<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<p>OK, Let's drop mine, then Could you draft a solution for that?</p>
<p><br>
</p>
<p>-David<br>
</p>
<br>
<div class="moz-cite-prefix">On 2019年04月24日 16:59, Koenig, Christian
wrote:<br>
</div>
<blockquote type="cite"
cite="mid:9e6e865f-29dd-be69-fada-f6a6b35e6c69@amd.com">
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<div class="moz-cite-prefix">Am 24.04.19 um 10:11 schrieb zhoucm1:<br>
</div>
<blockquote type="cite"
cite="mid:cc8a8a83-a200-528f-9d31-a16115c3c00f@amd.com">
<p><br>
</p>
<br>
<div class="moz-cite-prefix">On 2019年04月24日 16:07, Christian
König wrote:<br>
</div>
<blockquote type="cite"
cite="mid:f4f27f33-5e9f-985e-ca89-5cef39bf7aaf@gmail.com">
<div class="moz-cite-prefix">This is used in a work item, so
you don't need to check for signals.<br>
</div>
</blockquote>
will remove.<br>
<blockquote type="cite"
cite="mid:f4f27f33-5e9f-985e-ca89-5cef39bf7aaf@gmail.com">
<div class="moz-cite-prefix"><br>
And checking if the LRU is populated is mandatory here </div>
</blockquote>
How to check it outside of TTM? because the code is in dm.<br>
</blockquote>
<br>
Well just use a static cast on the first entry of the LRU.<br>
<br>
We can't upstream that solution anyway, so just a hack should do
for now.<br>
<br>
<blockquote type="cite"
cite="mid:cc8a8a83-a200-528f-9d31-a16115c3c00f@amd.com"><br>
<blockquote type="cite"
cite="mid:f4f27f33-5e9f-985e-ca89-5cef39bf7aaf@gmail.com">
<div class="moz-cite-prefix">or otherwise you can run into an
endless loop.<br>
</div>
</blockquote>
I already add a timeout for that.<br>
</blockquote>
<br>
Thinking more about it we most likely actually need to grab the
lock on the first BO entry from the LRU.<br>
<br>
<blockquote type="cite"
cite="mid:cc8a8a83-a200-528f-9d31-a16115c3c00f@amd.com"><br>
-David<br>
<blockquote type="cite"
cite="mid:f4f27f33-5e9f-985e-ca89-5cef39bf7aaf@gmail.com">
<div class="moz-cite-prefix"><br>
Christian.<br>
<br>
Am 24.04.19 um 09:59 schrieb zhoucm1:<br>
</div>
<blockquote type="cite"
cite="mid:26341685-6e73-55c4-2609-52616d92c06a@amd.com">
<p>how about new attached?</p>
<p><br>
</p>
<p>-David<br>
</p>
<br>
<div class="moz-cite-prefix">On 2019年04月24日 15:30, Christian
König wrote:<br>
</div>
<blockquote type="cite"
cite="mid:8db16fe8-78fe-7f25-4acd-51e37e645f3b@gmail.com">
<div class="moz-cite-prefix">That would change the
semantics of ttm_bo_mem_space() and so could change the
return code in an IOCTL as well. Not a good idea, cause
that could have a lot of side effects.<br>
<br>
Instead in the calling DC code you could check if you
get an -ENOMEM and then call schedule().<br>
<br>
If after the schedule() we see that we have now BOs on
the LRU we can try again and see if pinning the frame
buffer now succeeds.<br>
<br>
Christian.<br>
<br>
Am 24.04.19 um 09:17 schrieb zhoucm1:<br>
</div>
<blockquote type="cite"
cite="mid:b57d94e6-6e20-d809-7ac2-89a39a57a365@amd.com">
<p>I made a patch as attached.</p>
<p>I'm not sure how to make patch as your proposal,
Could you make a patch for that if mine isn't enough?<br>
</p>
-David<br>
<br>
<div class="moz-cite-prefix">On 2019年04月24日 15:12,
Christian König wrote:<br>
</div>
<blockquote type="cite"
cite="mid:32f2a1b7-99f8-de9a-799c-f98af308842b@gmail.com">
<div class="moz-cite-prefix">
<blockquote type="cite">how about just adding a
wrapper for pin function as below?</blockquote>
I considered this as well and don't think it will
work reliable.<br>
<br>
We could use it as a band aid for this specific
problem, but in general we need to improve the
handling in TTM to resolve those kind of resource
conflicts.<br>
<br>
Regards,<br>
Christian.<br>
<br>
Am 23.04.19 um 17:09 schrieb Zhou, David(ChunMing):<br>
</div>
<blockquote type="cite"
cite="mid:-jm8957cqk536djh1631fvvv-xx4wzb5q21ak-v8q7rd2l14aq-f68kxr7kbea18a7xceae626b-8s84c4d1mgrg53g0bhq3ahee89h70qrv4ly1-vf5a7d63x4mbquxnfmiehuk2-rwaw2m-qc2chu.1556032141262@email.android.com">
<meta content="text/html; charset=Windows-1252">
