Skip to contentSkip to conversation

perf/x86: Add Intel Tiger Lake uncore support

bug report Dan Carpenter 2 participants 2 messages Jan 25, 2022

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hello Kan Liang,

The patch fdb64822443e: "perf/x86: Add Intel Tiger Lake uncore support" from Feb 6, 2020, leads to the following Smatch static checker warning:

	arch/x86/events/intel/uncore_snb.c:1486 __uncore_imc_init_box()
	warn: cast after binop

arch/x86/events/intel/uncore_snb.c

    1465 static void __uncore_imc_init_box(struct intel_uncore_box *box,
    1466                                   unsigned int base_offset)
    1467 {
    1468         struct pci_dev *pdev = tgl_uncore_get_mc_dev();
    1469         struct intel_uncore_pmu *pmu = box->pmu;
    1470         struct intel_uncore_type *type = pmu->type;
    1471         resource_size_t addr;
    1472         u32 mch_bar;
    1473 
    1474         if (!pdev) {
    1475                 pr_warn("perf uncore: Cannot find matched IMC device.\n");
    1476                 return;
    1477         }
    1478 
    1479         pci_read_config_dword(pdev, SNB_UNCORE_PCI_IMC_BAR_OFFSET, &mch_bar);
    1480         /* MCHBAR is disabled */
    1481         if (!(mch_bar & BIT(0))) {
    1482                 pr_warn("perf uncore: MCHBAR is disabled. Failed to map IMC free-running counters.\n");
    1483                 return;
    1484         }
    1485         mch_bar &= ~BIT(0);
--> 1486         addr = (resource_size_t)(mch_bar + TGL_UNCORE_MMIO_IMC_MEM_OFFSET * pmu->pmu_idx);
                                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

All of these are 32 bit values. If we really need a 64bit type, these will overflow before the cast to resource_size_t. This code is 2 years old, so I suspect the cast is only required on line 1490 and that one is correct.

    1487 
    1488 #ifdef CONFIG_PHYS_ADDR_T_64BIT
    1489         pci_read_config_dword(pdev, SNB_UNCORE_PCI_IMC_BAR_OFFSET + 4, &mch_bar);
    1490         addr |= ((resource_size_t)mch_bar << 32);
    1491 #endif
    1492 
    1493         addr += base_offset;
    1494         box->io_addr = ioremap(addr, type->mmio_map_size);
    1495         if (!box->io_addr)
    1496                 pr_warn("perf uncore: Failed to ioremap for %s.\n", type->name);
    1497 }

regards,
dan carpenter

#1 lore raw

Hi Dan

Quoted 33 lines · Dan Carpenter

Hello Kan Liang,

The patch fdb64822443e: "perf/x86: Add Intel Tiger Lake uncore support" from Feb 6, 2020, leads to the following Smatch static checker warning:

	arch/x86/events/intel/uncore_snb.c:1486 __uncore_imc_init_box()
	warn: cast after binop

arch/x86/events/intel/uncore_snb.c

     1465 static void __uncore_imc_init_box(struct intel_uncore_box *box,
     1466                                   unsigned int base_offset)
     1467 {
     1468         struct pci_dev *pdev = tgl_uncore_get_mc_dev();
     1469         struct intel_uncore_pmu *pmu = box->pmu;
     1470         struct intel_uncore_type *type = pmu->type;
     1471         resource_size_t addr;
     1472         u32 mch_bar;
     1473
     1474         if (!pdev) {
     1475                 pr_warn("perf uncore: Cannot find matched IMC device.\n");
     1476                 return;
     1477         }
     1478
     1479         pci_read_config_dword(pdev, SNB_UNCORE_PCI_IMC_BAR_OFFSET, &mch_bar);
     1480         /* MCHBAR is disabled */
     1481         if (!(mch_bar & BIT(0))) {
     1482                 pr_warn("perf uncore: MCHBAR is disabled. Failed to map IMC free-running counters.\n");
     1483                 return;
     1484         }
     1485         mch_bar &= ~BIT(0);
--> 1486         addr = (resource_size_t)(mch_bar + TGL_UNCORE_MMIO_IMC_MEM_OFFSET * pmu->pmu_idx);
                                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Quoted 5 lines

All of these are 32 bit values. If we really need a 64bit type, these will overflow before the cast to resource_size_t. This code is 2 years old, so I suspect the cast is only required on line 1490 and that one is correct.

Thanks for the report.

For 32bit system, I'm not sure whether it's OK to remove the resource_size_t for 1486? Shouldn't we explicitly cast to resource_size_t for the possible overflow part for a 32bit system as below?

arch/x86/events/intel/uncore_snb.c +1 -1
index 36b6c20..fc6cfcb 100644
@@ -1720,7 +1720,7 @@ static void __uncore_imc_init_box(struct intel_uncore_box *box,
                return;
        }
        mch_bar &= ~BIT(0);
-       addr = (resource_size_t)(mch_bar + TGL_UNCORE_MMIO_IMC_MEM_OFFSET * pmu->pmu_idx);
+       addr = mch_bar + (resource_size_t)(TGL_UNCORE_MMIO_IMC_MEM_OFFSET * pmu->pmu_idx);
 
 #ifdef CONFIG_PHYS_ADDR_T_64BIT
        pci_read_config_dword(pdev, SNB_UNCORE_PCI_IMC_BAR_OFFSET + 4, 

&mch_bar);

Thanks,
Kan

Quoted 14 lines
     1487
     1488 #ifdef CONFIG_PHYS_ADDR_T_64BIT
     1489         pci_read_config_dword(pdev, SNB_UNCORE_PCI_IMC_BAR_OFFSET + 4, &mch_bar);
     1490         addr |= ((resource_size_t)mch_bar << 32);
     1491 #endif
     1492
     1493         addr += base_offset;
     1494         box->io_addr = ioremap(addr, type->mmio_map_size);
     1495         if (!box->io_addr)
     1496                 pr_warn("perf uncore: Failed to ioremap for %s.\n", type->name);
     1497 }

regards,
dan carpenter