[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH] xen-blkfront: use old rinfo after enomem during migration


  • To: Manjunath Patil <manjunath.b.patil@xxxxxxxxxx>, jgross@xxxxxxxx, konrad.wilk@xxxxxxxxxx, roger.pau@xxxxxxxxxx, axboe@xxxxxxxxx
  • From: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
  • Date: Fri, 30 Nov 2018 17:33:46 -0500
  • Autocrypt: addr=boris.ostrovsky@xxxxxxxxxx; prefer-encrypt=mutual; keydata= xsFNBFH8CgsBEAC0KiOi9siOvlXatK2xX99e/J3OvApoYWjieVQ9232Eb7GzCWrItCzP8FUV PQg8rMsSd0OzIvvjbEAvaWLlbs8wa3MtVLysHY/DfqRK9Zvr/RgrsYC6ukOB7igy2PGqZd+M MDnSmVzik0sPvB6xPV7QyFsykEgpnHbvdZAUy/vyys8xgT0PVYR5hyvhyf6VIfGuvqIsvJw5 C8+P71CHI+U/IhsKrLrsiYHpAhQkw+Zvyeml6XSi5w4LXDbF+3oholKYCkPwxmGdK8MUIdkM d7iYdKqiP4W6FKQou/lC3jvOceGupEoDV9botSWEIIlKdtm6C4GfL45RD8V4B9iy24JHPlom woVWc0xBZboQguhauQqrBFooHO3roEeM1pxXjLUbDtH4t3SAI3gt4dpSyT3EvzhyNQVVIxj2 FXnIChrYxR6S0ijSqUKO0cAduenhBrpYbz9qFcB/GyxD+ZWY7OgQKHUZMWapx5bHGQ8bUZz2 SfjZwK+GETGhfkvNMf6zXbZkDq4kKB/ywaKvVPodS1Poa44+B9sxbUp1jMfFtlOJ3AYB0WDS Op3d7F2ry20CIf1Ifh0nIxkQPkTX7aX5rI92oZeu5u038dHUu/dO2EcuCjl1eDMGm5PLHDSP 0QUw5xzk1Y8MG1JQ56PtqReO33inBXG63yTIikJmUXFTw6lLJwARAQABzTNCb3JpcyBPc3Ry b3Zza3kgKFdvcmspIDxib3Jpcy5vc3Ryb3Zza3lAb3JhY2xlLmNvbT7CwXgEEwECACIFAlH8 CgsCGwMGCwkIBwMCBhUIAgkKCwQWAgMBAh4BAheAAAoJEIredpCGysGyasEP/j5xApopUf4g 9Fl3UxZuBx+oduuw3JHqgbGZ2siA3EA4bKwtKq8eT7ekpApn4c0HA8TWTDtgZtLSV5IdH+9z JimBDrhLkDI3Zsx2CafL4pMJvpUavhc5mEU8myp4dWCuIylHiWG65agvUeFZYK4P33fGqoaS VGx3tsQIAr7MsQxilMfRiTEoYH0WWthhE0YVQzV6kx4wj4yLGYPPBtFqnrapKKC8yFTpgjaK jImqWhU9CSUAXdNEs/oKVR1XlkDpMCFDl88vKAuJwugnixjbPFTVPyoC7+4Bm/FnL3iwlJVE qIGQRspt09r+datFzPqSbp5Fo/9m4JSvgtPp2X2+gIGgLPWp2ft1NXHHVWP19sPgEsEJXSr9 tskM8ScxEkqAUuDs6+x/ISX8wa5Pvmo65drN+JWA8EqKOHQG6LUsUdJolFM2i4Z0k40BnFU/ kjTARjrXW94LwokVy4x+ZYgImrnKWeKac6fMfMwH2aKpCQLlVxdO4qvJkv92SzZz4538az1T