>3. If we have a ticket we grab a reference to
the first BO on the LRU, drop the LRU lock and try
to grab the reservation lock with the ticket.<br>
<br>
The BO on LRU is already locked by cs user, can it
be dropped here by DC user? and then DC user grab
its lock with ticket, how does CS grab it again?<br>
<br>
If you think waiting in ttm has this risk, how about
just adding a wrapper for pin function as below?<br>
amdgpu_get_pin_bo_timeout()<br>
{<br>
do {<br>
amdgpo_bo_reserve();<br>
r=amdgpu_bo_pin();<br>
<br>
if(!r)<br>
break;<br>
amdgpu_bo_unreserve();<br>
timeout--;<br>
<br>
} while(timeout>0);<br>
<br>
}<br>
<br>
-------- Original Message --------<br>
Subject: Re: [PATCH] ttm: wait mem space if user
allow while gpu busy<br>
From: Christian König <br>
To: "Zhou, David(ChunMing)" ,"Koenig, Christian"
,"Liang, Prike" ,<a class="moz-txt-link-abbreviated"
href="mailto:dri-devel@lists.freedesktop.org"
moz-do-not-send="true">dri-devel@lists.freedesktop.org</a><br>
CC: <br>
<br>
<div>
<div class="moz-cite-prefix">Well that's not so
easy of hand.<br>
<br>
The basic problem here is that when you busy
wait at this place you can easily run into
situations where application A busy waits for B
while B busy waits for A -> deadlock.<br>
<br>
So what we need here is the deadlock detection
logic of the ww_mutex. To use this we at least
need to do the following steps:<br>
<br>
1. Reserve the BO in DC using a ww_mutex ticket
(trivial).<br>
<br>
2. If we then run into this EBUSY condition in
TTM check if the BO we need memory for (or
rather the ww_mutex of its reservation object)
has a ticket assigned.<br>
<br>
3. If we have a ticket we grab a reference to
the first BO on the LRU, drop the LRU lock and
try to grab the reservation lock with the
ticket.<br>
<br>
4. If getting the reservation lock with the
ticket succeeded we check if the BO is still the
first one on the LRU in question (the BO could
have moved).<br>
<br>
5. If the BO is still the first one on the LRU
in question we try to evict it as we would evict
any other BO.<br>
<br>
6. If any of the "If's" above fail we just back
off and return -EBUSY.<br>
<br>
Steps 2-5 are certainly not trivial, but doable
as far as I can see.<br>
<br>
Have fun :)<br>
Christian.<br>
<br>
Am 23.04.19 um 15:19 schrieb Zhou,
David(ChunMing):<br>
</div>
<blockquote type="cite">
<meta name="Generator" content="Microsoft
Exchange Server">
<style>
<!--
.EmailQuote
{margin-left:1pt;
padding-left:4pt;
border-left:#800000 2px solid}
-->
</style>
<div>How about adding more condition
ctx->resv inline to address your concern?
As well as don't wait from same user,
shouldn't lead to deadlock.<br>
<br>
Otherwise, any other idea?<br>
<br>
-------- Original Message --------<br>
Subject: Re: [PATCH] ttm: wait mem space if
user allow while gpu busy<br>
From: Christian König <br>
To: "Liang, Prike" ,"Zhou, David(ChunMing)" ,<a
class="moz-txt-link-abbreviated"
href="mailto:dri-devel@lists.freedesktop.org"
moz-do-not-send="true">dri-devel@lists.freedesktop.org</a><br>
CC: <br>
<br>
</div>
<font size="2"><span style="font-size:11pt">
<div class="PlainText">Well that is
certainly a NAK because it can lead to
deadlock in the
<br>
memory management.<br>
<br>
You can't just busy wait with all those
locks held.<br>
<br>
Regards,<br>
Christian.<br>
<br>
Am 23.04.19 um 03:45 schrieb Liang, Prike:<br>
> Acked-by: Prike Liang <a
class="moz-txt-link-rfc2396E"
href="mailto:Prike.Liang@amd.com"
moz-do-not-send="true">
<Prike.Liang@amd.com></a><br>
><br>
> Thanks,<br>
> Prike<br>
> -----Original Message-----<br>
> From: Chunming Zhou <a
class="moz-txt-link-rfc2396E"
href="mailto:david1.zhou@amd.com"
moz-do-not-send="true">
<david1.zhou@amd.com></a><br>
> Sent: Monday, April 22, 2019 6:39 PM<br>
> To: <a
class="moz-txt-link-abbreviated"
href="mailto:dri-devel@lists.freedesktop.org"
moz-do-not-send="true">
dri-devel@lists.freedesktop.org</a><br>
> Cc: Liang, Prike <a
class="moz-txt-link-rfc2396E"
href="mailto:Prike.Liang@amd.com"
