On 4/23/25 1:15 PM, Baochen Qiang wrote: > > > On 4/23/2025 2:59 PM, Muhammad Usama Anjum wrote: >> Firmware requests 2 segments at first. The first segment is of 6799360 >> whose allocation fails due to dma remapping not available. The success >> is returned to firmware. Then firmware asks for 22 smaller segments >> instead of 2 big ones. Those get allocated successfully. At suspend/ >> hibernation time, these segments aren't freed as they will be reused >> by firmware after resuming. >> >> After resume the firmware asks for 2 segments again with first segment >> of 6799360 and vaddr is not NULL. We compare the type and size with > > suggest to rephrase as: > > After resume the firmware asks for 2 segments again with first segment > of 6799360. Since chunk->vaddr is not NULL, we compare the type and size with > >> previous type and size to know if it can be reused or not. >> Unfortunately, we detect that it cannot be reuses and this first smaller > > s/reuses/reused/ > >> segment is freed. Then we continue to allocate 6799360 size memory from >> dma which fails and we call ath11k_qmi_free_target_mem_chunk() which > > it is odd with 'from dma' ... > > I think just say 'allocate 6799360 size memory' is good enough. > >> frees the second smaller segment as well. Later success is returned to >> firmware which asks for 22 smaller segments again. But as we had freed 2 >> segments already, we'll allocate the first 2 new smaller segments again >> and reuse the remaining 20. Hence we aren't reusing the all 22 small >> segments, but only 20. >> >> This patch is correcting the skip logic when vaddr is set, but size/type >> don't match. In this case, we should use the same skip and success logic >> as used when dma_alloc_coherent fails without freeing the memory area. >> >> We had got reports that memory allocation in this function failed at >> resume [1] which made us debug why the reuse logic is wrong. Those > > The link is just v1 of this patch, it is not the report. If there is no public report, > just don't mention it. > >> failures weren't because of the bigger chunk allocation failure as they >> are skipped. Rather these failures were because of smaller chunk >> allocation failures. This issue is in the kernel side as because of >> memory pressure or fragmentation, the dma memory allocation fails. This >> patch fixes freeing and allocation of 2 smaller chunks. > > I know you are describing why you start to debug this issue. But I don't think it is > needed in the commit message. No matter kernel allocation fails or succeeds, the issue is > there, and the description above is sufficient to make the issue clear. > >> >> Tested-on: WCN6855 WLAN.HSP.1.1-03926.13-QCAHSPSWPL_V2_SILICONZ_CE-2.52297.6 > > blank line needed. > >> [1] https://lore.kernel.org/all/b30bc7f6-845d-4f9d-967e-c04a2b5f13f5@xxxxxxxxxxxxx >> >> Signed-off-by: Muhammad Usama Anjum <usama.anjum@xxxxxxxxxxxxx> >> --- >> Changes since v1: >> - Update description >> >> Fixes: 5962f370ce41 ("ath11k: Reuse the available memory after firmware reload") >> I think we should keep fixes tag as ^ claimed that its adding reuse >> support. But it left a bug in reuse which we are fixing. >> >> Feel free to add it or leave it as it is. > > Jeff, what do you think? I'll update the description as mentioned. Let's wait for Jeff to review before I send v3. > >> --- >> drivers/net/wireless/ath/ath11k/qmi.c | 10 +++++++++- >> 1 file changed, 9 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/net/wireless/ath/ath11k/qmi.c b/drivers/net/wireless/ath/ath11k/qmi.c >> index 47b9d4126d3a9..3c26f4dcf5d29 100644 >> --- a/drivers/net/wireless/ath/ath11k/qmi.c >> +++ b/drivers/net/wireless/ath/ath11k/qmi.c >> @@ -1990,8 +1990,16 @@ static int ath11k_qmi_alloc_target_mem_chunk(struct ath11k_base *ab) >> */ >> if (chunk->vaddr) { >> if (chunk->prev_type == chunk->type && >> - chunk->prev_size == chunk->size) >> + chunk->prev_size == chunk->size) { >> continue; >> + } else if (ab->qmi.mem_seg_count <= ATH11K_QMI_FW_MEM_REQ_SEGMENT_CNT) { >> + ath11k_dbg(ab, ATH11K_DBG_QMI, >> + "size/type mismatch (current %d %u) (prev %d %u), try later with small size\n", >> + chunk->size, chunk->type, >> + chunk->prev_size, chunk->prev_type); >> + ab->qmi.target_mem_delayed = true; >> + return 0; >> + } >> >> /* cannot reuse the existing chunk */ >> dma_free_coherent(ab->dev, chunk->prev_size, > -- Regards, Usama
