- kernelnewbies@kernelnewbies.org - Kernelnewbies Mailing Lists

radix_tree_next_chunk: redundant search for next slot in hole
by Probir Roy May 4, 2019

May 4, 2019

Hi, I am running Phoronix-fio benchmark on Linux kernel 4.18.0-rc5. I observe that radix_tree_next_chunk search shows some form of a redundant walk while called from radix_tree_for_each_slot iterator. While searching for next slot in a hole, it walks through the same slots over n over. void __rcu **radix_tree_next_chunk(const struct radix_tree_root *root, struct radix_tree_iter *iter, unsigned flags) { ... if ((flags & RADIX_TREE_ITER_TAGGED) ? !tag_get(node, tag, offset) : !child) { ... while (++offset < RADIX_TREE_MAP_SIZE) { void *slot = rcu_dereference_raw( /* redundant slot walk */ node->slots[offset]); if (slot) break; } } ... } Can you please explain such behavior? Any feedback/recommendation about this case is very welcome. -- Probir Roy

2 3

ULIMIT: limiting user virtual address space (stack inluded)
by Karaoui mohamed lamine May 2, 2019

May 2, 2019

Hi all, According to the man page of "ulimit", it is possible to limit the virtual address space of a user process: https://ss64.com/bash/ulimit.html To limit the virtual address space to 2 GB: $ ulimit -SH -v 1048576 However, it seems that "stack" allocation ignores this limit (I tried with both kernel 4.4 and 4.15). I found this article (patch) that seems to fix the problem: https://lwn.net/Articles/373701/ My questions are: - Why is the stack allocation not placed within the "ulimit" limit? (is it a bug?) - Is there a way, in user space, to force the stack to be allocated at a certain address? - Will the patch above (or similar) ever be integrated into the Linux kernel? Regards, Mohamed

3 2

Re: debugfs question ...
by 'Greg KH' May 1, 2019

May 1, 2019

On Wed, May 01, 2019 at 03:18:51PM +0100, rdq(a)metamail.co wrote: > Hello and thanks, > > > > > > > The pattern for the implementation is (AFAICT) right out of the book > > > > > > > You are returning a "short" read, and then disallowing ppos to be set to a > > non-zero value? That's a recipie for disaster :( > > > > Also, you allow userspace to allocate as much memory as it asks for? > > Not good :( > > Well I started here: > https://github.com/torvalds/linux/blob/v4.14/sound/soc/soc-core.c#L251. > > > Why not just use the simple_read_from_buffer() call? That handles all of > the > > "housekeeping" for you, and your function can be _much_ simpler. > > Possibly but it does not explain why returning 0 means cat displays nothing You returned nothing, so what is cat supposed to do? It worked properly. > and returning the string length causes it to loop indefinitely. You have to give userspace what it asks for, don't ignore the ppos variable. Have you tried writing a userspace program that does a correct read() call on a file or device node like this? You have to keep reading until you get all of the data you asked for, or the kernel returns end-of-file, or an error happens. You don't just do a single read and hope for the best :) > My example now reduced to: > > int ret; > char buf[64]; > int n = snprintf(buf,64,"Hello World\n"); > if (n < 0) > return -EINVAL; > // zero on success > ret = copy_to_user(user_buf,buf,n); > *ppos += n; > return ret; > > Is this the canonical form? Not at all, that's a recipe for disaster. > Using a 4.14 kernel, cat displays nothing. Or has something changed? > If so mighty odd as there are ancient example as well as the new: > > https://sites.google.com/site/linuxkernel88/sample-code/writing-a-character… in char_device_read() A char device is MUCH different from a file interface that you are using here. It kind of looks the same, but it has some subtle ways in which things can go sideways very easily. Also, that code is really really wrong, if that char_device_read() function in the above link was ever submitted to me, I would have a field day with it. Nothing like bad random internet examples to steer people in the wrong direction :( > https://github.com/torvalds/linux/blob/master/drivers/base/regmap/regmap-de… That one is not "quite" like your example here either. You have the option to handle all of the fun corner cases and error values by writing it yourself, or just use the helper function I pointed you at which does all of what you need to do, correctly. To do something in the middle is to guarantee that something will not work properly with some sort of userspace program, or be totally insecure. thanks, greg k-h

