Find several obvious bugs during code review in exception paths. Provide
this patchset to fix them. Not tested, just compiled.
Ziyang Xuan (2):
octeontx2-pf: Fix potential memory leak in otx2_init_tc()
octeontx2-pf: Fix a potential double free in otx2_sq_free_sqbs()
drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c | 1 +
drivers/net/ethernet/marvell/octeontx2/nic/otx2_tc.c | 7 ++++++-
2 files changed, 7 insertions(+), 1 deletion(-)
--
2.25.1
otx2_sq_free_sqbs() will be called twice when goto "err_free_nix_queues"
label in otx2_init_hw_resources(). The first calling is within
otx2_free_sq_res() at "err_free_nix_queues" label, and the second calling
is at later "err_free_sq_ptrs" label.
In otx2_sq_free_sqbs(), If sq->sqb_ptrs[i] is not 0, the memory page it
points to will be freed, and sq->sqb_ptrs[i] do not be assigned 0 after
memory page be freed. If otx2_sq_free_sqbs() is called twice, the memory
page pointed by sq->sqb_ptrs[i] will be freeed twice. To fix the bug,
assign 0 to sq->sqb_ptrs[i] after memory page be freed.
Fixes: caa2da34fd25 ("octeontx2-pf: Initialize and config queues")
Signed-off-by: Ziyang Xuan <[email protected]>
---
drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
index 9e10e7471b88..5a25fe51d102 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
@@ -1146,6 +1146,7 @@ void otx2_sq_free_sqbs(struct otx2_nic *pfvf)
DMA_FROM_DEVICE,
DMA_ATTR_SKIP_CPU_SYNC);
put_page(virt_to_page(phys_to_virt(pa)));
+ sq->sqb_ptrs[sqb] = 0;
}
sq->sqb_count = 0;
}
--
2.25.1
Hello,
On Fri, 2022-11-25 at 15:45 +0800, Ziyang Xuan wrote:
> otx2_sq_free_sqbs() will be called twice when goto "err_free_nix_queues"
> label in otx2_init_hw_resources(). The first calling is within
> otx2_free_sq_res() at "err_free_nix_queues" label, and the second calling
> is at later "err_free_sq_ptrs" label.
>
> In otx2_sq_free_sqbs(), If sq->sqb_ptrs[i] is not 0, the memory page it
> points to will be freed, and sq->sqb_ptrs[i] do not be assigned 0 after
> memory page be freed. If otx2_sq_free_sqbs() is called twice, the memory
> page pointed by sq->sqb_ptrs[i] will be freeed twice. To fix the bug,
> assign 0 to sq->sqb_ptrs[i] after memory page be freed.
>
> Fixes: caa2da34fd25 ("octeontx2-pf: Initialize and config queues")
> Signed-off-by: Ziyang Xuan <[email protected]>
> ---
> drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
> index 9e10e7471b88..5a25fe51d102 100644
> --- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
> +++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
> @@ -1146,6 +1146,7 @@ void otx2_sq_free_sqbs(struct otx2_nic *pfvf)
> DMA_FROM_DEVICE,
> DMA_ATTR_SKIP_CPU_SYNC);
> put_page(virt_to_page(phys_to_virt(pa)));
> + sq->sqb_ptrs[sqb] = 0;
The above looks not needed...
> }
> sq->sqb_count = 0;
... as this will prevent the next invocation of otx2_sq_free_sqbs()
from traversing and freeing any sq->sqb_ptrs[] element.
Cheers,
Paolo
> }
> Hello,
>
> On Fri, 2022-11-25 at 15:45 +0800, Ziyang Xuan wrote:
>> otx2_sq_free_sqbs() will be called twice when goto "err_free_nix_queues"
>> label in otx2_init_hw_resources(). The first calling is within
>> otx2_free_sq_res() at "err_free_nix_queues" label, and the second calling
>> is at later "err_free_sq_ptrs" label.
>>
>> In otx2_sq_free_sqbs(), If sq->sqb_ptrs[i] is not 0, the memory page it
>> points to will be freed, and sq->sqb_ptrs[i] do not be assigned 0 after
>> memory page be freed. If otx2_sq_free_sqbs() is called twice, the memory
>> page pointed by sq->sqb_ptrs[i] will be freeed twice. To fix the bug,
>> assign 0 to sq->sqb_ptrs[i] after memory page be freed.
>>
>> Fixes: caa2da34fd25 ("octeontx2-pf: Initialize and config queues")
>> Signed-off-by: Ziyang Xuan <[email protected]>
>> ---
>> drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
>> index 9e10e7471b88..5a25fe51d102 100644
>> --- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
>> +++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
>> @@ -1146,6 +1146,7 @@ void otx2_sq_free_sqbs(struct otx2_nic *pfvf)
>> DMA_FROM_DEVICE,
>> DMA_ATTR_SKIP_CPU_SYNC);
>> put_page(virt_to_page(phys_to_virt(pa)));
>> + sq->sqb_ptrs[sqb] = 0;
>
> The above looks not needed...
>> }
>> sq->sqb_count = 0;
>
> ... as this will prevent the next invocation of otx2_sq_free_sqbs()
> from traversing and freeing any sq->sqb_ptrs[] element.
Yes, you are right. I did pay much attention to sq->sqb_ptrs[],
and omitted the for loop condition.
Thank you!
>
> Cheers,
>
> Paolo
>> }
>
>
> .
>