BUSY_SHIFT) & 0x1)) + return 0; - dev_dbg(&qm->pdev->dev, "QM mailbox request to q%u: %u\n", queue, cmd); + usleep_range(WAIT_PERIOD_US_MIN, WAIT_PERIOD_US_MAX); + } - mailbox.w0 = cpu_to_le16(cmd | - (op ? 0x1 << QM_MB_OP_SHIFT : 0) | - (0x1 << QM_MB_BUSY_SHIFT)); - mailbox.queue_num = cpu_to_le16(queue); - mailbox.base_l = cpu_to_le32(lower_32_bits(dma_addr)); - mailbox.base_h = cpu_to_le32(upper_32_bits(dma_addr)); - mailbox.rsvd = 0; + dev_err(&qm->pdev->dev, "QM mailbox is busy to start!\n"); - mutex_lock(&qm->mailbox_lock); + return -EBUSY; +} + +static int qm_wait_mb_finish(struct hisi_qm *qm, struct qm_mailbox *mailbox) +{ + int i = 0; + + while (++i) { + qm_mb_read(qm, mailbox); + if (!((le16_to_cpu(mailbox->w0) >> QM_MB_BUSY_SHIFT) & 0x1)) + break; + + if (i > QM_MB_MAX_WAIT_CNT) { + dev_err(&qm->pdev->dev, + "QM mailbox operation timeout!\n"); + return -ETIMEDOUT; + } - if (unlikely(qm_wait_mb_ready(qm))) { - ret = -EBUSY; - dev_err(&qm->pdev->dev, "QM mailbox is busy to start!\n"); - goto busy_unlock; + usleep_range(WAIT_PERIOD_US_MIN, WAIT_PERIOD_US_MAX); + } + + if (le16_to_cpu(mailbox->w0) & QM_MB_STATUS_MASK) { + dev_err(&qm->pdev->dev, "QM mailbox operation failed!\n"); + return -EIO; } - qm_mb_write(qm, &mailbox); + return 0; +} - if (unlikely(qm_wait_mb_ready(qm))) { - ret = -EBUSY; - dev_err(&qm->pdev->dev, "QM mailbox operation timeout!\n"); - goto busy_unlock; +static int qm_mb_nolock(struct hisi_qm *qm, struct qm_mailbox *mailbox) +{ + int ret; + + ret = qm_wait_mb_ready(qm); + if (ret) + goto mb_busy; + + qm_mb_write(qm, mailbox); + + ret = qm_wait_mb_finish(qm, mailbox); + if (ret) + goto mb_busy; + + return 0; + +mb_busy: + atomic64_inc(&qm->debug.dfx.mb_err_cnt); + return ret; +} + +static void qm_mb_pre_init(struct qm_mailbox *mailbox, u8 cmd, + u64 base, u16 queue, bool op) +{ + mailbox->w0 = cpu_to_le16((cmd) | + ((op) ? 0x1 << QM_MB_OP_SHIFT : 0) | + (0x1 << QM_MB_BUSY_SHIFT)); + mailbox->queue_num = cpu_to_le16(queue); + mailbox->base_l = cpu_to_le32(lower_32_bits(base)); + mailbox->base_h = cpu_to_le32(upper_32_bits(base)); + mailbox->rsvd = 0; +} + +static int qm_check_dev_error(struct hisi_qm *qm); +static int hisi_qm_mb_write(struct hisi_qm *qm, u8 cmd, dma_addr_t dma_addr, + u16 queue, bool op) +{ + struct hisi_qm *pf_qm = pci_get_drvdata(pci_physfn(qm->pdev)); + struct qm_mailbox mailbox; + int ret; + + dev_dbg(&qm->pdev->dev, "QM mailbox request to q%u: %u.