https://bugs.freedesktop.org/show_bug.cgi?id=108609

Bug ID: 108609 Summary: vegam_smumgr.c: accessing mvdd_voltage_table.entries[] array out of bounds in function vegam_populate_smc_mvdd_table Product: DRI Version: unspecified Hardware: x86-64 (AMD64) OS: Linux (All) Status: NEW Severity: normal Priority: medium Component: DRM/AMDgpu Assignee: dri-devel@lists.freedesktop.org Reporter: rstrube@gmail.com

Created attachment 142298 --> https://bugs.freedesktop.org/attachment.cgi?id=142298&action=edit Patch to fix accessing mvdd_voltage_table.entries[] array out of bounds in vegam_smumgr.c

I believe I've discovered a small bug in the vegam_smumgr.c, specifically the following function:

static int vegam_populate_smc_mvdd_table(struct pp_hwmgr *hwmgr, SMU75_Discrete_DpmTable *table) { struct smu7_hwmgr *data = (struct smu7_hwmgr *)(hwmgr->backend); uint32_t count, level;

if (SMU7_VOLTAGE_CONTROL_BY_GPIO == data->mvdd_control) { count = data->mvdd_voltage_table.count; if (count > SMU_MAX_SMIO_LEVELS) count = SMU_MAX_SMIO_LEVELS; for (level = 0; level < count; level++) { table->SmioTable2.Pattern[level].Voltage = PP_HOST_TO_SMC_US(

data->mvdd_voltage_table.entries[count].value * VOLTAGE_SCALE); /* Index into DpmTable.Smio. Drive bits from Smio entry to get this voltage level.*/ table->SmioTable2.Pattern[level].Smio = (uint8_t) level; table->Smio[level] |=

data->mvdd_voltage_table.entries[level].smio_low; } table->SmioMask2 = data->mvdd_voltage_table.mask_low;

table->MvddLevelCount = (uint32_t) PP_HOST_TO_SMC_UL(count); }

return 0; }

With the lines (within the for loop):

table->SmioTable2.Pattern[level].Voltage = PP_HOST_TO_SMC_US( data->mvdd_voltage_table.entries[count].value * VOLTAGE_SCALE);

If this code was executed it would try to access the mvdd_voltage_table.entries[] array out of bounds, because count > than the max value for level.

I believe:

data->mvdd_voltage_table.entries[count].value

should actually be:

data->mvdd_voltage_table.entries[level].value

You can see in a similar function within vegam_smumgr.c, this bug is *not* present:

static int vegam_populate_smc_vddci_table(struct pp_hwmgr *hwmgr, struct SMU75_Discrete_DpmTable *table) { uint32_t count, level; struct smu7_hwmgr *data = (struct smu7_hwmgr *)(hwmgr->backend);

count = data->vddci_voltage_table.count;

if (SMU7_VOLTAGE_CONTROL_BY_GPIO == data->vddci_control) { if (count > SMU_MAX_SMIO_LEVELS) count = SMU_MAX_SMIO_LEVELS; for (level = 0; level < count; ++level) { table->SmioTable1.Pattern[level].Voltage = PP_HOST_TO_SMC_US(

data->vddci_voltage_table.entries[level].value * VOLTAGE_SCALE); table->SmioTable1.Pattern[level].Smio = (uint8_t) level;

table->Smio[level] |= data->vddci_voltage_table.entries[level].smio_low; } }

table->SmioMask1 = data->vddci_voltage_table.mask_low;

return 0; }

I've attached a patch for kernel 4.19, admittedly the change is trivial but I figured I would try to do things the right way :)

Thanks! Rob