[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <el52wvltm2ptkyjhiajeo564sa6kcwqihdttvutem2qoegj5rg@wnqe7flapgbf>
Date: Fri, 28 Mar 2025 12:54:42 +0200
From: "Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>
To: Ard Biesheuvel <ardb@...nel.org>
Cc: "Kirill A. Shutemov" <kirill@...temov.name>,
Steven Rostedt <rostedt@...dmis.org>, Tom Lendacky <thomas.lendacky@....com>,
"Aithal, Srikanth" <sraithal@....com>, Jason Baron <jbaron@...mai.com>,
Peter Zijlstra <peterz@...radead.org>, Josh Poimboeuf <jpoimboe@...nel.org>,
Linux-Next Mailing List <linux-next@...r.kernel.org>, open list <linux-kernel@...r.kernel.org>,
"Roth, Michael" <Michael.Roth@....com>
Subject: Re: linux-next regression: SNP Guest boot hangs with certain cpu/mem
config combination
On Fri, Mar 28, 2025 at 10:33:31AM +0100, Ard Biesheuvel wrote:
> Can you quantify the speedup?
Test is below. I run it 10 times on a VM without unaccepted memory. With
and without has_unaccepted_memory() check in cond_accept_memory().
The difference is not huge, but it is there:
Without static branch: Mean: 35559993 us, StdDev: 167264
With static branch: Mean: 35286227 us, StdDev: 207595
Diff: -273766 us / -0.77%
static int __init alloc_test(void)
{
ktime_t start_time, end_time;
s64 delta, total = 0;
struct page *page;
int i, j;
for (i = 0; i < 10; i++) {
start_time = ktime_get();
for (j = 0; j < 100000; j++) {
page = alloc_pages(GFP_KERNEL, PMD_ORDER);
__free_pages(page, PMD_ORDER);
}
end_time = ktime_get();
printk("%d: %lld us\n", i, ktime_us_delta(end_time, start_time));
}
for (i = 0; i < 100; i++) {
start_time = ktime_get();
for (j = 0; j < 100000; j++) {
page = alloc_pages(GFP_KERNEL, PMD_ORDER);
__free_pages(page, PMD_ORDER);
}
end_time = ktime_get();
delta = ktime_us_delta(end_time, start_time);
total += delta;
printk("%d: %lld us\n", i, delta);
}
printk("total: %lld us\n", total);
return 0;
}
late_initcall(alloc_test);
--
Kiryl Shutsemau / Kirill A. Shutemov
Powered by blists - more mailing lists