Re: [PATCH v26 2/5] fs/proc/task_mmu: Implement IOCTL to get and optionally clear info about PTEs

From: Muhammad Usama Anjum <usama.anjum@collabora.com>
To: Andrei Vagin <avagin@gmail.com>
Cc: "Muhammad Usama Anjum" <usama.anjum@collabora.com>,
	"Peter Xu" <peterx@redhat.com>,
	"David Hildenbrand" <david@redhat.com>,
	"Andrew Morton" <akpm@linux-foundation.org>,
	"Michał Mirosław" <emmir@google.com>,
	"Danylo Mocherniuk" <mdanylo@google.com>,
	"Paul Gofman" <pgofman@codeweavers.com>,
	"Cyrill Gorcunov" <gorcunov@gmail.com>,
	"Mike Rapoport" <rppt@kernel.org>,
	"Nadav Amit" <namit@vmware.com>,
	"Alexander Viro" <viro@zeniv.linux.org.uk>,
	"Shuah Khan" <shuah@kernel.org>,
	"Christian Brauner" <brauner@kernel.org>,
	"Yang Shi" <shy828301@gmail.com>,
	"Vlastimil Babka" <vbabka@suse.cz>,
	"Liam R . Howlett" <Liam.Howlett@oracle.com>,
	"Yun Zhou" <yun.zhou@windriver.com>,
	"Suren Baghdasaryan" <surenb@google.com>,
	"Alex Sierra" <alex.sierra@amd.com>,
	"Matthew Wilcox" <willy@infradead.org>,
	"Pasha Tatashin" <pasha.tatashin@soleen.com>,
	"Axel Rasmussen" <axelrasmussen@google.com>,
	"Gustavo A . R . Silva" <gustavoars@kernel.org>,
	"Dan Williams" <dan.j.williams@intel.com>,
	linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
	linux-mm@kvack.org, linux-kselftest@vger.kernel.org,
	"Greg KH" <gregkh@linuxfoundation.org>,
	kernel@collabora.com, "Michał Mirosław" <mirq-linux@rere.qmqm.pl>
Subject: Re: [PATCH v26 2/5] fs/proc/task_mmu: Implement IOCTL to get and optionally clear info about PTEs
Date: Mon, 7 Aug 2023 09:28:19 +0500	[thread overview]
Message-ID: <10f947a2-3917-a5fe-837d-214d70991bcb@collabora.com> (raw)
In-Reply-To: <CANaxB-zZFq7VD7tBBUmACUJPE9iVuTyQKfg4Jw82-U_1qw6ALg@mail.gmail.com>

On 8/5/23 2:53 AM, Andrei Vagin wrote:
> On Thu, Jul 27, 2023 at 2:37 AM Muhammad Usama Anjum
> <usama.anjum@collabora.com> wrote:
>>
> 
> <snip>
> 
>> +static long do_pagemap_scan(struct mm_struct *mm, unsigned long uarg)
>> +{
>> +       unsigned long walk_start, walk_end;
>> +       struct mmu_notifier_range range;
>> +       struct pagemap_scan_private p;
>> +       size_t n_ranges_out = 0;
>> +       int ret;
>> +
>> +       memset(&p, 0, sizeof(p));
>> +       ret = pagemap_scan_get_args(&p.arg, uarg);
>> +       if (ret)
>> +               return ret;
>> +
>> +       ret = pagemap_scan_init_bounce_buffer(&p);
>> +       if (ret)
>> +               return ret;
>> +
>> +       /* Protection change for the range is going to happen. */
>> +       if (p.arg.flags & PM_SCAN_WP_MATCHING) {
>> +               mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_VMA, 0,
>> +                                       mm, p.arg.start, p.arg.end);
>> +               mmu_notifier_invalidate_range_start(&range);
>> +       }
>> +
>> +       walk_start = walk_end = p.arg.start;
>> +       for (; walk_end != p.arg.end; walk_start = walk_end) {
>> +               int n_out;
>> +
>> +               walk_end = min_t(unsigned long,
>> +                                (walk_start + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK,
>> +                                p.arg.end);
> 
> This approach has performance implications. The basic program that scans
> its address space takes around 20-30 seconds, but it has just a few
> small mappings. The first optimization that comes to mind is to remove
> the PAGEMAP_WALK_SIZE limit and instead halt walk_page_range when the
> bounce buffer is full. After draining the buffer, the walk_page_range
> function can be restarted.
Yeah, I've this implemented in WIP and will be posting in next revision.

> 
> The test program and perf data can be found here:
> https://gist.github.com/avagin/c5a22f3c78f8cb34281602dfe9c43d10
> 
>> +
>> +               ret = mmap_read_lock_killable(mm);
>> +               if (ret)
>> +                       break;
>> +               ret = walk_page_range(mm, walk_start, walk_end,
>> +                                     &pagemap_scan_ops, &p);
>> +               mmap_read_unlock(mm);
>> +
>> +               n_out = pagemap_scan_flush_buffer(&p);
>> +               if (n_out < 0)
>> +                       ret = n_out;
>> +               else
>> +                       n_ranges_out += n_out;
>> +
>> +               if (ret)
>> +                       break;
>> +       }
>> +
> 
> Thanks,
> Andrei

-- 
BR,
Muhammad Usama Anjum