moz-do-not-send="true">
<Prike.Liang@amd.com></a>; Zhou,
David(ChunMing) <a
class="moz-txt-link-rfc2396E"
href="mailto:David1.Zhou@amd.com"
moz-do-not-send="true">
<David1.Zhou@amd.com></a><br>
> Subject: [PATCH] ttm: wait mem space
if user allow while gpu busy<br>
><br>
> heavy gpu job could occupy memory
long time, which could lead to other user
fail to get memory.<br>
><br>
> Change-Id:
I0b322d98cd76e5ac32b00462bbae8008d76c5e11<br>
> Signed-off-by: Chunming Zhou <a
class="moz-txt-link-rfc2396E"
href="mailto:david1.zhou@amd.com"
moz-do-not-send="true">
<david1.zhou@amd.com></a><br>
> ---<br>
> drivers/gpu/drm/ttm/ttm_bo.c | 6
++++--<br>
> 1 file changed, 4 insertions(+), 2
deletions(-)<br>
><br>
> diff --git
a/drivers/gpu/drm/ttm/ttm_bo.c
b/drivers/gpu/drm/ttm/ttm_bo.c index
7c484729f9b2..6c596cc24bec 100644<br>
> --- a/drivers/gpu/drm/ttm/ttm_bo.c<br>
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c<br>
> @@ -830,8 +830,10 @@ static int
ttm_bo_mem_force_space(struct
ttm_buffer_object *bo,<br>
> if (mem->mm_node)<br>
> break;<br>
> ret =
ttm_mem_evict_first(bdev, mem_type, place,
ctx);<br>
> - if (unlikely(ret != 0))<br>
> - return ret;<br>
> + if (unlikely(ret != 0))
{<br>
> + if (!ctx ||
ctx->no_wait_gpu || ret != -EBUSY)<br>
> + return
ret;<br>
> + }<br>
> } while (1);<br>
> mem->mem_type = mem_type;<br>
> return
ttm_bo_add_move_fence(bo, man, mem);<br>
> --<br>
> 2.17.1<br>
><br>
>
_______________________________________________<br>
> dri-devel mailing list<br>
> <a class="moz-txt-link-abbreviated"
href="mailto:dri-devel@lists.freedesktop.org" moz-do-not-send="true">
dri-devel@lists.freedesktop.org</a><br>
> <a
href="https://lists.freedesktop.org/mailman/listinfo/dri-devel"
moz-do-not-send="true">
https://lists.freedesktop.org/mailman/listinfo/dri-devel</a><br>
<br>
</div>
</span></font><br>
<fieldset class="mimeAttachmentHeader"></fieldset>
<pre class="moz-quote-pre">_______________________________________________
dri-devel mailing list
<a class="moz-txt-link-abbreviated" href="mailto:dri-devel@lists.freedesktop.org" moz-do-not-send="true">dri-devel@lists.freedesktop.org</a>
<a class="moz-txt-link-freetext" href="https://lists.freedesktop.org/mailman/listinfo/dri-devel" moz-do-not-send="true">https://lists.freedesktop.org/mailman/listinfo/dri-devel</a></pre>
</blockquote>
<br>
</div>
<br>
<fieldset class="mimeAttachmentHeader"></fieldset>
<pre class="moz-quote-pre" wrap="">_______________________________________________
dri-devel mailing list
<a class="moz-txt-link-abbreviated" href="mailto:dri-devel@lists.freedesktop.org" moz-do-not-send="true">dri-devel@lists.freedesktop.org</a>
<a class="moz-txt-link-freetext" href="https://lists.freedesktop.org/mailman/listinfo/dri-devel" moz-do-not-send="true">https://lists.freedesktop.org/mailman/listinfo/dri-devel</a></pre>
</blockquote>
<br>
</blockquote>
<br>
<br>
<fieldset class="mimeAttachmentHeader"></fieldset>
<pre class="moz-quote-pre" wrap="">_______________________________________________
dri-devel mailing list
<a class="moz-txt-link-abbreviated" href="mailto:dri-devel@lists.freedesktop.org" moz-do-not-send="true">dri-devel@lists.freedesktop.org</a>
<a class="moz-txt-link-freetext" href="https://lists.freedesktop.org/mailman/listinfo/dri-devel" moz-do-not-send="true">https://lists.freedesktop.org/mailman/listinfo/dri-devel</a></pre>
</blockquote>
<br>
</blockquote>
<br>
<br>
<fieldset class="mimeAttachmentHeader"></fieldset>
<pre class="moz-quote-pre" wrap="">_______________________________________________
dri-devel mailing list
<a class="moz-txt-link-abbreviated" href="mailto:dri-devel@lists.freedesktop.org" moz-do-not-send="true">dri-devel@lists.freedesktop.org</a>
<a class="moz-txt-link-freetext" href="https://lists.freedesktop.org/mailman/listinfo/dri-devel" moz-do-not-send="true">https://lists.freedesktop.org/mailman/listinfo/dri-devel</a></pre>
</blockquote>
<br>
</blockquote>
<br>
</blockquote>
<br>
</blockquote>
<br>
</body>
</html>