NVMe driver init sequence

Fri Aug 25 16:11:39 UTC 2017

I have observed this behavior in our lab on 10.3-STABLE. So I have reported against it.
If you can bring in the change for 10.4R that will be great (lot of users stick to 10 based releases for their development). And in case, if customer faces this issue, I can point to this commit ☺.

Regards,
Kinjal Patel

From: wlosh at bsdimp.com [mailto:wlosh at bsdimp.com] On Behalf Of Warner Losh
Sent: Friday, August 25, 2017 9:00 AM
To: Kinjal Patel
Cc: Jim Harris; freebsd-drivers at freebsd.org
Subject: Re: NVMe driver init sequence

BTW, I just noticed you reported this against 10.3R. Is there some reason (perhaps a customer with an issue) that you reported it there? I'm trying to decide if I need to MFC this change to 11.x and 10.x (the latter has some time urgency due to upcoming 10.4R release).

Warner

On Fri, Aug 25, 2017 at 9:54 AM, Kinjal Patel <Kinjal.Patel at taec.toshiba.com<mailto:Kinjal.Patel at taec.toshiba.com>> wrote:
Great!
Thanks Warner.

Regards,
Kinjal Patel

From: wlosh at bsdimp.com<mailto:wlosh at bsdimp.com> [mailto:wlosh at bsdimp.com<mailto:wlosh at bsdimp.com>] On Behalf Of Warner Losh
Sent: Thursday, August 24, 2017 8:17 PM
To: Jim Harris
Cc: Kinjal Patel; freebsd-drivers at freebsd.org<mailto:freebsd-drivers at freebsd.org>

Subject: Re: NVMe driver init sequence

Greetings,

I've just committed r322872 which implements this, with a slightly tweaked comment.

Warner

On Fri, Aug 18, 2017 at 4:09 PM, Jim Harris <jim.harris at gmail.com<mailto:jim.harris at gmail.com>> wrote:
On Fri, Aug 18, 2017 at 2:04 PM, Kinjal Patel
<Kinjal.Patel at taec.toshiba.com<mailto:Kinjal.Patel at taec.toshiba.com>> wrote:
> Created a bug #22166 (https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=221616)
> Updated the description and patch to it.
>
> Regards,
> Kinjal Patel
>
> From: Kinjal Patel
> Sent: Friday, August 18, 2017 1:51 PM
> To: 'Warner Losh'
> Cc: freebsd-drivers at freebsd.org<mailto:freebsd-drivers at freebsd.org>
> Subject: RE: NVMe driver init sequence
>
> Here is a the change I propose. This will make the init sequence right.

Looks good to me.

>
> Index: sys/dev/nvme/nvme.c
> ===================================================================
> --- sys/dev/nvme/nvme.c (revision 322672)
> +++ sys/dev/nvme/nvme.c (working copy)
> @@ -253,6 +253,9 @@
>                 return (status);
>         }
>
> +    /* make device bus-master before attempting to enable controller */
> +    pci_enable_busmaster(dev);
> +
>         /*
>          * Reset controller twice to ensure we do a transition from cc.en==1
>          *  to cc.en==0.  This is because we don't really know what status
> @@ -270,8 +273,6 @@
>                 return (status);
>         }
>
> -       pci_enable_busmaster(dev);
> -
>         ctrlr->config_hook.ich_func = nvme_ctrlr_start_config_hook;
>         ctrlr->config_hook.ich_arg = ctrlr;
>
>
>
> regards,
> Kinjal Patel
>
> From: wlosh at bsdimp.com<mailto:wlosh at bsdimp.com><mailto:wlosh at bsdimp.com<mailto:wlosh at bsdimp.com>> [mailto:wlosh at bsdimp.com<mailto:wlosh at bsdimp.com>] On Behalf Of Warner Losh
> Sent: Thursday, August 17, 2017 3:10 PM
> To: Kinjal Patel
> Cc: freebsd-drivers at freebsd.org<mailto:freebsd-drivers at freebsd.org><mailto:freebsd-drivers at freebsd.org<mailto:freebsd-drivers at freebsd.org>>
> Subject: Re: NVMe driver init sequence
>
> This sounds right to me....  Most BIOSes seem to enable BME=1.
>
> Warner
>
> On Thu, Aug 17, 2017 at 2:45 PM, Kinjal Patel <Kinjal.Patel at taec.toshiba.com<mailto:Kinjal.Patel at taec.toshiba.com><mailto:Kinjal.Patel at taec.toshiba.com<mailto:Kinjal.Patel at taec.toshiba.com>>> wrote:
> Hi,
>
> I have a query on NVMe driver init sequence in FreeBSD 10.3(stable).
>
> The driver init sequence is,
>
> 1)      Enable controller (CC.EN=1)
>
> 2)      Wait for controller ready (CSTS.RDY=1)
>
> 3)      Set PCI bus master enable (BME=1)
>
> As per NVMe spec, when NVMe controller becomes ready it has to process command.
>
> "Enable (EN): When set to '1', then the controller shall process commands based on Submission Queue Tail doorbell writes"
>
> And per PCI Express spec when BME is not set, the PCI Express Function is not allowed to issue any Memory or I/O requests.
> "Bus Master Enable - Controls the ability of a PCI Express Endpoint to issue Memory95 and I/O Read/Write Requests, and
> the ability of a Root or Switch Port to forward Memory and I/O Read/Write Requests in the Upstream direction"
>
> Enabling controller before setting BME=1 is violation of spec, as controller has to accept commands but BME is prerequisite for that.
>
> The Linux device driver init sequence is,
>
> 1)     Set PCI bus master enable (BME=1)
>
> 2)     Enable Controller (CC.EN=1)
>
> 3)     Wait for controller ready (CSTS.RDY=1)
>
> The FreeBSD NVMe driver sequence should be changed to set BME=1 before attempting to enable controller.
>
>
> Regards,
> Kinjal Patel
> _______________________________________________
> freebsd-drivers at freebsd.org<mailto:freebsd-drivers at freebsd.org><mailto:freebsd-drivers at freebsd.org<mailto:freebsd-drivers at freebsd.org>> mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-drivers
> To unsubscribe, send any mail to "freebsd-drivers-unsubscribe at freebsd.org<mailto:freebsd-drivers-unsubscribe at freebsd.org><mailto:freebsd-drivers-unsubscribe at freebsd.org<mailto:freebsd-drivers-unsubscribe at freebsd.org>>"
>
> _______________________________________________
> freebsd-drivers at freebsd.org<mailto:freebsd-drivers at freebsd.org> mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-drivers
> To unsubscribe, send any mail to "freebsd-drivers-unsubscribe at freebsd.org<mailto:freebsd-drivers-unsubscribe at freebsd.org>"