\n", queue, cmd); + + /* No need to judge if master OOO is blocked. */ + if (qm_check_dev_error(pf_qm)) { + dev_err(&qm->pdev->dev, + "QM mailbox operation failed since qm is stop!\n"); + return -EIO; } -busy_unlock: + qm_mb_pre_init(&mailbox, cmd, dma_addr, queue, op); + mutex_lock(&qm->mailbox_lock); + ret = qm_mb_nolock(qm, &mailbox); mutex_unlock(&qm->mailbox_lock); - if (ret) - atomic64_inc(&qm->debug.dfx.mb_err_cnt); + return ret; } +static int hisi_qm_mb_read(struct hisi_qm *qm, u64 *base, u8 cmd, u16 queue) +{ + struct qm_mailbox mailbox; + int ret; + + qm_mb_pre_init(&mailbox, cmd, 0, queue, 1); + mutex_lock(&qm->mailbox_lock); + ret = qm_mb_nolock(qm, &mailbox); + mutex_unlock(&qm->mailbox_lock); + if (ret) + return ret; + + *base = le32_to_cpu(mailbox.base_l) | + ((u64)le32_to_cpu(mailbox.base_h) << 32); + + return 0; +} + static void qm_db_v1(struct hisi_qm *qm, u16 qn, u8 cmd, u16 index, u8 priority) { u64 doorbell; @@ -826,12 +916,10 @@ static int qm_get_vft_v2(struct hisi_qm *qm, u32 *base, u32 *number) u64 sqc_vft; int ret; - ret = qm_mb(qm, QM_MB_CMD_SQC_VFT_V2, 0, 0, 1); + ret = hisi_qm_mb_read(qm, &sqc_vft, QM_MB_CMD_SQC_VFT_V2, 0); if (ret) return ret; - sqc_vft = readl(qm->io_base + QM_MB_CMD_DATA_ADDR_L) | - ((u64)readl(qm->io_base + QM_MB_CMD_DATA_ADDR_H) << 32); *base = QM_SQC_VFT_BASE_MASK_V2 & (sqc_vft >> QM_SQC_VFT_BASE_SHIFT_V2); *number = (QM_SQC_VFT_NUM_MASK_v2 & (sqc_vft >> QM_SQC_VFT_NUM_SHIFT_V2)) + 1; @@ -1144,12 +1232,12 @@ static int dump_show(struct hisi_qm *qm, void *info, static int qm_dump_sqc_raw(struct hisi_qm *qm, dma_addr_t dma_addr, u16 qp_id) { - return qm_mb(qm, QM_MB_CMD_SQC, dma_addr, qp_id, 1); + return hisi_qm_mb_write(qm, QM_MB_CMD_SQC, dma_addr, qp_id, 1); } static int qm_dump_cqc_raw(struct hisi_qm *qm, dma_addr_t dma_addr, u16 qp_id) { - return qm_mb(qm, QM_MB_CMD_CQC, dma_addr, qp_id, 1); + return hisi_qm_mb_write(qm, QM_MB_CMD_CQC, dma_addr, qp_id, 1); } static int qm_sqc_dump(struct hisi_qm *qm, const char *s) @@ -1260,7 +1348,7 @@ static int qm_eqc_dump(struct hisi_qm *qm, char *s) if (IS_ERR(eqc)) return PTR_ERR(eqc); - ret = qm_mb(qm, QM_MB_CMD_EQC, eqc_dma, 0, 1); + ret = hisi_qm_mb_write(qm, QM_MB_CMD_EQC, eqc_dma, 0, 1); if (ret) goto mailbox_fail; @@ -1289,7 +1377,7 @@ static int qm_aeqc_dump(struct hisi_qm *qm, char *s) if (IS_ERR(aeqc)) return PTR_ERR(aeqc); - ret = qm_mb(qm, QM_MB_CMD_AEQC, aeqc_dma, 0, 1); + ret = hisi_qm_mb_write(qm, QM_MB_CMD_AEQC, aeqc_dma, 0, 1); if (ret) goto mailbox_fail; @@ -1834,7 +1922,7 @@ static int qm_sq_ctx_cfg(struct hisi_qp *qp, int qp_id, int pasid) sqc->cq_num = cpu_to_le16(qp_id); sqc->w13 = cpu_to_le16(QM_MK_SQC_W13(0, 1, qp->alg_type)); - ret = qm_mb(qm, QM_MB_CMD_SQC, sqc_dma, qp_id, 0); + ret = hisi_qm_mb_write(qm, QM_MB_CMD_SQC, sqc_dma, qp_id, 0); dma_unmap_single(dev, sqc_dma, sizeof(struct qm_sqc), DMA_TO_DEVICE); kfree(sqc); @@ -1874,7 +1962,7 @@ static int qm_cq_ctx_cfg(struct hisi_qp *qp, int qp_id, int pasid) cqc->dw6 = cpu_to_le32(1 << QM_CQ_PHASE_SHIFT | qp->c_flag << QM_CQ_FLAG_SHIFT); - ret = qm_mb(qm, QM_MB_CMD_CQC, cqc_dma, qp_id, 0); + ret = hisi_qm_mb_write(qm, QM_MB_CMD_CQC, cqc_dma, qp_id, 0); dma_unmap_single(dev, cqc_dma, sizeof(struct qm_cqc), DMA_TO_DEVICE); kfree(cqc); @@ -2894,7 +2982,7 @@ static int qm_eq_ctx_cfg(struct hisi_qm *qm) if (qm->ver == QM_HW_V1) eqc->dw3 = cpu_to_le32(QM_EQE_AEQE_SIZE); eqc->dw6 = cpu_to_le32((QM_EQ_DEPTH - 1) | (1 << QM_EQC_PHASE_SHIFT)); - ret = qm_mb(qm, QM_MB_CMD_EQC, eqc_dma, 0, 0); + ret = hisi_qm_mb_write(qm, QM_MB_CMD_EQC, eqc_dma, 0, 0); dma_unmap_single(dev, eqc_dma, sizeof(struct qm_eqc), DMA_TO_DEVICE); kfree(eqc); @@ -2921,7 +3009,7 @@ static int qm_aeq_ctx_cfg(struct hisi_qm *qm) aeqc->base_l = cpu_to_le32(lower_32_bits(qm->aeqe_dma)); aeqc->base_h = cpu_to_le32(upper_32_bits(qm->aeqe_dma)); aeqc->dw6 = cpu_to_le32((QM_Q_DEPTH - 1) | (1 << QM_EQC_PHASE_SHIFT)); - ret = qm_mb(qm, QM_MB_CMD_AEQC, aeqc_dma, 0, 0); + ret = hisi_qm_mb_write(qm, QM_MB_CMD_AEQC, aeqc_dma, 0, 0); dma_unmap_single(dev, aeqc_dma, sizeof(struct qm_aeqc), DMA_TO_DEVICE); kfree(aeqc); @@ -2969,11 +3057,11 @@ static int __hisi_qm_start(struct hisi_qm *qm) if (ret) return ret; - ret = qm_mb(qm, QM_MB_CMD_SQC_BT, qm->sqc_dma, 0, 0); + ret = hisi_qm_mb_write(qm, QM_MB_CMD_SQC_BT, qm->sqc_dma, 0, 0); if (ret) return ret; - ret = qm_mb(qm, QM_MB_CMD_CQC_BT, qm->cqc_dma, 0, 0); + ret = hisi_qm_mb_write(qm, QM_MB_CMD_CQC_BT, qm->cqc_dma, 0, 0); if (ret) return ret; @@ -3194,7 +3282,6 @@ static ssize_t qm_status_read(struct file *filp, char __user *buffer, return (*pos = len); } - static const struct file_operations qm_status_fops = { .owner = THIS_MODULE, .open = simple_open, -- 2.30.0

