IB/hfi1: Define variables as unsigned long to fix KASAN warning
authorIra Weiny <ira.weiny@intel.com>
Wed, 11 Sep 2019 11:30:53 +0000 (07:30 -0400)
committerJason Gunthorpe <jgg@mellanox.com>
Fri, 13 Sep 2019 19:59:55 +0000 (16:59 -0300)
Define the working variables to be unsigned long to be compatible with
for_each_set_bit and change types as needed.

While we are at it remove unused variables from a couple of functions.

This was found because of the following KASAN warning:
 ==================================================================
   BUG: KASAN: stack-out-of-bounds in find_first_bit+0x19/0x70
   Read of size 8 at addr ffff888362d778d0 by task kworker/u308:2/1889

   CPU: 21 PID: 1889 Comm: kworker/u308:2 Tainted: G W         5.3.0-rc2-mm1+ #2
   Hardware name: Intel Corporation W2600CR/W2600CR, BIOS SE5C600.86B.02.04.0003.102320141138 10/23/2014
   Workqueue: ib-comp-unb-wq ib_cq_poll_work [ib_core]
   Call Trace:
    dump_stack+0x9a/0xf0
    ? find_first_bit+0x19/0x70
    print_address_description+0x6c/0x332
    ? find_first_bit+0x19/0x70
    ? find_first_bit+0x19/0x70
    __kasan_report.cold.6+0x1a/0x3b
    ? find_first_bit+0x19/0x70
    kasan_report+0xe/0x12
    find_first_bit+0x19/0x70
    pma_get_opa_portstatus+0x5cc/0xa80 [hfi1]
    ? ret_from_fork+0x3a/0x50
    ? pma_get_opa_port_ectrs+0x200/0x200 [hfi1]
    ? stack_trace_consume_entry+0x80/0x80
    hfi1_process_mad+0x39b/0x26c0 [hfi1]
    ? __lock_acquire+0x65e/0x21b0
    ? clear_linkup_counters+0xb0/0xb0 [hfi1]
    ? check_chain_key+0x1d7/0x2e0
    ? lock_downgrade+0x3a0/0x3a0
    ? match_held_lock+0x2e/0x250
    ib_mad_recv_done+0x698/0x15e0 [ib_core]
    ? clear_linkup_counters+0xb0/0xb0 [hfi1]
    ? ib_mad_send_done+0xc80/0xc80 [ib_core]
    ? mark_held_locks+0x79/0xa0
    ? _raw_spin_unlock_irqrestore+0x44/0x60
    ? rvt_poll_cq+0x1e1/0x340 [rdmavt]
    __ib_process_cq+0x97/0x100 [ib_core]
    ib_cq_poll_work+0x31/0xb0 [ib_core]
    process_one_work+0x4ee/0xa00
    ? pwq_dec_nr_in_flight+0x110/0x110
    ? do_raw_spin_lock+0x113/0x1d0
    worker_thread+0x57/0x5a0
    ? process_one_work+0xa00/0xa00
    kthread+0x1bb/0x1e0
    ? kthread_create_on_node+0xc0/0xc0
    ret_from_fork+0x3a/0x50

   The buggy address belongs to the page:
   page:ffffea000d8b5dc0 refcount:0 mapcount:0 mapping:0000000000000000 index:0x0
   flags: 0x17ffffc0000000()
   raw: 0017ffffc0000000 0000000000000000 ffffea000d8b5dc8 0000000000000000
   raw: 0000000000000000 0000000000000000 00000000ffffffff 0000000000000000
   page dumped because: kasan: bad access detected

   addr ffff888362d778d0 is located in stack of task kworker/u308:2/1889 at offset 32 in frame:
    pma_get_opa_portstatus+0x0/0xa80 [hfi1]

   this frame has 1 object:
    [32, 36) 'vl_select_mask'

