| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <bed8564e-0801-4c55-a9ce-32956db68b98@oss.qualcomm.com>
Date: Tue, 18 Nov 2025 19:02:24 +0800
From: Baochen Qiang <baochen.qiang@....qualcomm.com>
To: Alexander Wilhelm <alexander.wilhelm@...termo.com>
Cc: Jeff Johnson <jjohnson@...nel.org>, linux-wireless@...r.kernel.org,
ath12k@...ts.infradead.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2] wifi: ath12k: fix endianness handling for SRNG ring
pointer accesses
On 11/18/2025 6:49 PM, Alexander Wilhelm wrote:
> On Tue, Nov 18, 2025 at 06:43:44PM +0800, Baochen Qiang wrote:
>>
>>
>> On 11/18/2025 6:17 PM, Alexander Wilhelm wrote:
>>> The SRNG head and tail ring pointers are stored in device memory as
>>> little-endian values. On big-endian systems, direct dereferencing of these
>>> pointers leads to incorrect values being read or written, causing ring
>>> management issues and potentially breaking data flow.
>>>
>>> This patch ensures all accesses to SRNG ring pointers use the appropriate
>>> endianness conversions. This affects both read and write paths for source
>>> and destination rings, as well as debug output. The changes guarantee
>>> correct operation on both little- and big-endian architectures.
>>>
>>> Signed-off-by: Alexander Wilhelm <alexander.wilhelm@...termo.com>
>>> ---
>>> Changes in v2:
>>> - Set '__le32 *' type for 'hp_addr/tp_addr' in both 'dst_ring' and 'src_ring'
>>> ---
>>> drivers/net/wireless/ath/ath12k/hal.c | 35 +++++++++++++++------------
>>> drivers/net/wireless/ath/ath12k/hal.h | 8 +++---
>>> 2 files changed, 24 insertions(+), 19 deletions(-)
>>>
>>> diff --git a/drivers/net/wireless/ath/ath12k/hal.c b/drivers/net/wireless/ath/ath12k/hal.c
>>> index 6406fcf5d69f..bd4d1de9eb1a 100644
>>> --- a/drivers/net/wireless/ath/ath12k/hal.c
>>> +++ b/drivers/net/wireless/ath/ath12k/hal.c
>>> @@ -2007,7 +2007,7 @@ int ath12k_hal_srng_dst_num_free(struct ath12k_base *ab, struct hal_srng *srng,
>>> tp = srng->u.dst_ring.tp;
>>>
>>> if (sync_hw_ptr) {
>>> - hp = *srng->u.dst_ring.hp_addr;
>>> + hp = le32_to_cpu(*srng->u.dst_ring.hp_addr);
>>> srng->u.dst_ring.cached_hp = hp;
>>> } else {
>>> hp = srng->u.dst_ring.cached_hp;
>>> @@ -2030,7 +2030,7 @@ int ath12k_hal_srng_src_num_free(struct ath12k_base *ab, struct hal_srng *srng,
>>> hp = srng->u.src_ring.hp;
>>>
>>> if (sync_hw_ptr) {
>>> - tp = *srng->u.src_ring.tp_addr;
>>> + tp = le32_to_cpu(*srng->u.src_ring.tp_addr);
>>> srng->u.src_ring.cached_tp = tp;
>>> } else {
>>> tp = srng->u.src_ring.cached_tp;
>>> @@ -2149,9 +2149,9 @@ void ath12k_hal_srng_access_begin(struct ath12k_base *ab, struct hal_srng *srng)
>>>
>>> if (srng->ring_dir == HAL_SRNG_DIR_SRC) {
>>> srng->u.src_ring.cached_tp =
>>> - *(volatile u32 *)srng->u.src_ring.tp_addr;
>>> + le32_to_cpu(*(volatile u32 *)srng->u.src_ring.tp_addr);
>>
>> s/volatile u32 */volatile __le32 */ ?
>
> I got it. I'll fix it in v3.
>
>>> } else {
>>> - hp = READ_ONCE(*srng->u.dst_ring.hp_addr);
>>> + hp = le32_to_cpu(READ_ONCE(*srng->u.dst_ring.hp_addr));
>>>
>>> if (hp != srng->u.dst_ring.cached_hp) {
>>> srng->u.dst_ring.cached_hp = hp;
>>> @@ -2175,25 +2175,28 @@ void ath12k_hal_srng_access_end(struct ath12k_base *ab, struct hal_srng *srng)
>>> * hence written to a shared memory location that is read by FW
>>> */
>>> if (srng->ring_dir == HAL_SRNG_DIR_SRC) {
>>> - srng->u.src_ring.last_tp =
>>> - *(volatile u32 *)srng->u.src_ring.tp_addr;
>>> + srng->u.src_ring.last_tp = le32_to_cpu(
>>> + *(volatile u32 *)srng->u.src_ring.tp_addr);
>>
>> s/volatile u32 */volatile __le32 */ ?