2 1

[PATCH V5 OLK-5.10] soc: hisilicon: hisi_hbmdev: Add hbm acls repair and query methods
by Zhang Zekun 06 Nov '23

06 Nov '23

hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I8CCP5 ---------------------------------- Hbm memory device add support for acls hot repair. The patch add two methods for userpace: - query a paddr if it support acls repair - repair a paddr in hbm memory device The feature of ACLS hot repair can help to fix a memory error from userspace by passing through the error physical address to HBM hardware. Signed-off-by: Zhang Zekun <zhangzekun11(a)huawei.com> --- drivers/soc/hisilicon/Kconfig | 10 +++ drivers/soc/hisilicon/hisi_hbmdev.c | 134 ++++++++++++++++++++++++++++ 2 files changed, 144 insertions(+) diff --git a/drivers/soc/hisilicon/Kconfig b/drivers/soc/hisilicon/Kconfig index 383375f20cac..45754528873e 100644 --- a/drivers/soc/hisilicon/Kconfig +++ b/drivers/soc/hisilicon/Kconfig @@ -44,4 +44,14 @@ config HISI_HBMCACHE To compile the driver as a module, choose M here: the module will be called hisi_hbmcache. +config HISI_HBMDEV_ACLS + bool "Add support for HISI ACLS repair" + depends on HISI_HBMDEV + help + Add ACLS support for hbm device, which can be used to query and + repair hardware error in HBM devices. This feature need to work with + hardware firmwares. + + If not sure say no. + endmenu diff --git a/drivers/soc/hisilicon/hisi_hbmdev.c b/drivers/soc/hisilicon/hisi_hbmdev.c index 5b6b1618148c..a9cc78bde81b 100644 --- a/drivers/soc/hisilicon/hisi_hbmdev.c +++ b/drivers/soc/hisilicon/hisi_hbmdev.c @@ -11,6 +11,7 @@ #include <linux/node.h> #include <linux/arch_topology.h> #include <linux/memory_hotplug.h> +#include <linux/mm.h> #include "hisi_internal.h" @@ -25,6 +26,9 @@ struct cdev_node { struct memory_dev { struct kobject *memdev_kobj; struct kobject *topo_kobj; +#ifdef CONFIG_HISI_HBMDEV_ACLS + struct kobject *acls_kobj; +#endif struct cdev_node cdev_list; nodemask_t cluster_cpumask[MAX_NUMNODES]; }; @@ -85,6 +89,134 @@ static void memory_topo_init(void) kobject_put(mdev->topo_kobj); } +#ifdef CONFIG_HISI_HBMDEV_ACLS +static struct acpi_device *paddr_to_acpi_device(u64 paddr) +{ + unsigned long pfn; + int nid; + + pfn = __phys_to_pfn(paddr); + if (!pfn_valid(pfn)) + return NULL; + + nid = pfn_to_nid(pfn); + if (nid < 0 && nid >= MAX_NUMNODES) + return NULL; + + return hotplug_mdev[nid]; +} + +static ssize_t acls_query_store(struct kobject *kobj, + struct kobj_attribute *attr, + const char *buf, size_t count) +{ + struct acpi_object_list arg_list; + struct acpi_device *adev; + union acpi_object obj; + acpi_status status; + u64 paddr, res; + + if (kstrtoull(buf, 16, &paddr)) + return -EINVAL; + + adev = paddr_to_acpi_device(paddr); + if (!adev) + return -EINVAL; + + obj.type = ACPI_TYPE_INTEGER; + obj.integer.value = paddr; + arg_list.count = 1; + arg_list.pointer = &obj; + + status = acpi_evaluate_integer(adev->handle, "AQRY", &arg_list, &res); + if (ACPI_FAILURE(status)) + return -ENODEV; + + /* AQRY will return a positive error code to represent error status */ + if (IS_ERR_VALUE(-res)) + return -res; + else if (res) + return -ENODEV; + + return count; +} + +static struct kobj_attribute acls_query_store_attribute = + __ATTR(acls_query, 0200, NULL, acls_query_store); + +static ssize_t acls_repair_store(struct kobject *kobj, + struct kobj_attribute *attr, + const char *buf, size_t count) +{ + struct acpi_object_list arg_list; + struct acpi_device *adev; + union acpi_object obj; + acpi_status status; + u64 paddr, res; + + if (kstrtoull(buf, 16, &paddr)) + return -EINVAL; + + adev = paddr_to_acpi_device(paddr); + if (!adev) + return -EINVAL; + + obj.type = ACPI_TYPE_INTEGER; + obj.integer.value = paddr; + arg_list.count = 1; + arg_list.pointer = &obj; + + status = acpi_evaluate_integer(adev->handle, "AREP", &arg_list, &res); + if (ACPI_FAILURE(status)) + return -ENODEV; + + /* AREP will return a positive error code to represent error status */ + if (IS_ERR_VALUE(-res)) + return -res; + else if (res) + return -ENODEV; + + return count; +} +static struct kobj_attribute acls_repair_store_attribute = + __ATTR(acls_repair, 0200, NULL, acls_repair_store); + +static struct attribute *acls_attrs[] = { + &acls_query_store_attribute.attr, + &acls_repair_store_attribute.attr, + NULL, +}; + +static struct attribute_group acls_attr_group = { + .attrs = acls_attrs, +}; + +static void acls_init(void) +{ + int ret = -ENOMEM; + + mdev->acls_kobj = kobject_create_and_add("acls", mdev->memdev_kobj); + if (!mdev->acls_kobj) + goto out; + + ret = sysfs_create_group(mdev->acls_kobj, &acls_attr_group); + if (ret) + kobject_put(mdev->acls_kobj); + +out: + if (ret) + pr_err("ACLS hot repair is not enabled\n"); +} + +static void acls_remove(void) +{ + kobject_put(mdev->acls_kobj); +} +#else +static void acls_init(void) {} +static void acls_remove(void) {} +#endif + static int get_pxm(struct acpi_device *acpi_device, void *arg) { acpi_handle handle = acpi_device->handle; @@ -284,6 +416,7 @@ static int __init mdev_init(void) } memory_topo_init(); + acls_init(); return ret; } module_init(mdev_init); @@ -293,6 +426,7 @@ static void __exit mdev_exit(void) container_remove(); kobject_put(mdev->memdev_kobj); kobject_put(mdev->topo_kobj); + acls_remove(); kfree(mdev); } module_exit(mdev_exit); -- 2.17.1

