Re: [bug report] dmaengine: uniphier-xdmac: Add UniPhier external DMA controller driver

[Kunihiko Hayashi <hayashi.kunihiko@socionext.com>]

Hi,

On 2021/07/27 13:05, Kunihiko Hayashi wrote:
> Hi Dan,
> Thank you for reporting.
> 
> On 2021/07/26 16:34, Dan Carpenter wrote:
>> Hello Kunihiko Hayashi,
>>
>> The patch 667b9251440b: "dmaengine: uniphier-xdmac: Add UniPhier
>> external DMA controller driver" from Feb 21, 2020, leads to the
>> following static checker warning:
>>
>>     drivers/dma/uniphier-xdmac.c:212 uniphier_xdmac_chan_stop()
>>     warn: sleeping in atomic context
>>
>> drivers/dma/uniphier-xdmac.c
>>      197 static int uniphier_xdmac_chan_stop(struct uniphier_xdmac_chan *xc)
>>      198 {
>>      199     u32 val;
>>      200
>>      201     /* disable interrupt */
>>      202     val = readl(xc->reg_ch_base + XDMAC_IEN);
>>      203     val &= ~(XDMAC_IEN_ENDIEN | XDMAC_IEN_ERRIEN);
>>      204     writel(val, xc->reg_ch_base + XDMAC_IEN);
>>      205
>>      206     /* stop XDMAC */
>>      207     val = readl(xc->reg_ch_base + XDMAC_TSS);
>>      208     val &= ~XDMAC_TSS_REQ;
>>      209     writel(0, xc->reg_ch_base + XDMAC_TSS);
>>      210
>>      211     /* wait until transfer is stopped */
>> --> 212     return readl_poll_timeout(xc->reg_ch_base + XDMAC_STAT, val,
>>      213                   !(val & XDMAC_STAT_TENF), 100, 1000);
>>                                                                      ^^^
>> This is doing a 100 us sleep but both callers hold a spin lock.
> 
> Yes, this waiting code is in spin lock context.
> I think that the function needs to be rewritten in tasklet.

I found this can be simply replaced with readl_poll_timeout_atomic().
I'll fix it.

Thank you,
---
Best Regards
Kunihiko Hayashi