Regression: QCA6390 fails with "mm/page_alloc: place pages to tail in __free_pages_core()"

[Carl Huang]

On 2020-11-13 19:08, Carl Huang wrote:
> On 2020-11-13 16:17, Pavel Procopiuc wrote:
>> Op 12.11.2020 om 11:48 schreef David Hildenbrand:
>>> Trying to understand the code, it looks like there are always two 
>>> rounds of reqests. The first one always fails ("requesting one big 
>>> chunk of DMA memory"), the second one (providing multiple chunks of 
>>> DMA memory) is supposed to work - and we do allocate memory.
>>> 
>>> 
>>> In the *working* cases we have
>>> 
>>> Respond mem req failed, result: 1, err: 0
>>> qmi failed to respond fw mem req:-22
>>> ...
>>> chip_id 0x0 chip_family 0xb board_id 0xff soc_id 0xffffffff
>>> 
>>> We don't fail in qmi_txn_wait() - second request w
>>> 
>>> 
>>> In the *non-working* cases we have
>>> 
>>> Respond mem req failed, result: 1, err: 0
>>> qmi failed to respond fw mem req:-22
>>> ...
>>> qmi failed memory request, err = -110
>>> qmi failed to respond fw mem req:-110
>>> 
>>> We fail in qmi_txn_wait(). We run into a timeout (ETIMEDOUT).
>>> 
>>> Can we bump up the timeout limit and see if things change? Maybe FW 
>>> needs more time with other addresses.
>> 
>> I tried increasing ATH11K_QMI_WLANFW_TIMEOUT_MS 20 times to 100000
>> (i.e. 100 seconds) and it didn't have any positive effect, the second
>> error (-110) just came 100 seconds later and not 5.
>> 
> Checked some logs. Looks when the error happens, the physical address 
> are
> very small. Its' between 20M - 30M.
> 
> So could you have a try to reserve the memory starting from 20M?
> Add "memmap=10M\$20M" to your grub.cfg or edit in kernel parameters. so 
> ath11k
> can't allocate from these address.
> 
> Or you can try to reserve even larger memory starting from 20M.
> 
To guarantee ath11k doesn't get physical address below 32M, reserve some 
more, for
example "memmap=12M\$20M".