[Bug 280037] KTLS with Intel QAT may trigger kernel panics

[email protected]
Newsgroups gmane.os.freebsd.devel.net
Message-ID <[email protected]/bugzilla/>
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=280037

--- Comment #17 from ss3bsd <[email protected]> ---
Created attachment 256343
  --> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=256343&action=edit
experimental patch that utilizes extra IRQ instances of QAT driver

I have created an experimental patch that utilizes extra IRQ instances (with a
queue for each) of QAT driver, which might improve the performance on high load
where driver stalls (EAGAIN) occur.

This patch introduces a sysctl entry to switch the behavior.

With the option enabled (dev.qat.0.use_extra_instances=1), 
 - I see fewer driver stalls in question
 - I could not actually confirm the throughput improvement because 10GbE is
saturated even with the option disabled


IRQ statistics shown in sysctl after some traffic:

<dev.qat.0.use_extra_instances: 0>

% sysctl -a | grep qat | grep irq

irq105: qat0:b0:217 @cpu0(domain0): 315445
irq106: qat0:b1:219 @cpu0(domain0): 197384
irq107: qat0:b2:221 @cpu0(domain0): 0
irq108: qat0:b3:223 @cpu0(domain0): 0
irq109: qat0:b4:225 @cpu0(domain0): 0
irq110: qat0:b5:227 @cpu0(domain0): 0
irq111: qat0:b6:229 @cpu0(domain0): 0
irq112: qat0:b7:231 @cpu0(domain0): 0
irq113: qat0:b8:233 @cpu0(domain0): 0
irq114: qat0:b9:235 @cpu0(domain0): 0
irq115: qat0:b10:237 @cpu0(domain0): 0
irq116: qat0:b11:239 @cpu0(domain0): 0
irq117: qat0:b12:241 @cpu0(domain0): 0
irq118: qat0:b13:243 @cpu0(domain0): 0
irq119: qat0:b14:245 @cpu0(domain0): 0
irq120: qat0:b15:247 @cpu0(domain0): 0
irq121: qat0:ae:249 @cpu0(domain0): 0


<dev.qat.0.use_extra_instances: 1>

% sysctl -a | grep qat | grep irq
irq105: qat0:b0:217 @cpu0(domain0): 117087
irq106: qat0:b1:219 @cpu1(domain0): 303001
irq107: qat0:b2:221 @cpu2(domain0): 631575
irq108: qat0:b3:223 @cpu3(domain0): 536653
irq109: qat0:b4:225 @cpu4(domain0): 534724
irq110: qat0:b5:227 @cpu5(domain0): 697166
irq111: qat0:b6:229 @cpu6(domain0): 708141
irq112: qat0:b7:231 @cpu7(domain0): 726079
irq113: qat0:b8:233 @cpu0(domain0): 0
irq114: qat0:b9:235 @cpu1(domain0): 0
irq115: qat0:b10:237 @cpu2(domain0): 0
irq116: qat0:b11:239 @cpu3(domain0): 0
irq117: qat0:b12:241 @cpu4(domain0): 0
irq118: qat0:b13:243 @cpu5(domain0): 0
irq119: qat0:b14:245 @cpu6(domain0): 0
irq120: qat0:b15:247 @cpu7(domain0): 0
irq121: qat0:ae:249 @cpu0(domain0): 0


Related topic found:
https://forums.freebsd.org/threads/qat-engines-and-cryptodev-usage-underutilised-card.95415/

Note that how many engines used (discussed at the topic) is perhaps irrelevant
to how many IRQs used.

-- 
You are receiving this mail because:
You are the assignee for the bug.
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.