   Memory state around the buggy address:
    ffff888362d77780: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    ffff888362d77800: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
   >ffff888362d77880: 00 00 00 00 00 00 f1 f1 f1 f1 04 f2 f2 f2 00 00
                                                    ^
    ffff888362d77900: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    ffff888362d77980: 00 00 00 00 00 00 00 00 f1 f1 f1 f1 04 f2 f2 f2

 ==================================================================

Cc: <stable@vger.kernel.org>
Fixes: 7724105686e7 ("IB/hfi1: add driver files")
Link: https://lore.kernel.org/r/20190911113053.126040.47327.stgit@awfm-01.aw.intel.com
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
drivers/infiniband/hw/hfi1/mad.c

index 184dba3..d8ff063 100644 (file)
@@ -2326,7 +2326,7 @@ struct opa_port_status_req {
        __be32 vl_select_mask;
 };
 
-#define VL_MASK_ALL            0x000080ff
+#define VL_MASK_ALL            0x00000000000080ffUL
 
 struct opa_port_status_rsp {
        __u8 port_num;
@@ -2625,15 +2625,14 @@ static int pma_get_opa_classportinfo(struct opa_pma_mad *pmp,
 }
 
 static void a0_portstatus(struct hfi1_pportdata *ppd,
-                         struct opa_port_status_rsp *rsp, u32 vl_select_mask)
+                         struct opa_port_status_rsp *rsp)
 {
        if (!is_bx(ppd->dd)) {
                unsigned long vl;
                u64 sum_vl_xmit_wait = 0;
-               u32 vl_all_mask = VL_MASK_ALL;
+               unsigned long vl_all_mask = VL_MASK_ALL;
 
-               for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
-                                8 * sizeof(vl_all_mask)) {
+               for_each_set_bit(vl, &vl_all_mask, BITS_PER_LONG) {
                        u64 tmp = sum_vl_xmit_wait +
                                  read_port_cntr(ppd, C_TX_WAIT_VL,
                                                 idx_from_vl(vl));
@@ -2730,12 +2729,12 @@ static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
                (struct opa_port_status_req *)pmp->data;
        struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
        struct opa_port_status_rsp *rsp;
-       u32 vl_select_mask = be32_to_cpu(req->vl_select_mask);
+       unsigned long vl_select_mask = be32_to_cpu(req->vl_select_mask);
        unsigned long vl;
        size_t response_data_size;
        u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
        u8 port_num = req->port_num;
-       u8 num_vls = hweight32(vl_select_mask);
+       u8 num_vls = hweight64(vl_select_mask);
        struct _vls_pctrs *vlinfo;
        struct hfi1_ibport *ibp = to_iport(ibdev, port);
        struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
@@ -2770,7 +2769,7 @@ static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
 
        hfi1_read_link_quality(dd, &rsp->link_quality_indicator);
 
-       rsp->vl_select_mask = cpu_to_be32(vl_select_mask);
+       rsp->vl_select_mask = cpu_to_be32((u32)vl_select_mask);
        rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
                                          CNTR_INVALID_VL));
        rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
@@ -2841,8 +2840,7 @@ static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
         * So in the for_each_set_bit() loop below, we don't need
         * any additional checks for vl.
         */
-       for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
-                        8 * sizeof(vl_select_mask)) {
+       for_each_set_bit(vl, &vl_select_mask, BITS_PER_LONG) {
                memset(vlinfo, 0, sizeof(*vlinfo));
 
                tmp = read_dev_cntr(dd, C_DC_RX_FLIT_VL, idx_from_vl(vl));
@@ -2883,7 +2881,7 @@ static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
                vfi++;
        }
 
-       a0_portstatus(ppd, rsp, vl_select_mask);
+       a0_portstatus(ppd, rsp);
 
        if (resp_len)
                *resp_len += response_data_size;
@@ -2930,16 +2928,14 @@ static u64 get_error_counter_summary(struct ib_device *ibdev, u8 port,
        return error_counter_summary;
 }
 
