From: Alexey Kardashevskiy <aik@ozlabs•ru>
To: Wei Yang <weiyang@linux•vnet.ibm.com>
Cc: Gavin Shan <gwshan@linux•vnet.ibm.com>,
benh@kernel•crashing.org, linuxppc-dev@ozlabs•org
Subject: Re: [PATCH V2 1/6] powerpc/powernv: don't enable SRIOV when VF BAR contains non M64 BAR
Date: Fri, 7 Aug 2015 11:24:42 +1000 [thread overview]
Message-ID: <55C408DA.1010406@ozlabs.ru> (raw)
In-Reply-To: <20150806141332.GE6235@richard>
On 08/07/2015 12:13 AM, Wei Yang wrote:
> On Thu, Aug 06, 2015 at 05:47:42PM +1000, Alexey Kardashevskiy wrote:
>> On 08/06/2015 04:57 PM, Gavin Shan wrote:
>>> On Thu, Aug 06, 2015 at 04:10:21PM +1000, Alexey Kardashevskiy wrote:
>>>> On 08/06/2015 02:35 PM, Gavin Shan wrote:
>>>>> On Wed, Aug 05, 2015 at 09:24:58AM +0800, Wei Yang wrote:
>>>>>> On PHB_IODA2, we enable SRIOV devices by mapping IOV BAR with M64 BARs. If
>>>>>> a SRIOV device's BAR is not 64-bit prefetchable, this is not assigned from
>>>>>> M64 windwo, which means M64 BAR can't work on it.
>>
>>
>> The proper text would be something like this:
>>
>> ===
>> SRIOV only supports 64bit MMIO. So if we fail to assign 64bit BAR, we
>> cannot enable the device.
>> ===
>>
>>
>>>>>>
>>>>>
>>>>> s/PHB_IODA2/PHB3
>>>>
>>>>
>>>> No, it is IODA2. OPEL does PHB3-specific bits, the host kernel just uses OPAL.
>>>>
>>>
>>> Ok.
>>>
>>>>
>>>>> s/windwo/window
>>>>>
>>>>>> This patch makes this explicit.
>>>>>>
>>>>>> Signed-off-by: Wei Yang <weiyang@linux•vnet.ibm.com>
>>>>>
>>>>> The idea sounds right, but there is one question as below.
>>>>>
>>>>>> ---
>>>>>> arch/powerpc/platforms/powernv/pci-ioda.c | 25 +++++++++----------------
>>>>>> 1 file changed, 9 insertions(+), 16 deletions(-)
>>>>>>
>>>>>> diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c b/arch/powerpc/platforms/powernv/pci-ioda.c
>>>>>> index 5738d31..9b41dba 100644
>>>>>> --- a/arch/powerpc/platforms/powernv/pci-ioda.c
>>>>>> +++ b/arch/powerpc/platforms/powernv/pci-ioda.c
>>>>>> @@ -908,9 +908,6 @@ static int pnv_pci_vf_resource_shift(struct pci_dev *dev, int offset)
>>>>>> if (!res->flags || !res->parent)
>>>>>> continue;
>>>>>>
>>>>>> - if (!pnv_pci_is_mem_pref_64(res->flags))
>>>>>> - continue;
>>>>>> -
>>>>>> /*
>>>>>> * The actual IOV BAR range is determined by the start address
>>>>>> * and the actual size for num_vfs VFs BAR. This check is to
>>>>>> @@ -939,9 +936,6 @@ static int pnv_pci_vf_resource_shift(struct pci_dev *dev, int offset)
>>>>>> if (!res->flags || !res->parent)
>>>>>> continue;
>>>>>>
>>>>>> - if (!pnv_pci_is_mem_pref_64(res->flags))
>>>>>> - continue;
>>>>>> -
>>>>>> size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
>>>>>> res2 = *res;
>>>>>> res->start += size * offset;
>>>>>> @@ -1221,9 +1215,6 @@ static int pnv_pci_vf_assign_m64(struct pci_dev *pdev, u16 num_vfs)
>>>>>> if (!res->flags || !res->parent)
>>>>>> continue;
>>>>>>
>>>>>> - if (!pnv_pci_is_mem_pref_64(res->flags))
>>>>>> - continue;
>>>>>> -
>>>>>> for (j = 0; j < vf_groups; j++) {
>>>>>> do {
>>>>>> win = find_next_zero_bit(&phb->ioda.m64_bar_alloc,
>>>>>> @@ -1510,6 +1501,12 @@ int pnv_pci_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
>>>>>> pdn = pci_get_pdn(pdev);
>>>>>>
>>>>>> if (phb->type == PNV_PHB_IODA2) {
>>>>>> + if (!pdn->vfs_expanded) {
>>>>>> + dev_info(&pdev->dev, "don't support this SRIOV device"
>>>>>> + " with non M64 VF BAR\n");
>>>>>> + return -EBUSY;
>>>>>> + }
>>>>>> +
>>>>>
>>>>> It would be -ENOSPC since -EBUSY indicates the devices (VFs) are temparily
>>>>> unavailable. For this case, the VFs are permanently unavailable because of
>>>>> running out of space to accomodate M64 and non-M64 VF BARs.