m+3ekJAimou89cXwXHCFb5WqJcyjDfdQF857vTn1z4qu7udYCuuV/4xDEhslUq1+GcNDjAhB nNYPzD+SvhWEsrjuXv+fDONdJtmLUpKs4Jtak3smGGhZsqpcNv8nQzUGDQZjuCSmDqW8vn2o hWwveNeRTkxh+2x1Qb3GT46uzsFNBFH8CgsBEADGC/yx5ctcLQlB9hbq7KNqCDyZNoYu1HAB Hal3MuxPfoGKObEktawQPQaSTB5vNlDxKihezLnlT/PKjcXC2R1OjSDinlu5XNGc6mnky03q yymUPyiMtWhBBftezTRxWRslPaFWlg/h/Y1iDuOcklhpr7K1h1jRPCrf1yIoxbIpDbffnuyz kuto4AahRvBU4Js4sU7f/btU+h+e0AcLVzIhTVPIz7PM+Gk2LNzZ3/on4dnEc/qd+ZZFlOQ4 KDN/hPqlwA/YJsKzAPX51L6Vv344pqTm6Z0f9M7YALB/11FO2nBB7zw7HAUYqJeHutCwxm7i BDNt0g9fhviNcJzagqJ1R7aPjtjBoYvKkbwNu5sWDpQ4idnsnck4YT6ctzN4I+6lfkU8zMzC gM2R4qqUXmxFIS4Bee+gnJi0Pc3KcBYBZsDK44FtM//5Cp9DrxRQOh19kNHBlxkmEb8kL/pw XIDcEq8MXzPBbxwHKJ3QRWRe5jPNpf8HCjnZz0XyJV0/4M1JvOua7IZftOttQ6KnM4m6WNIZ 2ydg7dBhDa6iv1oKdL7wdp/rCulVWn8R7+3cRK95SnWiJ0qKDlMbIN8oGMhHdin8cSRYdmHK kTnvSGJNlkis5a+048o0C6jI3LozQYD/W9wq7MvgChgVQw1iEOB4u/3FXDEGulRVko6xCBU4 SQARAQABwsFfBBgBAgAJBQJR/AoLAhsMAAoJEIredpCGysGyfvMQAIywR6jTqix6/fL0Ip8G jpt3uk//QNxGJE3ZkUNLX6N786vnEJvc1beCu6EwqD1ezG9fJKMl7F3SEgpYaiKEcHfoKGdh 30B3Hsq44vOoxR6zxw2B/giADjhmWTP5tWQ9548N4VhIZMYQMQCkdqaueSL+8asp8tBNP+TJ PAIIANYvJaD8xA7sYUXGTzOXDh2THWSvmEWWmzok8er/u6ZKdS1YmZkUy8cfzrll/9hiGCTj u3qcaOM6i/m4hqtvsI1cOORMVwjJF4+IkC5ZBoeRs/xW5zIBdSUoC8L+OCyj5JETWTt40+lu qoqAF/AEGsNZTrwHJYu9rbHH260C0KYCNqmxDdcROUqIzJdzDKOrDmebkEVnxVeLJBIhYZUd t3Iq9hdjpU50TA6sQ3mZxzBdfRgg+vaj2DsJqI5Xla9QGKD+xNT6v14cZuIMZzO7w0DoojM4 ByrabFsOQxGvE0w9Dch2BDSI2Xyk1zjPKxG1VNBQVx3flH37QDWpL2zlJikW29Ws86PHdthh Fm5PY8YtX576DchSP6qJC57/eAAe/9ztZdVAdesQwGb9hZHJc75B+VNm4xrh/PJO6c1THqdQ 19WVJ+7rDx3PhVncGlbAOiiiE3NOFPJ1OQYxPKtpBUukAlOTnkKE6QcA4zckFepUkfmBV1wM Jg6OxFYd01z+a+oL
  • Cc: linux-block@xxxxxxxxxxxxxxx, xen-devel@xxxxxxxxxxxxxxxxxxxx
  • Delivery-date: Fri, 30 Nov 2018 22:34:05 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>
  • Openpgp: preference=signencrypt