-static void a0_datacounters(struct hfi1_pportdata *ppd, struct _port_dctrs *rsp,
-                           u32 vl_select_mask)
+static void a0_datacounters(struct hfi1_pportdata *ppd, struct _port_dctrs *rsp)
 {
        if (!is_bx(ppd->dd)) {
                unsigned long vl;
                u64 sum_vl_xmit_wait = 0;
-               u32 vl_all_mask = VL_MASK_ALL;
+               unsigned long vl_all_mask = VL_MASK_ALL;
 
-               for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
-                                8 * sizeof(vl_all_mask)) {
+               for_each_set_bit(vl, &vl_all_mask, BITS_PER_LONG) {
                        u64 tmp = sum_vl_xmit_wait +
                                  read_port_cntr(ppd, C_TX_WAIT_VL,
                                                 idx_from_vl(vl));
@@ -2994,7 +2990,7 @@ static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
        u64 port_mask;
        u8 port_num;
        unsigned long vl;
-       u32 vl_select_mask;
+       unsigned long vl_select_mask;
        int vfi;
        u16 link_width;
        u16 link_speed;
@@ -3071,8 +3067,7 @@ static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
         * So in the for_each_set_bit() loop below, we don't need
         * any additional checks for vl.
         */
-       for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
-                        8 * sizeof(req->vl_select_mask)) {
+       for_each_set_bit(vl, &vl_select_mask, BITS_PER_LONG) {
                memset(vlinfo, 0, sizeof(*vlinfo));
 
                rsp->vls[vfi].port_vl_xmit_data =
@@ -3120,7 +3115,7 @@ static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
                vfi++;
        }
 
-       a0_datacounters(ppd, rsp, vl_select_mask);
+       a0_datacounters(ppd, rsp);
 
        if (resp_len)
                *resp_len += response_data_size;
@@ -3215,7 +3210,7 @@ static int pma_get_opa_porterrors(struct opa_pma_mad *pmp,
        struct _vls_ectrs *vlinfo;
        unsigned long vl;
        u64 port_mask, tmp;
-       u32 vl_select_mask;
+       unsigned long vl_select_mask;
        int vfi;
 
        req = (struct opa_port_error_counters64_msg *)pmp->data;
@@ -3273,8 +3268,7 @@ static int pma_get_opa_porterrors(struct opa_pma_mad *pmp,
        vlinfo = &rsp->vls[0];
        vfi = 0;
        vl_select_mask = be32_to_cpu(req->vl_select_mask);
-       for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
-                        8 * sizeof(req->vl_select_mask)) {
+       for_each_set_bit(vl, &vl_select_mask, BITS_PER_LONG) {
                memset(vlinfo, 0, sizeof(*vlinfo));
                rsp->vls[vfi].port_vl_xmit_discards =
                        cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
@@ -3485,7 +3479,7 @@ static int pma_set_opa_portstatus(struct opa_pma_mad *pmp,
        u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
        u64 portn = be64_to_cpu(req->port_select_mask[3]);
        u32 counter_select = be32_to_cpu(req->counter_select_mask);
-       u32 vl_select_mask = VL_MASK_ALL; /* clear all per-vl cnts */
+       unsigned long vl_select_mask = VL_MASK_ALL; /* clear all per-vl cnts */
        unsigned long vl;
 
        if ((nports != 1) || (portn != 1 << port)) {
@@ -3579,8 +3573,7 @@ static int pma_set_opa_portstatus(struct opa_pma_mad *pmp,
        if (counter_select & CS_UNCORRECTABLE_ERRORS)
                write_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL, 0);
 
-       for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
-                        8 * sizeof(vl_select_mask)) {
+       for_each_set_bit(vl, &vl_select_mask, BITS_PER_LONG) {
                if (counter_select & CS_PORT_XMIT_DATA)
                        write_port_cntr(ppd, C_TX_FLIT_VL, idx_from_vl(vl), 0);