3 2

[PATCH v4 OLK-5.10 00/29] Introduce some vdpa ops to support vdpa device live migrate
by Jiang Dongxu 06 Nov '23

06 Nov '23

From: jiangdongxu <jiangdongxu1(a)huawei.com> Patch 1-17: some bugfix and ops intruduced by upstream Patch 18-19: introduce vdpa device logging ops Patch 20-21: introduce vdpa device state ops Patch 22-23: introduce vdpa device migrate state ops Patch 24: introduce new vhost feature BYTEPMAPLOG Patch 25: export iommu_get_resv_regions/iommu_set_resv_regions Patch 26-27: some optimization about vhost-vdpa Patch 28: add vdpa/vhost-vdpa build config Patch 29: fix vhost-vdpa compile warnings Arnaldo Carvalho de Melo (1): tools include UAPI: Sync linux/vhost.h with the kernel sources Cindy Lu (2): vhost_vdpa: fix the crash in unmap a large memory vhost_vdpa: fix unmap process in no-batch mode Eugenio Pérez (1): vdpa: add get_backend_features vdpa operation Gautam Dawar (1): vhost-vdpa: free iommu domain after last use during cleanup Greg Kroah-Hartman (1): vhost-vdpa: vhost_vdpa_alloc_domain() should be using a const struct bus_type * Jason Gunthorpe (1): PCI/IOV: Add pci_iov_vf_id() to get VF index Sebastien Boeuf (3): vdpa: Add resume operation vhost-vdpa: Introduce RESUME backend feature bit vhost-vdpa: uAPI to resume the device Shannon Nelson (2): vhost_vdpa: tell vqs about the negotiated vhost_vdpa: support PACKED when setting-getting vring_base Shunsuke Mie (1): virtio: fix virtio transitional ids Stefano Garzarella (3): vhost-vdpa: fix an iotlb memory leak vdpa: add bind_mm/unbind_mm callbacks vhost-vdpa: use bind_mm/unbind_mm device callbacks Zhu Lingshan (1): virtio: update virtio id table, add transitional ids jiangdongxu (12): vdpa: add log operations vhost-vdpa: add uAPI for logging vdpa: add device state operations vhost-vdpa: add uAPI for device buffer vdpa: add vdpa device migration status ops vhost-vdpa: add uAPI for device migration status vhost: add VHOST feature VHOST_BACKEND_F_BYTEMAPLOG export iommu_get_resv_regions and iommu_set_resv_regions vhost-vdpa: Allow transparent MSI IOV vhost-vdpa: fix msi irq request err arm64: openeuler_defconfig: add VDPA config vhost-vdpa: fix compile warnings arch/arm64/configs/openeuler_defconfig | 6 +- drivers/iommu/iommu.c | 2 + drivers/pci/iov.c | 14 + drivers/vhost/vdpa.c | 383 ++++++++++++++++++++++--- include/linux/pci.h | 8 +- include/linux/vdpa.h | 60 +++- include/uapi/linux/vhost.h | 20 ++ include/uapi/linux/vhost_types.h | 21 ++ include/uapi/linux/virtio_ids.h | 12 + tools/include/uapi/linux/vhost.h | 8 + 10 files changed, 489 insertions(+), 45 deletions(-) -- 2.27.0