1 0

debugfs question ...
by rdq＠metamail.co May 1, 2019

May 1, 2019

Greetings, My I2C sensor driver has a debugfs entry for development purposes. Everything works fine with the exception of the read operation. When 'cat' is used, the read operation is called repeatedly and indefinitely. If the read() is changed to return 0 then, as expected, nothing is displayed. The pattern for the implementation is (AFAICT) right out of the book (shown below). What am I missing? Any thoughts much appreciated. TAIA. RDQ static ssize_t sc031gs_reg_read_file(struct file *file, char __user *user_buf, size_t count, loff_t *ppos) { char *buf = 0; ssize_t total = 0; struct sc031gs_dev *sensor = file->private_data; if (!sensor) return -EINVAL; if (*ppos < 0 || !count) return -EINVAL; buf = kmalloc(count, GFP_KERNEL); if (!buf) return -ENOMEM; total = snprintf(buf,count,"Hello world\n"); if (total >= 0) { if (copy_to_user(user_buf, buf, total)) { kfree(buf); return -EFAULT; } *ppos += total; } kfree(buf); return total; }

3 3

vmscan.c: Reclaim unevictable pages.
by Pankaj Suryawanshi April 28, 2019

April 28, 2019

Hello , shrink_page_list() returns , number of pages reclaimed, when pages is unevictable it returns VM_BUG_ON_PAGE(PageLRU(page) || PageUnevicatble(page),page); We can add the unevictable pages in reclaim list in shrink_page_list(), return total number of reclaim pages including unevictable pages, let the caller handle unevictable pages. I think the problem is shrink_page_list is awkard. If page is unevictable it goto activate_locked->keep_locked->keep lables, keep lable list_add the unevictable pages and throw the VM_BUG instead of passing it to caller while it relies on caller for non-reclaimed-non-unevictable page's putback. I think we can make it consistent so that shrink_page_list could return non-reclaimed pages via page_list and caller can handle it. As an advance, it could try to migrate mlocked pages without retrial. Below is the issue i observed of CMA_ALLOC of large size buffer : (Kernel version - 4.14.65 With Android Pie. [ 24.718792] page dumped because: VM_BUG_ON_PAGE(PageLRU(page) || PageUnevictable(page)) [ 24.726949] page->mem_cgroup:bd008c00 [ 24.730693] ------------[ cut here ]------------ [ 24.735304] kernel BUG at mm/vmscan.c:1350! [ 24.739478] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM Below is the patch which solved this issue : diff --git a/mm/vmscan.c b/mm/vmscan.c index be56e2e..12ac353 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c @@ -998,7 +998,7 @@ static unsigned long shrink_page_list(struct list_head *page_list, sc->nr_scanned++; if (unlikely(!page_evictable(page))) - goto activate_locked; + goto cull_mlocked; if (!sc->may_unmap && page_mapped(page)) goto keep_locked; @@ -1331,7 +1331,12 @@ static unsigned long shrink_page_list(struct list_head *page_list, } else list_add(&page->lru, &free_pages); continue; - +cull_mlocked: + if (PageSwapCache(page)) + try_to_free_swap(page); + unlock_page(page); + list_add(&page->lru, &ret_pages); + continue; activate_locked: /* Not a candidate for swapping, so reclaim swap space. */ if (PageSwapCache(page) && (mem_cgroup_swap_full(page) || It fixes the below issue. 1. Large size buffer allocation using cma_alloc successful with unevictable pages. cma_alloc of current kernel will fail due to unevictable page Please let me know if anything i am missing. Regards, Pankaj

2 2

Best practice to upgrade kernel
by Xiaolong Wang April 26, 2019

April 26, 2019

Hi all, I have a newbie question. I’m using a 3.18.66 kernel branch from Qualcomm. Over the year I’ve applied some of patched here and there manually. But It seems Qualcomm does not patch lot of newly released vulnerabilities. I would love to upgrade to the latest 3.18.138 (not 4.x) What is the best practice here? Thanks -Dan

