Hi Js, I read your attached log and patch, yes, dma memory leak will happen when more than one pre_request execute. The method of ++next->cookie is not good, your patch seems good, but I still need some time to test the patch, because you unmap the dma in sdhci_finish_data rather than the sdhci_post_req. Anyway, thanks for report and debug this issue. I will give you my test result ASAP. Best Regards Haibo Chen > -----Original Message----- > From: Jiri Slaby [mailto:jslaby@xxxxxxx] > Sent: Wednesday, August 05, 2015 11:12 PM > To: Chen Haibo-B51421; Ulf Hansson > Cc: linux-mmc@xxxxxxxxxxxxxxx; Linux kernel mailing list > Subject: [RFC] sdhci: fix DMA leaks [was: [SHDCI] Heavy (thousands) DMA > leaks] > > On 08/05/2015, 01:52 PM, Jiri Slaby wrote: > >> Yes, I see: > >> sdhci-pci 0000:02:00.0: swiotlb buffer is full (sz: 65536 bytes) > >> after some time. The driver falls back to non-DMA transfers after that. > >> It also generates a warning about that: > >> WARNING: CPU: 0 PID: 0 at drivers/mmc/host/sdhci.c:857 > >> sdhci_prepare_data+0x8ec/0x900 [sdhci]() > > > > I am attaching a debug patch and a debug log. You can see where > > 0x00000000fffb0000 and 0x00000000fffe0000 is leaked. It is when > > 'invalid cookie' error happens. > > And you could see the cookie handling is totally bogus. > > With this rewrite, I no longer see the problems. Could you confirm it > still does the good job with respect to performance -- the numbers you > mentioned in your commit. > > Ulf, what do you think about the attached patch? (Do not look at the > commented info prints.) > > thanks, > -- > js > suse labs ��.n��������+%������w��{.n�����{��i��)�� jg��� �����ݢj����G���� ���j:+v���w�m������w�������h�����٥