>>>>>
>>>>> The error message could be printed with dev_warn() and it would be precise
>>>>> as below or something else you prefer:
>>>>>
>>>>> dev_warn(&pdev->dev, "SRIOV not supported because of non-M64 VF BAR\n");
>>>>
>>>>
>>>> Both messages are cryptic.
>>>>
>>>> If it is not M64 BAR, then what is it? It is always in one of M64 BARs (in
>>>> the worst case - BAR#15?), the difference is if it is segmented or not, no?
>>>>
>>>
>>> The VF BAR could be one of IO, M32, M64. If it's not M64, the VFs are supposed
>>> to be disabled and the (IO and M32) resources won't be allocted, but for sure,
>>> the IO and M32 resources can't be put into any one of the 16 PHB's M64 BARs.
>>> would you recommend one better message then?
>>
>>
>>
>> dev_warn(&pdev->dev, "SRIOV is disabled as no space is left in 64bit
>> MMIO window\n");
>>
>> Or it is not "MMIO window"?
>>
>
> The reason is not "no space left in 64bit MMIO window".
>
> The reason is the IOV BAR is not 64bit prefetchable, then in linux kernel this
> can't be allocated from M64 Space, then we can't use M64 BAR to cover it.
Oh. So now it is not M64 vs. M32 and IO, it is about prefetchable vs.
non-prefetchable. Please choose one.
Should it be this then?
dev_warn(&pdev->dev, "Non-prefetchable BARs are not supported for SRIOV")
But Gavin keeps insisting on mentioning "non-M64 BAR" - this part I do not
understand.
And where does this limit come from? Is it POWER8, IODA2, PHB3, SRIOV or
something else? Is it all about isolation or the host _without_ KVM but
with SRIOV also cannot use VFs if they have non-prefetchable BARs? Is this
because of POWER8 or IODA2 or PHB3 or SRIOV or something else?
>>
>>
>>>
>>>>>
>>>>>
>>>>>> /* Calculate available PE for required VFs */
>>>>>> mutex_lock(&phb->ioda.pe_alloc_mutex);
>>>>>> pdn->offset = bitmap_find_next_zero_area(
>>>>>> @@ -2774,9 +2771,10 @@ static void pnv_pci_ioda_fixup_iov_resources(struct pci_dev *pdev)
>>>>>> if (!res->flags || res->parent)
>>>>>> continue;
>>>>>> if (!pnv_pci_is_mem_pref_64(res->flags)) {
>>>>>> - dev_warn(&pdev->dev, " non M64 VF BAR%d: %pR\n",
>>>>>> + dev_warn(&pdev->dev, "Don't support SR-IOV with"
>>>>>> + " non M64 VF BAR%d: %pR. \n",
>>>>>> i, res);
>>>>>> - continue;
>>>>>> + return;
>>>>>> }
>>>>>>
>>>>>> size = pci_iov_resource_size(pdev, i + PCI_IOV_RESOURCES);
>>>>>> @@ -2795,11 +2793,6 @@ static void pnv_pci_ioda_fixup_iov_resources(struct pci_dev *pdev)
>>>>>> res = &pdev->resource[i + PCI_IOV_RESOURCES];
>>>>>> if (!res->flags || res->parent)
>>>>>> continue;
>>>>>> - if (!pnv_pci_is_mem_pref_64(res->flags)) {
>>>>>> - dev_warn(&pdev->dev, "Skipping expanding VF BAR%d: %pR\n",
>>>>>> - i, res);
>>>>>> - continue;
>>>>>> - }
>>>>>
>>>>> When any one IOV BAR on the PF is non-M64, none of the VFs can be enabled.
>>>>> Will we still allocate/assign M64 or M32 resources for the IOV BARs? If so,
>>>>> I think it can be avoided.