3 2

Partition linux page cache for each VM
by Chirag Chauhan April 26, 2019

April 26, 2019

Hey, I need to modify pagecache such that each VM gets their own LRU list and eviction is done from that list. I need to verify whether I'm on the right path, so can anyone please verify my approach. The first step is to identify which VM is making the call to allocate a page. As VMs are stored as a file, I can identify each VM with their filename, right? For this, I'm trying to get the filename associated with an address_space object when add_to_page_cache_lru() is called in filemap.c >From address_space object, I can get inode and therefore dentry. I'm not able to figure out how to get the name of a file. And can you please provide brief overview of what the overall approach should be? Any comments would help. Thanks. Chirag

3 4

__clksrc_of_table and "clocksource_probe: no matching clocksources found"
by Guilherme Costa April 23, 2019

April 23, 2019

Hi, I'm working with kernel 4.9.51 LTSI on a custom board with an AM3352. The system seems to be working just fine, but I'm getting the following message on boot: "clocksource_probe: no matching clocksources found" I did some snooping around and this message is printed by clocksource_probe(), called during time_init(). This is the function code: void __init clocksource_probe(void) { struct device_node *np; const struct of_device_id *match; of_init_fn_1_ret init_func_ret; unsigned clocksources = 0; int ret; for_each_matching_node_and_match(np, __clksrc_of_table, &match) { if (!of_device_is_available(np)) continue; init_func_ret = match->data; ret = init_func_ret(np); if (ret) { pr_err("Failed to initialize '%s': %d", of_node_full_name(np), ret); continue; } clocksources++; } clocksources += acpi_probe_device_table(clksrc); if (!clocksources) pr_crit("%s: no matching clocksources found\n", __func__); } >From the code, we can see that the message was printed because clocksources == 0. This implies that there were no matching nodes on the device tree, and that acpi_probe_device_table returned 0 (which is correct, seeing that the kernel has no ACPI support). This leaves me with two questions: 1 - Is this message a problem indicator? I'm assuming it is, because it's printed with a pr_crit... 2 - Why is for_each_matching_node_and_match not getting any matches? I did not find where __clksrc_of_table is initialized, so maybe it's because it is empty? Cheers, Guilherme

2 2

is there a simple graphic of SATA port multiplier structures?
by Robert P. J. Day April 23, 2019

April 23, 2019

was just asked to take a look at how SATA II port multipliers are supported in the linux kernel and, in the midst of all the code and header files, what seems to be missing is a simple graphic of the main structures and how many of them there are. let me elucidate. imagine, say, a 1-to-4 PMP -- single port multiplexing to four SATA HDDs. from include/linux/libata.h, i can see the ATA-related structures that will be involved, but it would be just ducky if there was a short description as to *how* *many* of each would be involved in this scenario; that is, a simple counting demonstration would be massively informative. first, it appears that there would be a single ata_port structure: struct ata_port { struct Scsi_Host *scsi_host; /* our co-allocated scsi host */ struct ata_port_operations *ops; spinlock_t *lock; /* Flags owned by the EH context. Only EH should touch these once the port is active */ unsigned long flags; /* ATA_FLAG_xxx */ ... snip ... where one of the possible "flags" settings is: ATA_FLAG_PMP = (1 << 19), /* controller supports PMP */ so, having only glanced at all of this so far, i assume i would have one "struct ata_port" with that flag set. on the other hand, it would appear i would have four "struct ata_link"s: struct ata_link { struct ata_port *ap; int pmp; /* port multiplier port # */ ... snip ... finally, what about "struct ata_device": struct ata_device { struct ata_link *link; unsigned int devno; /* 0 or 1 */ ... snip ... not having read any further just yet, it's not clear whether there would be one or four of those, depending on how libata views the storage devices themselves, but i'm sure i'll figure it out as i keep reading. in the end, it would be terrifically useful if there was a simple example of how all these structures connected to support a basic scenario. is that written down anywhere? it doesn't need to be crazy comprehensive, just the enumeration of the structures involved. thoughts? rday -- ======================================================================== Robert P. J. Day Ottawa, Ontario, CANADA http://crashcourse.ca Twitter: http://twitter.com/rpjday LinkedIn: http://ca.linkedin.com/in/rpjday ========================================================================

1 0

Need advice about do we need this if condition
by Саша Ковтуненко April 17, 2019

April 17, 2019

hello I am trying to fix https://bugzilla.kernel.org/show_bug.cgi?id=202935 I found cause https://github.com/torvalds/linux/blob/master/fs/cifs/cifsfs.c#L1076 not sure is it correct to remove such condition ? Thanks AK

3 2