>
> Same as above, sure!
>
>>> /* Make sure descriptor is written before updating the
>>> * head pointer.
>>> */
>>> dma_wmb();
>>> - WRITE_ONCE(*srng->u.src_ring.hp_addr, srng->u.src_ring.hp);
>>> + WRITE_ONCE(*srng->u.src_ring.hp_addr,
>>> + cpu_to_le32(srng->u.src_ring.hp));
>>> } else {
>>> - srng->u.dst_ring.last_hp = *srng->u.dst_ring.hp_addr;
>>> + srng->u.dst_ring.last_hp =
>>> + le32_to_cpu(*srng->u.dst_ring.hp_addr);
>>> /* Make sure descriptor is read before updating the
>>> * tail pointer.
>>> */
>>> dma_mb();
>>> - WRITE_ONCE(*srng->u.dst_ring.tp_addr, srng->u.dst_ring.tp);
>>> + WRITE_ONCE(*srng->u.dst_ring.tp_addr,
>>> + cpu_to_le32(srng->u.dst_ring.tp));
>>> }
>>> } else {
>>> if (srng->ring_dir == HAL_SRNG_DIR_SRC) {
>>> - srng->u.src_ring.last_tp =
>>> - *(volatile u32 *)srng->u.src_ring.tp_addr;
>>> + srng->u.src_ring.last_tp = le32_to_cpu(
>>> + *(volatile u32 *)srng->u.src_ring.tp_addr);
>
> Same as above, sure!
>
>>> /* Assume implementation use an MMIO write accessor
>>> * which has the required wmb() so that the descriptor
>>> * is written before the updating the head pointer.
>>> @@ -2203,7 +2206,8 @@ void ath12k_hal_srng_access_end(struct ath12k_base *ab, struct hal_srng *srng)
>>> (unsigned long)ab->mem,
>>> srng->u.src_ring.hp);
>>> } else {
>>> - srng->u.dst_ring.last_hp = *srng->u.dst_ring.hp_addr;
>>> + srng->u.dst_ring.last_hp =
>>> + le32_to_cpu(*srng->u.dst_ring.hp_addr);
>>> /* Make sure descriptor is read before updating the
>>> * tail pointer.
>>> */
>>> @@ -2547,7 +2551,7 @@ void ath12k_hal_srng_shadow_update_hp_tp(struct ath12k_base *ab,
>>> * HP only when then ring isn't' empty.
>>> */
>>> if (srng->ring_dir == HAL_SRNG_DIR_SRC &&
>>> - *srng->u.src_ring.tp_addr != srng->u.src_ring.hp)
>>> + le32_to_cpu(*srng->u.src_ring.tp_addr) != srng->u.src_ring.hp)
>>> ath12k_hal_srng_access_end(ab, srng);
>>> }
>>>
>>> @@ -2648,14 +2652,15 @@ void ath12k_hal_dump_srng_stats(struct ath12k_base *ab)
>>> "src srng id %u hp %u, reap_hp %u, cur tp %u, cached tp %u last tp %u napi processed before %ums\n",
>>> srng->ring_id, srng->u.src_ring.hp,
>>> srng->u.src_ring.reap_hp,
>>> - *srng->u.src_ring.tp_addr, srng->u.src_ring.cached_tp,
>>> + __le32_to_cpu(*srng->u.src_ring.tp_addr),
>>> + srng->u.src_ring.cached_tp,
>>> srng->u.src_ring.last_tp,
>>> jiffies_to_msecs(jiffies - srng->timestamp));
>>> else if (srng->ring_dir == HAL_SRNG_DIR_DST)
>>> ath12k_err(ab,
>>> "dst srng id %u tp %u, cur hp %u, cached hp %u last hp %u napi processed before %ums\n",
>>> srng->ring_id, srng->u.dst_ring.tp,
>>> - *srng->u.dst_ring.hp_addr,
>>> + __le32_to_cpu(*srng->u.dst_ring.hp_addr),
>>
>> still my v1 comment does not get addressed:
>>
>>
>> why __le32_to_cpu() only in logging, while le32_to_cpu() elsewhere?
>
> Sorry, I was confused with the previous issue. Yes, I saw this form on an
> upstream patch where '__le32_to_cpu()' was used instead of 'le32_to_cpu()'.
> Which one should I prefer? I'll unify that in v3.
I am not sure here -- Copilot tells me that le32_to_cpu() does argument type check while
__le32_to_cpu() does not, in that case we should use the latter one because we are sure
the type is safe.
However actual code shows they are actually the same:
include/linux/byteorder/generic.h:
#define le32_to_cpu __le32_to_cpu
Let's see if others can give some insights.
>
>
> Best regards
> Alexander Wilhelm
Powered by blists - more mailing lists