On 11/30/18 4:49 PM, Manjunath Patil wrote:
> Thank you Boris for your comments. I removed faulty email of mine.
>
> replies inline.
> On 11/30/2018 12:42 PM, Boris Ostrovsky wrote:
>> On 11/29/18 12:17 AM, Manjunath Patil wrote:
>>> Hi,
>>> Feel free to suggest/comment on this.
>>>
>>> I am trying to do the following at dst during the migration now.
>>> 1. Dont clear the old rinfo in blkif_free(). Instead just clean it.
>>> 2. Store the old rinfo and nr_rings into temp variables in
>>> negotiate_mq()
>>> 3. let nr_rings get re-calculated based on backend data
>>> 4. try allocating new memory based on new nr_rings
>> Since I suspect number of rings will likely be the same why not reuse
>> the rings in the common case?
> I thought attaching devices will be more often than migration. Hence
> did not want add to an extra check for
>   - if I am inside migration code path and
>   - if new nr_rings is equal to old nr_rings or not
>
> Sure addition of such a thing would avoid the memory allocation
> altogether in migration path,
> but it would add a little overhead for normal device addition.
>
> Do you think its worth adding that change?


IMO a couple of extra checks are not going to make much difference.

I wonder though --- have you actually seen the case where you did fail
allocation and changes provided in this patch made things work? I am
asking because right after negotiate_mq() we will call setup_blkring()
and it will want to allocate bunch of memory. A failure there is fatal
(to ring setup). So it seems to me that you will survive negotiate_mq()
but then will likely fail soon after.


>>
>>
>>> 5.
>>>    a. If memory allocation is a success
>>>       - free the old rinfo and proceed to use the new rinfo
>>>    b. If memory allocation is a failure
>>>       - use the old the rinfo
>>>       - adjust the nr_rings to the lowest of new nr_rings and old
>>> nr_rings
>>
>>> @@ -1918,10 +1936,24 @@ static int negotiate_mq(struct blkfront_info
>>> *info)
>>>                     sizeof(struct blkfront_ring_info),
>>>                     GFP_KERNEL);
>>>       if (!info->rinfo) {
>>> -        xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating
>>> ring_info structure");
>>> -        info->nr_rings = 0;
>>> -        return -ENOMEM;
>>> -    }
>>> +        if (unlikely(nr_rings_old)) {
>>> +            /* We might waste some memory if
>>> +             * info->nr_rings < nr_rings_old
>>> +             */
>>> +            info->rinfo = rinfo_old;
>>> +            if (info->nr_rings > nr_rings_old)
>>> +                info->nr_rings = nr_rings_old;
>>> +            xenbus_dev_fatal(info->xbdev, -ENOMEM,
>>
>> Why xenbus_dev_fatal()?
> I wanted to make sure that this msg is seen on console by default. So
> that we know there was a enomem event happened and we recovered from it.
> What do you suggest instead? xenbus_dev_error?

Neither. xenbus_dev_fatal() is going to change connection state so it is
certainly not what we want. And even xenbus_dev_error() doesn't look
like the right thing to do since as far as block device setup is
concerned there are no errors.

Maybe pr_warn().

-boris


>>
>> -boris
>>
>>
>>> +            "reusing old ring_info structure(new ring size=%d)",
>>> +                info->nr_rings);
>>> +        } else {
>>> +            xenbus_dev_fatal(info->xbdev, -ENOMEM,
>>> +                "allocating ring_info structure");
>>> +            info->nr_rings = 0;
>>> +            return -ENOMEM;
>>> +        }
>>> +    } else if (unlikely(nr_rings_old))
>>> +        kfree(rinfo_old);
>>>         for (i = 0; i < info->nr_rings; i++) {
>>>           struct blkfront_ring_info *rinfo;
>


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.