2 30

[PATCH V4 OLK-5.10] soc: hisilicon: hisi_hbmdev: Add hbm acls repair and query methods
by Zhang Zekun 06 Nov '23

06 Nov '23

hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I8CCP5 ---------------------------------- Hbm memory device add support for acls hot repair. The patch add two methods for userpace: - query a paddr if it support acls repair - repair a paddr in hbm memory device The feature of ACLS hot repair can help to fix a memory error from userspace by passing through the error physical address to HBM hardware. Signed-off-by: Zhang Zekun <zhangzekun11(a)huawei.com> --- v4 - remove "Offering HULK" to pass CI patch format check v3 - Add #include <linux/mm.h> to avoid compile failure in arch x86_64 v2: - Change "u64 pfn" ot "unsigned long pfn" to avoid compile error - Add two stub to pretiffy the code drivers/soc/hisilicon/Kconfig | 10 +++ drivers/soc/hisilicon/hisi_hbmdev.c | 132 ++++++++++++++++++++++++++++ 2 files changed, 142 insertions(+) diff --git a/drivers/soc/hisilicon/Kconfig b/drivers/soc/hisilicon/Kconfig index 383375f20cac..45754528873e 100644 --- a/drivers/soc/hisilicon/Kconfig +++ b/drivers/soc/hisilicon/Kconfig @@ -44,4 +44,14 @@ config HISI_HBMCACHE To compile the driver as a module, choose M here: the module will be called hisi_hbmcache. +config HISI_HBMDEV_ACLS + bool "Add support for HISI ACLS repair" + depends on HISI_HBMDEV + help + Add ACLS support for hbm device, which can be used to query and + repair hardware error in HBM devices. This feature need to work with + hardware firmwares. + + If not sure say no. + endmenu diff --git a/drivers/soc/hisilicon/hisi_hbmdev.c b/drivers/soc/hisilicon/hisi_hbmdev.c index 5b6b1618148c..8348abc6144c 100644 --- a/drivers/soc/hisilicon/hisi_hbmdev.c +++ b/drivers/soc/hisilicon/hisi_hbmdev.c @@ -11,6 +11,7 @@ #include <linux/node.h> #include <linux/arch_topology.h> #include <linux/memory_hotplug.h> +#include <linux/mm.h> #include "hisi_internal.h" @@ -25,6 +26,9 @@ struct cdev_node { struct memory_dev { struct kobject *memdev_kobj; struct kobject *topo_kobj; +#ifdef CONFIG_HISI_HBMDEV_ACLS + struct kobject *acls_kobj; +#endif struct cdev_node cdev_list; nodemask_t cluster_cpumask[MAX_NUMNODES]; }; @@ -85,6 +89,132 @@ static void memory_topo_init(void) kobject_put(mdev->topo_kobj); } +#ifdef CONFIG_HISI_HBMDEV_ACLS +static struct acpi_device *paddr_to_acpi_device(u64 paddr) +{ + unsigned long pfn; + int nid; + + pfn = __phys_to_pfn(paddr); + if (!pfn_valid(pfn)) + return NULL; + + nid = pfn_to_nid(pfn); + if (nid < 0 && nid >= MAX_NUMNODES) + return NULL; + + return hotplug_mdev[nid]; +} + +static ssize_t acls_query_store(struct kobject *kobj, + struct kobj_attribute *attr, + const char *buf, size_t count) +{ + struct acpi_object_list arg_list; + struct acpi_device *adev; + union acpi_object obj; + acpi_status status; + u64 paddr, res; + + if (kstrtoull(buf, 16, &paddr)) + return -EINVAL; + + adev = paddr_to_acpi_device(paddr); + if (!adev) + return -EINVAL; + + obj.type = ACPI_TYPE_INTEGER; + obj.integer.value = paddr; + arg_list.count = 1; + arg_list.pointer = &obj; + + status = acpi_evaluate_integer(adev->handle, "AQRY", &arg_list, &res); + if (ACPI_FAILURE(status)) + return -ENODEV; + + if (IS_ERR_VALUE(-res)) + return -res; + else if (res) + return -ENODEV; + + return count; +} + +static struct kobj_attribute acls_query_store_attribute = + __ATTR(acls_query, 0200, NULL, acls_query_store); + +static ssize_t acls_repair_store(struct kobject *kobj, + struct kobj_attribute *attr, + const char *buf, size_t count) +{ + struct acpi_object_list arg_list; + struct acpi_device *adev; + union acpi_object obj; + acpi_status status; + u64 paddr, res; + + if (kstrtoull(buf, 16, &paddr)) + return -EINVAL; + + adev = paddr_to_acpi_device(paddr); + if (!adev) + return -EINVAL; + + obj.type = ACPI_TYPE_INTEGER; + obj.integer.value = paddr; + arg_list.count = 1; + arg_list.pointer = &obj; + + status = acpi_evaluate_integer(adev->handle, "AREP", &arg_list, &res); + if (ACPI_FAILURE(status)) + return -ENODEV; + + if (IS_ERR_VALUE(-res)) + return -res; + else if (res) + return -ENODEV; + + return count; +} +static struct kobj_attribute acls_repair_store_attribute = + __ATTR(acls_repair, 0200, NULL, acls_repair_store); + +static struct attribute *acls_attrs[] = { + &acls_query_store_attribute.attr, + &acls_repair_store_attribute.attr, + NULL, +}; + +static struct attribute_group acls_attr_group = { + .attrs = acls_attrs, +}; + +static void acls_init(void) +{ + int ret = -ENOMEM; + + mdev->acls_kobj = kobject_create_and_add("acls", mdev->memdev_kobj); + if (!mdev->acls_kobj) + goto out; + + ret = sysfs_create_group(mdev->acls_kobj, &acls_attr_group); + if (ret) + kobject_put(mdev->acls_kobj); + +out: + if (ret) + pr_err("ACLS hot repair is not enabled\n"); +} + +static void acls_remove(void) +{ + kobject_put(mdev->acls_kobj); +} +#else +static void acls_init(void) {} +static void acls_remove(void) {} +#endif + static int get_pxm(struct acpi_device *acpi_device, void *arg) { acpi_handle handle = acpi_device->handle; @@ -284,6 +414,7 @@ static int __init mdev_init(void) } memory_topo_init(); + acls_init(); return ret; } module_init(mdev_init); @@ -293,6 +424,7 @@ static void __exit mdev_exit(void) container_remove(); kobject_put(mdev->memdev_kobj); kobject_put(mdev->topo_kobj); + acls_remove(); kfree(mdev); } module_exit(mdev_exit); -- 2.17.1

2 1