>>>>>
>>>>>>
>>>>>> dev_dbg(&pdev->dev, " Fixing VF BAR%d: %pR to\n", i, res);
>>>>>> size = pci_iov_resource_size(pdev, i + PCI_IOV_RESOURCES);
>>
>>
>> --
>> Alexey
>
--
Alexey
next prev parent reply other threads:[~2015-08-07 1:24 UTC|newest]
Thread overview: 56+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-07-29 7:22 [PATCH] powerpc/powernv: use one M64 BAR in Single PE mode for one VF BAR Wei Yang
2015-07-30 1:15 ` Gavin Shan
2015-07-30 5:43 ` Wei Yang
2015-07-31 0:13 ` Gavin Shan
2015-07-31 2:01 ` Wei Yang
2015-08-05 1:24 ` [PATCH V2 0/6] Redesign SR-IOV on PowerNV Wei Yang
2015-08-05 1:24 ` [PATCH V2 1/6] powerpc/powernv: don't enable SRIOV when VF BAR contains non M64 BAR Wei Yang
2015-08-06 4:35 ` Gavin Shan
2015-08-06 6:10 ` Alexey Kardashevskiy
2015-08-06 6:57 ` Gavin Shan
2015-08-06 7:47 ` Alexey Kardashevskiy
2015-08-06 11:07 ` Gavin Shan
2015-08-06 14:13 ` Wei Yang
2015-08-07 1:24 ` Alexey Kardashevskiy [this message]
2015-08-06 14:10 ` Wei Yang
2015-08-07 1:20 ` Gavin Shan
2015-08-07 2:24 ` Wei Yang
2015-08-07 3:50 ` Gavin Shan
2015-08-07 7:14 ` Alexey Kardashevskiy
2015-08-10 1:40 ` Wei Yang
2015-08-05 1:24 ` [PATCH V2 2/6] powerpc/powernv: simplify the calculation of iov resource Wei Yang
2015-08-06 4:51 ` Gavin Shan
2015-08-06 9:00 ` Alexey Kardashevskiy
2015-08-06 9:41 ` Wei Yang
2015-08-06 10:15 ` Alexey Kardashevskiy
2015-08-07 1:36 ` Wei Yang
2015-08-06 13:49 ` Wei Yang
2015-08-07 1:08 ` Gavin Shan
2015-08-05 1:25 ` [PATCH V2 3/6] powerpc/powernv: use one M64 BAR in Single PE mode for one VF BAR Wei Yang
2015-08-06 5:20 ` Gavin Shan
2015-08-06 9:36 ` Wei Yang
2015-08-06 10:07 ` Gavin Shan
2015-08-07 1:48 ` Wei Yang
2015-08-07 8:13 ` Alexey Kardashevskiy
2015-08-06 10:04 ` Alexey Kardashevskiy
2015-08-07 2:01 ` Wei Yang
2015-08-07 8:59 ` Alexey Kardashevskiy
2015-08-10 1:48 ` Wei Yang
2015-08-05 1:25 ` [PATCH V2 4/6] powerpc/powernv: replace the hard coded boundary with gate Wei Yang
2015-08-06 5:26 ` Gavin Shan
2015-08-07 9:11 ` Alexey Kardashevskiy
2015-08-05 1:25 ` [PATCH V2 5/6] powerpc/powernv: boundary the total vf bar size instead of the individual one Wei Yang
2015-08-06 5:28 ` Gavin Shan
2015-08-06 14:03 ` Wei Yang
2015-08-07 1:23 ` Gavin Shan
2015-08-07 2:25 ` Wei Yang
2015-08-05 1:25 ` [PATCH V2 6/6] powerpc/powernv: allocate discrete PE# when using M64 BAR in Single PE mode Wei Yang
2015-08-06 5:36 ` Gavin Shan
2015-08-06 13:41 ` Wei Yang
2015-08-07 1:36 ` Gavin Shan
2015-08-07 2:33 ` Wei Yang
2015-08-07 3:43 ` Gavin Shan
2015-08-07 5:44 ` Wei Yang
2015-08-07 5:54 ` Gavin Shan
2015-08-07 6:25 ` Wei Yang
2015-08-07 10:00 ` Alexey Kardashevskiy
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=55C408DA.1010406@ozlabs.ru \
--to=aik@ozlabs$(echo .)ru \
--cc=benh@kernel$(echo .)crashing.org \
--cc=gwshan@linux$(echo .)vnet.ibm.com \
--cc=linuxppc-dev@ozlabs$(echo .)org \
--cc=weiyang@linux$(echo .)vnet.ibm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox