Message ID | 20241018054728.116519-1-kanie@linux.alibaba.com (mailing list archive) |
---|---|
State | Changes Requested |
Headers | show |
Series | PCI: optimize proc sequential file read | expand |
On Fri, Oct 18, 2024 at 01:47:28PM +0800, Guixin Liu wrote: > PCI driver will traverse pci device list in pci_seq_start in every > sequential file reading, use xarry to store all pci devices to > accelerate finding the start. > > Use "time cat /proc/bus/pci/devices" to test on a machine with 13k > pci devices, get an increase of about 90%. s/pci/PCI/ (several times) s/pci_seq_start/pci_seq_start()/ s/xarry/xarray/ s/, /, / (remove extra space) > Without this patch: > real 0m0.917s > user 0m0.000s > sys 0m0.913s > With this patch: > real 0m0.093s > user 0m0.000s > sys 0m0.093s Nice speedup, for sure! > Signed-off-by: Guixin Liu <kanie@linux.alibaba.com> > --- > drivers/pci/pci.h | 3 +++ > drivers/pci/probe.c | 1 + > drivers/pci/proc.c | 64 +++++++++++++++++++++++++++++++++++++++----- > drivers/pci/remove.c | 1 + > include/linux/pci.h | 2 ++ > 5 files changed, 64 insertions(+), 7 deletions(-) > > diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h > index 14d00ce45bfa..1a7da91eeb80 100644 > --- a/drivers/pci/pci.h > +++ b/drivers/pci/pci.h > @@ -962,4 +962,7 @@ void pcim_release_region(struct pci_dev *pdev, int bar); > (PCI_CONF1_ADDRESS(bus, dev, func, reg) | \ > PCI_CONF1_EXT_REG(reg)) > > +void pci_seq_tree_add_dev(struct pci_dev *dev); > +void pci_seq_tree_remove_dev(struct pci_dev *dev); > + > #endif /* DRIVERS_PCI_H */ > diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c > index 4f68414c3086..1fd9e9022f70 100644 > --- a/drivers/pci/probe.c > +++ b/drivers/pci/probe.c > @@ -2592,6 +2592,7 @@ void pci_device_add(struct pci_dev *dev, struct pci_bus *bus) > WARN_ON(ret < 0); > > pci_npem_create(dev); > + pci_seq_tree_add_dev(dev); > } > > struct pci_dev *pci_scan_single_device(struct pci_bus *bus, int devfn) > diff --git a/drivers/pci/proc.c b/drivers/pci/proc.c > index f967709082d6..30ca071ccad5 100644 > --- a/drivers/pci/proc.c > +++ b/drivers/pci/proc.c > @@ -19,6 +19,9 @@ > > static int proc_initialized; /* = 0 */ > > +DEFINE_XARRAY_FLAGS(pci_seq_tree, 0); > +static unsigned long pci_max_idx; > + > static loff_t proc_bus_pci_lseek(struct file *file, loff_t off, int whence) > { > struct pci_dev *dev = pde_data(file_inode(file)); > @@ -334,25 +337,72 @@ static const struct proc_ops proc_bus_pci_ops = { > #endif /* HAVE_PCI_MMAP */ > }; > > +void pci_seq_tree_add_dev(struct pci_dev *dev) > +{ > + int ret; > + > + if (dev) { I don't think we should test "dev" for NULL here. If it's NULL, I think we have bigger problems and we should oops. > + xa_lock(&pci_seq_tree); > + pci_dev_get(dev); > + ret = __xa_insert(&pci_seq_tree, pci_max_idx, dev, GFP_KERNEL); > + if (!ret) { > + dev->proc_seq_idx = pci_max_idx; > + pci_max_idx++; > + } else { > + pci_dev_put(dev); > + WARN_ON(ret); > + } > + xa_unlock(&pci_seq_tree); > + } > +} > + > +void pci_seq_tree_remove_dev(struct pci_dev *dev) > +{ > + unsigned long idx = dev->proc_seq_idx; > + struct pci_dev *latest_dev = NULL; > + struct pci_dev *ret; > + > + if (!dev) > + return; Same comment about testing "dev" for NULL. > + xa_lock(&pci_seq_tree); > + __xa_erase(&pci_seq_tree, idx); > + pci_dev_put(dev); > + /* > + * Move the latest pci_dev to this idx to keep the continuity. > + */ > + if (idx != pci_max_idx - 1) { > + latest_dev = __xa_erase(&pci_seq_tree, pci_max_idx - 1); > + ret = __xa_cmpxchg(&pci_seq_tree, idx, NULL, latest_dev, > + GFP_KERNEL); > + if (!ret) > + latest_dev->proc_seq_idx = idx; > + WARN_ON(ret); > + } > + pci_max_idx--; > + xa_unlock(&pci_seq_tree); > +} > + > /* iterator */ > static void *pci_seq_start(struct seq_file *m, loff_t *pos) > { > - struct pci_dev *dev = NULL; > + struct pci_dev *dev; > loff_t n = *pos; > > - for_each_pci_dev(dev) { > - if (!n--) > - break; > - } > + dev = xa_load(&pci_seq_tree, n); > + if (dev) > + pci_dev_get(dev); > return dev; I'm a little hesitant to add another place that keeps track of every PCI device. It's a fair bit of code here, and it's redundant information, which means more work to keep them all synchronized. This proc interface feels inherently racy. We keep track of the current item (n) in a struct seq_file, but I don't think there's anything to prevent a pci_dev hot-add or -remove between calls to pci_seq_start(). Is the proc interface the only place to get this information? If there's a way to get it from sysfs, maybe that is better and we don't need to spend effort optimizing the less-desirable path? > } > > static void *pci_seq_next(struct seq_file *m, void *v, loff_t *pos) > { > - struct pci_dev *dev = v; > + struct pci_dev *dev; > > (*pos)++; > - dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev); > + dev = xa_load(&pci_seq_tree, *pos); > + if (dev) > + pci_dev_get(dev); Where is the pci_dev_put() that corresponds with this new pci_dev_get()? > return dev; > } > > diff --git a/drivers/pci/remove.c b/drivers/pci/remove.c > index e4ce1145aa3e..257ea46233a3 100644 > --- a/drivers/pci/remove.c > +++ b/drivers/pci/remove.c > @@ -53,6 +53,7 @@ static void pci_destroy_dev(struct pci_dev *dev) > pci_npem_remove(dev); > > device_del(&dev->dev); > + pci_seq_tree_remove_dev(dev); > > down_write(&pci_bus_sem); > list_del(&dev->bus_list); > diff --git a/include/linux/pci.h b/include/linux/pci.h > index 573b4c4c2be6..aeb3d4cce06a 100644 > --- a/include/linux/pci.h > +++ b/include/linux/pci.h > @@ -534,6 +534,8 @@ struct pci_dev { > > /* These methods index pci_reset_fn_methods[] */ > u8 reset_methods[PCI_NUM_RESET_METHODS]; /* In priority order */ > + > + unsigned long long proc_seq_idx; > }; > > static inline struct pci_dev *pci_physfn(struct pci_dev *dev) > -- > 2.43.0 >
Hi Guixin, kernel test robot noticed the following build errors: [auto build test ERROR on pci/next] [also build test ERROR on pci/for-linus linus/master v6.12-rc3 next-20241018] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in https://git-scm.com/docs/git-format-patch#_base_tree_information] url: https://github.com/intel-lab-lkp/linux/commits/Guixin-Liu/PCI-optimize-proc-sequential-file-read/20241018-135026 base: https://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git next patch link: https://lore.kernel.org/r/20241018054728.116519-1-kanie%40linux.alibaba.com patch subject: [PATCH] PCI: optimize proc sequential file read config: i386-buildonly-randconfig-003-20241019 (https://download.01.org/0day-ci/archive/20241019/202410191439.yQ27wvB6-lkp@intel.com/config) compiler: gcc-12 (Debian 12.2.0-14) 12.2.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20241019/202410191439.yQ27wvB6-lkp@intel.com/reproduce) If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <lkp@intel.com> | Closes: https://lore.kernel.org/oe-kbuild-all/202410191439.yQ27wvB6-lkp@intel.com/ All errors (new ones prefixed by >>): ld: drivers/pci/probe.o: in function `pci_device_add': >> drivers/pci/probe.c:2595: undefined reference to `pci_seq_tree_add_dev' ld: drivers/pci/remove.o: in function `pci_destroy_dev': >> drivers/pci/remove.c:56: undefined reference to `pci_seq_tree_remove_dev' vim +2595 drivers/pci/probe.c 2546 2547 void pci_device_add(struct pci_dev *dev, struct pci_bus *bus) 2548 { 2549 int ret; 2550 2551 pci_configure_device(dev); 2552 2553 device_initialize(&dev->dev); 2554 dev->dev.release = pci_release_dev; 2555 2556 set_dev_node(&dev->dev, pcibus_to_node(bus)); 2557 dev->dev.dma_mask = &dev->dma_mask; 2558 dev->dev.dma_parms = &dev->dma_parms; 2559 dev->dev.coherent_dma_mask = 0xffffffffull; 2560 2561 dma_set_max_seg_size(&dev->dev, 65536); 2562 dma_set_seg_boundary(&dev->dev, 0xffffffff); 2563 2564 pcie_failed_link_retrain(dev); 2565 2566 /* Fix up broken headers */ 2567 pci_fixup_device(pci_fixup_header, dev); 2568 2569 pci_reassigndev_resource_alignment(dev); 2570 2571 dev->state_saved = false; 2572 2573 pci_init_capabilities(dev); 2574 2575 /* 2576 * Add the device to our list of discovered devices 2577 * and the bus list for fixup functions, etc. 2578 */ 2579 down_write(&pci_bus_sem); 2580 list_add_tail(&dev->bus_list, &bus->devices); 2581 up_write(&pci_bus_sem); 2582 2583 ret = pcibios_device_add(dev); 2584 WARN_ON(ret < 0); 2585 2586 /* Set up MSI IRQ domain */ 2587 pci_set_msi_domain(dev); 2588 2589 /* Notifier could use PCI capabilities */ 2590 dev->match_driver = false; 2591 ret = device_add(&dev->dev); 2592 WARN_ON(ret < 0); 2593 2594 pci_npem_create(dev); > 2595 pci_seq_tree_add_dev(dev); 2596 } 2597
Hi Guixin, kernel test robot noticed the following build errors: [auto build test ERROR on pci/next] [also build test ERROR on pci/for-linus linus/master v6.12-rc3 next-20241018] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in https://git-scm.com/docs/git-format-patch#_base_tree_information] url: https://github.com/intel-lab-lkp/linux/commits/Guixin-Liu/PCI-optimize-proc-sequential-file-read/20241018-135026 base: https://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git next patch link: https://lore.kernel.org/r/20241018054728.116519-1-kanie%40linux.alibaba.com patch subject: [PATCH] PCI: optimize proc sequential file read config: loongarch-randconfig-r062-20241019 (https://download.01.org/0day-ci/archive/20241020/202410200211.RFdwjJSq-lkp@intel.com/config) compiler: loongarch64-linux-gcc (GCC) 14.1.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20241020/202410200211.RFdwjJSq-lkp@intel.com/reproduce) If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <lkp@intel.com> | Closes: https://lore.kernel.org/oe-kbuild-all/202410200211.RFdwjJSq-lkp@intel.com/ All errors (new ones prefixed by >>): loongarch64-linux-ld: drivers/pci/probe.o: in function `.L5491': >> probe.c:(.text+0x2ff8): undefined reference to `pci_seq_tree_add_dev' loongarch64-linux-ld: drivers/pci/remove.o: in function `.LBB262': >> drivers/pci/remove.c:56:(.text+0x398): undefined reference to `pci_seq_tree_remove_dev'
Hi Guixin, kernel test robot noticed the following build warnings: [auto build test WARNING on pci/next] [also build test WARNING on pci/for-linus linus/master v6.12-rc3 next-20241018] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in https://git-scm.com/docs/git-format-patch#_base_tree_information] url: https://github.com/intel-lab-lkp/linux/commits/Guixin-Liu/PCI-optimize-proc-sequential-file-read/20241018-135026 base: https://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git next patch link: https://lore.kernel.org/r/20241018054728.116519-1-kanie%40linux.alibaba.com patch subject: [PATCH] PCI: optimize proc sequential file read config: x86_64-randconfig-122-20241019 (https://download.01.org/0day-ci/archive/20241020/202410201436.9i93qAYF-lkp@intel.com/config) compiler: gcc-12 (Debian 12.2.0-14) 12.2.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20241020/202410201436.9i93qAYF-lkp@intel.com/reproduce) If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <lkp@intel.com> | Closes: https://lore.kernel.org/oe-kbuild-all/202410201436.9i93qAYF-lkp@intel.com/ sparse warnings: (new ones prefixed by >>) >> drivers/pci/proc.c:22:1: sparse: sparse: symbol 'pci_seq_tree' was not declared. Should it be static? vim +/pci_seq_tree +22 drivers/pci/proc.c 21 > 22 DEFINE_XARRAY_FLAGS(pci_seq_tree, 0); 23 static unsigned long pci_max_idx; 24
在 2024/10/19 06:22, Bjorn Helgaas 写道: > On Fri, Oct 18, 2024 at 01:47:28PM +0800, Guixin Liu wrote: >> PCI driver will traverse pci device list in pci_seq_start in every >> sequential file reading, use xarry to store all pci devices to >> accelerate finding the start. >> >> Use "time cat /proc/bus/pci/devices" to test on a machine with 13k >> pci devices, get an increase of about 90%. > s/pci/PCI/ (several times) > s/pci_seq_start/pci_seq_start()/ > s/xarry/xarray/ > s/, /, / (remove extra space) > OK. >> Without this patch: >> real 0m0.917s >> user 0m0.000s >> sys 0m0.913s >> With this patch: >> real 0m0.093s >> user 0m0.000s >> sys 0m0.093s > Nice speedup, for sure! > >> Signed-off-by: Guixin Liu <kanie@linux.alibaba.com> >> --- >> drivers/pci/pci.h | 3 +++ >> drivers/pci/probe.c | 1 + >> drivers/pci/proc.c | 64 +++++++++++++++++++++++++++++++++++++++----- >> drivers/pci/remove.c | 1 + >> include/linux/pci.h | 2 ++ >> 5 files changed, 64 insertions(+), 7 deletions(-) >> >> diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h >> index 14d00ce45bfa..1a7da91eeb80 100644 >> --- a/drivers/pci/pci.h >> +++ b/drivers/pci/pci.h >> @@ -962,4 +962,7 @@ void pcim_release_region(struct pci_dev *pdev, int bar); >> (PCI_CONF1_ADDRESS(bus, dev, func, reg) | \ >> PCI_CONF1_EXT_REG(reg)) >> >> +void pci_seq_tree_add_dev(struct pci_dev *dev); >> +void pci_seq_tree_remove_dev(struct pci_dev *dev); >> + >> #endif /* DRIVERS_PCI_H */ >> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c >> index 4f68414c3086..1fd9e9022f70 100644 >> --- a/drivers/pci/probe.c >> +++ b/drivers/pci/probe.c >> @@ -2592,6 +2592,7 @@ void pci_device_add(struct pci_dev *dev, struct pci_bus *bus) >> WARN_ON(ret < 0); >> >> pci_npem_create(dev); >> + pci_seq_tree_add_dev(dev); >> } >> >> struct pci_dev *pci_scan_single_device(struct pci_bus *bus, int devfn) >> diff --git a/drivers/pci/proc.c b/drivers/pci/proc.c >> index f967709082d6..30ca071ccad5 100644 >> --- a/drivers/pci/proc.c >> +++ b/drivers/pci/proc.c >> @@ -19,6 +19,9 @@ >> >> static int proc_initialized; /* = 0 */ >> >> +DEFINE_XARRAY_FLAGS(pci_seq_tree, 0); >> +static unsigned long pci_max_idx; >> + >> static loff_t proc_bus_pci_lseek(struct file *file, loff_t off, int whence) >> { >> struct pci_dev *dev = pde_data(file_inode(file)); >> @@ -334,25 +337,72 @@ static const struct proc_ops proc_bus_pci_ops = { >> #endif /* HAVE_PCI_MMAP */ >> }; >> >> +void pci_seq_tree_add_dev(struct pci_dev *dev) >> +{ >> + int ret; >> + >> + if (dev) { > I don't think we should test "dev" for NULL here. If it's NULL, I > think we have bigger problems and we should oops. Sure. >> + xa_lock(&pci_seq_tree); >> + pci_dev_get(dev); >> + ret = __xa_insert(&pci_seq_tree, pci_max_idx, dev, GFP_KERNEL); >> + if (!ret) { >> + dev->proc_seq_idx = pci_max_idx; >> + pci_max_idx++; >> + } else { >> + pci_dev_put(dev); >> + WARN_ON(ret); >> + } >> + xa_unlock(&pci_seq_tree); >> + } >> +} >> + >> +void pci_seq_tree_remove_dev(struct pci_dev *dev) >> +{ >> + unsigned long idx = dev->proc_seq_idx; >> + struct pci_dev *latest_dev = NULL; >> + struct pci_dev *ret; >> + >> + if (!dev) >> + return; > Same comment about testing "dev" for NULL. > Ok. >> + xa_lock(&pci_seq_tree); >> + __xa_erase(&pci_seq_tree, idx); >> + pci_dev_put(dev); >> + /* >> + * Move the latest pci_dev to this idx to keep the continuity. >> + */ >> + if (idx != pci_max_idx - 1) { >> + latest_dev = __xa_erase(&pci_seq_tree, pci_max_idx - 1); >> + ret = __xa_cmpxchg(&pci_seq_tree, idx, NULL, latest_dev, >> + GFP_KERNEL); >> + if (!ret) >> + latest_dev->proc_seq_idx = idx; >> + WARN_ON(ret); >> + } >> + pci_max_idx--; >> + xa_unlock(&pci_seq_tree); >> +} >> + >> /* iterator */ >> static void *pci_seq_start(struct seq_file *m, loff_t *pos) >> { >> - struct pci_dev *dev = NULL; >> + struct pci_dev *dev; >> loff_t n = *pos; >> >> - for_each_pci_dev(dev) { >> - if (!n--) >> - break; >> - } >> + dev = xa_load(&pci_seq_tree, n); >> + if (dev) >> + pci_dev_get(dev); >> return dev; > I'm a little hesitant to add another place that keeps track of every > PCI device. It's a fair bit of code here, and it's redundant > information, which means more work to keep them all synchronized. > > This proc interface feels inherently racy. We keep track of the > current item (n) in a struct seq_file, but I don't think there's > anything to prevent a pci_dev hot-add or -remove between calls to > pci_seq_start(). Yes, maybe lost some information this time. > > Is the proc interface the only place to get this information? If > there's a way to get it from sysfs, maybe that is better and we don't > need to spend effort optimizing the less-desirable path? This is the situation I encountered: in scenarios of rapid container scaling, when a container is started, it executes lscpu to traverse the /proc/bus/pci/devices file, or the container process directly traverses this file. When many containers are being started at once, it causes numerous containers to wait due to the locks on the klist used for traversing pci_dev, which greatly reduces the efficiency of container scaling and also causes other CPUs to become unresponsive. User-space programs, including Docker, are clients that we cannot easily modify. Therefore, I attempted to accelerate pci_seq_start() within the kernel. This indeed resulted in the need for more code to maintain, as we must ensure both fast access and ordering. Initially, I considered directly modifying the klist in the driver base module, but such changes would impact other drivers as well. Do you have any other good suggestions? Best Regards, Guixin liu >> } >> >> static void *pci_seq_next(struct seq_file *m, void *v, loff_t *pos) >> { >> - struct pci_dev *dev = v; >> + struct pci_dev *dev; >> >> (*pos)++; >> - dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev); >> + dev = xa_load(&pci_seq_tree, *pos); >> + if (dev) >> + pci_dev_get(dev); > Where is the pci_dev_put() that corresponds with this new > pci_dev_get()? In pci_seq_stop(), will call the pci_dev_put(). > >> return dev; >> } >> >> diff --git a/drivers/pci/remove.c b/drivers/pci/remove.c >> index e4ce1145aa3e..257ea46233a3 100644 >> --- a/drivers/pci/remove.c >> +++ b/drivers/pci/remove.c >> @@ -53,6 +53,7 @@ static void pci_destroy_dev(struct pci_dev *dev) >> pci_npem_remove(dev); >> >> device_del(&dev->dev); >> + pci_seq_tree_remove_dev(dev); >> >> down_write(&pci_bus_sem); >> list_del(&dev->bus_list); >> diff --git a/include/linux/pci.h b/include/linux/pci.h >> index 573b4c4c2be6..aeb3d4cce06a 100644 >> --- a/include/linux/pci.h >> +++ b/include/linux/pci.h >> @@ -534,6 +534,8 @@ struct pci_dev { >> >> /* These methods index pci_reset_fn_methods[] */ >> u8 reset_methods[PCI_NUM_RESET_METHODS]; /* In priority order */ >> + >> + unsigned long long proc_seq_idx; >> }; >> >> static inline struct pci_dev *pci_physfn(struct pci_dev *dev) >> -- >> 2.43.0 >>
Hi Guixin, kernel test robot noticed the following build warnings: https://git-scm.com/docs/git-format-patch#_base_tree_information] url: https://github.com/intel-lab-lkp/linux/commits/Guixin-Liu/PCI-optimize-proc-sequential-file-read/20241018-135026 base: https://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git next patch link: https://lore.kernel.org/r/20241018054728.116519-1-kanie%40linux.alibaba.com patch subject: [PATCH] PCI: optimize proc sequential file read config: i386-randconfig-141-20241019 (https://download.01.org/0day-ci/archive/20241019/202410190659.9MCI8EyL-lkp@intel.com/config) compiler: gcc-12 (Debian 12.2.0-14) 12.2.0 If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <lkp@intel.com> | Reported-by: Dan Carpenter <dan.carpenter@linaro.org> | Closes: https://lore.kernel.org/r/202410190659.9MCI8EyL-lkp@intel.com/ smatch warnings: drivers/pci/proc.c:365 pci_seq_tree_remove_dev() warn: variable dereferenced before check 'dev' (see line 361) vim +/dev +365 drivers/pci/proc.c 4ca256d9a0e58a Guixin Liu 2024-10-18 359 void pci_seq_tree_remove_dev(struct pci_dev *dev) 4ca256d9a0e58a Guixin Liu 2024-10-18 360 { 4ca256d9a0e58a Guixin Liu 2024-10-18 @361 unsigned long idx = dev->proc_seq_idx; ^^^^^^^^^^^^^^^^^ Dereferenced 4ca256d9a0e58a Guixin Liu 2024-10-18 362 struct pci_dev *latest_dev = NULL; 4ca256d9a0e58a Guixin Liu 2024-10-18 363 struct pci_dev *ret; 4ca256d9a0e58a Guixin Liu 2024-10-18 364 4ca256d9a0e58a Guixin Liu 2024-10-18 @365 if (!dev) ^^^^ Checked too late 4ca256d9a0e58a Guixin Liu 2024-10-18 366 return; 4ca256d9a0e58a Guixin Liu 2024-10-18 367 4ca256d9a0e58a Guixin Liu 2024-10-18 368 xa_lock(&pci_seq_tree); 4ca256d9a0e58a Guixin Liu 2024-10-18 369 __xa_erase(&pci_seq_tree, idx); 4ca256d9a0e58a Guixin Liu 2024-10-18 370 pci_dev_put(dev); 4ca256d9a0e58a Guixin Liu 2024-10-18 371 /* 4ca256d9a0e58a Guixin Liu 2024-10-18 372 * Move the latest pci_dev to this idx to keep the continuity. 4ca256d9a0e58a Guixin Liu 2024-10-18 373 */ 4ca256d9a0e58a Guixin Liu 2024-10-18 374 if (idx != pci_max_idx - 1) { 4ca256d9a0e58a Guixin Liu 2024-10-18 375 latest_dev = __xa_erase(&pci_seq_tree, pci_max_idx - 1); 4ca256d9a0e58a Guixin Liu 2024-10-18 376 ret = __xa_cmpxchg(&pci_seq_tree, idx, NULL, latest_dev, 4ca256d9a0e58a Guixin Liu 2024-10-18 377 GFP_KERNEL); 4ca256d9a0e58a Guixin Liu 2024-10-18 378 if (!ret) 4ca256d9a0e58a Guixin Liu 2024-10-18 379 latest_dev->proc_seq_idx = idx; 4ca256d9a0e58a Guixin Liu 2024-10-18 380 WARN_ON(ret); 4ca256d9a0e58a Guixin Liu 2024-10-18 381 } 4ca256d9a0e58a Guixin Liu 2024-10-18 382 pci_max_idx--; 4ca256d9a0e58a Guixin Liu 2024-10-18 383 xa_unlock(&pci_seq_tree); 4ca256d9a0e58a Guixin Liu 2024-10-18 384 }
On Mon, Oct 21, 2024 at 10:04:03AM +0800, Guixin Liu wrote: > > This proc interface feels inherently racy. We keep track of the > > current item (n) in a struct seq_file, but I don't think there's > > anything to prevent a pci_dev hot-add or -remove between calls to > > pci_seq_start(). > Yes, maybe lost some information this time. > > > > Is the proc interface the only place to get this information? If > > there's a way to get it from sysfs, maybe that is better and we don't > > need to spend effort optimizing the less-desirable path? > > This is the situation I encountered: in scenarios of rapid container > > scaling, when a container is started, it executes lscpu to traverse > > the /proc/bus/pci/devices file, or the container process directly > > traverses this file. When many containers are being started at once, > > it causes numerous containers to wait due to the locks on the klist > > used for traversing pci_dev, which greatly reduces the efficiency of > > container scaling and also causes other CPUs to become unresponsive. > > > User-space programs, including Docker, are clients that we cannot easily > modify. > > Therefore, I attempted to accelerate pci_seq_start() within the kernel. > > This indeed resulted in the need for more code to maintain, as we must > > ensure both fast access and ordering. Initially, I considered directly > > modifying the klist in the driver base module, but such changes would > > impact other drivers as well. I am not opposed to any driver core changes where we iterate over lists of objects on a bus, but usually that's not a real "hot path" that matters. Also, you need to keep the lists in a specific order, which I do not think an xarray will do very well (or at least not the last time I looked at it, I could be wrong.) I understand your need to want to make 'lspci' faster, but by default, 'lspci' does not access the proc files, I thought it went through sysfs. Why not just fix the userspace tool instead if that's the real issue? Just because you can modify the kernel, doesn't mean you should, often the better solution is to fix userspace from doing something dumb, and if it is doing something dumb, then let it and have it deal with the consequences :) thanks, greg k-h
在 2024/10/21 15:17, Dan Carpenter 写道: > Hi Guixin, > > kernel test robot noticed the following build warnings: > > https://git-scm.com/docs/git-format-patch#_base_tree_information] > > url: https://github.com/intel-lab-lkp/linux/commits/Guixin-Liu/PCI-optimize-proc-sequential-file-read/20241018-135026 > base: https://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git next > patch link: https://lore.kernel.org/r/20241018054728.116519-1-kanie%40linux.alibaba.com > patch subject: [PATCH] PCI: optimize proc sequential file read > config: i386-randconfig-141-20241019 (https://download.01.org/0day-ci/archive/20241019/202410190659.9MCI8EyL-lkp@intel.com/config) > compiler: gcc-12 (Debian 12.2.0-14) 12.2.0 > > If you fix the issue in a separate patch/commit (i.e. not just a new version of > the same patch/commit), kindly add following tags > | Reported-by: kernel test robot <lkp@intel.com> > | Reported-by: Dan Carpenter <dan.carpenter@linaro.org> > | Closes: https://lore.kernel.org/r/202410190659.9MCI8EyL-lkp@intel.com/ > > smatch warnings: > drivers/pci/proc.c:365 pci_seq_tree_remove_dev() warn: variable dereferenced before check 'dev' (see line 361) > > vim +/dev +365 drivers/pci/proc.c > > 4ca256d9a0e58a Guixin Liu 2024-10-18 359 void pci_seq_tree_remove_dev(struct pci_dev *dev) > 4ca256d9a0e58a Guixin Liu 2024-10-18 360 { > 4ca256d9a0e58a Guixin Liu 2024-10-18 @361 unsigned long idx = dev->proc_seq_idx; > ^^^^^^^^^^^^^^^^^ > Dereferenced Thanks, changed in v2. Best Regards, Guixin Liu > 4ca256d9a0e58a Guixin Liu 2024-10-18 362 struct pci_dev *latest_dev = NULL; > 4ca256d9a0e58a Guixin Liu 2024-10-18 363 struct pci_dev *ret; > 4ca256d9a0e58a Guixin Liu 2024-10-18 364 > 4ca256d9a0e58a Guixin Liu 2024-10-18 @365 if (!dev) > ^^^^ > Checked too late > > 4ca256d9a0e58a Guixin Liu 2024-10-18 366 return; > 4ca256d9a0e58a Guixin Liu 2024-10-18 367 > 4ca256d9a0e58a Guixin Liu 2024-10-18 368 xa_lock(&pci_seq_tree); > 4ca256d9a0e58a Guixin Liu 2024-10-18 369 __xa_erase(&pci_seq_tree, idx); > 4ca256d9a0e58a Guixin Liu 2024-10-18 370 pci_dev_put(dev); > 4ca256d9a0e58a Guixin Liu 2024-10-18 371 /* > 4ca256d9a0e58a Guixin Liu 2024-10-18 372 * Move the latest pci_dev to this idx to keep the continuity. > 4ca256d9a0e58a Guixin Liu 2024-10-18 373 */ > 4ca256d9a0e58a Guixin Liu 2024-10-18 374 if (idx != pci_max_idx - 1) { > 4ca256d9a0e58a Guixin Liu 2024-10-18 375 latest_dev = __xa_erase(&pci_seq_tree, pci_max_idx - 1); > 4ca256d9a0e58a Guixin Liu 2024-10-18 376 ret = __xa_cmpxchg(&pci_seq_tree, idx, NULL, latest_dev, > 4ca256d9a0e58a Guixin Liu 2024-10-18 377 GFP_KERNEL); > 4ca256d9a0e58a Guixin Liu 2024-10-18 378 if (!ret) > 4ca256d9a0e58a Guixin Liu 2024-10-18 379 latest_dev->proc_seq_idx = idx; > 4ca256d9a0e58a Guixin Liu 2024-10-18 380 WARN_ON(ret); > 4ca256d9a0e58a Guixin Liu 2024-10-18 381 } > 4ca256d9a0e58a Guixin Liu 2024-10-18 382 pci_max_idx--; > 4ca256d9a0e58a Guixin Liu 2024-10-18 383 xa_unlock(&pci_seq_tree); > 4ca256d9a0e58a Guixin Liu 2024-10-18 384 } >
在 2024/10/21 19:04, Greg KH 写道: > On Mon, Oct 21, 2024 at 10:04:03AM +0800, Guixin Liu wrote: >>> This proc interface feels inherently racy. We keep track of the >>> current item (n) in a struct seq_file, but I don't think there's >>> anything to prevent a pci_dev hot-add or -remove between calls to >>> pci_seq_start(). >> Yes, maybe lost some information this time. >>> Is the proc interface the only place to get this information? If >>> there's a way to get it from sysfs, maybe that is better and we don't >>> need to spend effort optimizing the less-desirable path? >> This is the situation I encountered: in scenarios of rapid container >> >> scaling, when a container is started, it executes lscpu to traverse >> >> the /proc/bus/pci/devices file, or the container process directly >> >> traverses this file. When many containers are being started at once, >> >> it causes numerous containers to wait due to the locks on the klist >> >> used for traversing pci_dev, which greatly reduces the efficiency of >> >> container scaling and also causes other CPUs to become unresponsive. >> >> >> User-space programs, including Docker, are clients that we cannot easily >> modify. >> >> Therefore, I attempted to accelerate pci_seq_start() within the kernel. >> >> This indeed resulted in the need for more code to maintain, as we must >> >> ensure both fast access and ordering. Initially, I considered directly >> >> modifying the klist in the driver base module, but such changes would >> >> impact other drivers as well. > I am not opposed to any driver core changes where we iterate over lists > of objects on a bus, but usually that's not a real "hot path" that > matters. Also, you need to keep the lists in a specific order, which I > do not think an xarray will do very well (or at least not the last time > I looked at it, I could be wrong.) > > I understand your need to want to make 'lspci' faster, but by default, > 'lspci' does not access the proc files, I thought it went through sysfs. > Why not just fix the userspace tool instead if that's the real issue? > > Just because you can modify the kernel, doesn't mean you should, often > the better solution is to fix userspace from doing something dumb, and > if it is doing something dumb, then let it and have it deal with the > consequences :) > > thanks, > > greg k-h Actually, lscpu may access PCI proc files and could potentially iterate three times. You can see this in the lscpu source code within the function lscpu_read_virtualization()->has_pci_device(). We've previously optimized it to combine those three iterations into one, as reflected in this PR: https://github.com/util-linux/util-linux/pull/3177. Additionally, many of our clients’ tools and programs also access PCI proc files, but we are not allowed to modify that part; they do not provide us with the source code. Clients tend to perceive that the kernel performance we provide is subpar(That's why we have stopped giving 'lscpu' direct access to sysfs). We have already implemented such optimizations in our operating system and achieved significant performance benefits. If this patch can be accepted into the community, it would be great for more people to benefit from it. Best Regards, Guixin Liu
[+cc Greg] On Fri, Oct 18, 2024 at 01:47:28PM +0800, Guixin Liu wrote: > PCI driver will traverse pci device list in pci_seq_start in every > sequential file reading, use xarry to store all pci devices to > accelerate finding the start. > /* iterator */ > static void *pci_seq_start(struct seq_file *m, loff_t *pos) > { > - struct pci_dev *dev = NULL; > + struct pci_dev *dev; > loff_t n = *pos; > > - for_each_pci_dev(dev) { > - if (!n--) > - break; > - } Maybe another approach would be to use pci_get_device() directly instead of for_each_pci_dev(). pci_get_device() takes a "from" starting point, so instead of keeping track of the index "n", you could keep track of the current pci_dev.
在 2024/10/22 23:44, Bjorn Helgaas 写道: > [+cc Greg] > > On Fri, Oct 18, 2024 at 01:47:28PM +0800, Guixin Liu wrote: >> PCI driver will traverse pci device list in pci_seq_start in every >> sequential file reading, use xarry to store all pci devices to >> accelerate finding the start. >> /* iterator */ >> static void *pci_seq_start(struct seq_file *m, loff_t *pos) >> { >> - struct pci_dev *dev = NULL; >> + struct pci_dev *dev; >> loff_t n = *pos; >> >> - for_each_pci_dev(dev) { >> - if (!n--) >> - break; >> - } > Maybe another approach would be to use pci_get_device() directly > instead of for_each_pci_dev(). > > pci_get_device() takes a "from" starting point, so instead of keeping > track of the index "n", you could keep track of the current pci_dev. You mean let struct seq_file to keep the pci_dev? Well, we also need keep the pci_dev which held by seq_file in klist_devices, because finding next rely on it, this make things complicated. In addition, in pci_seq_start(), the pos may not increased one by one, so we dont know how many times we should skip to find the pci_dev which number is *pos. Best Regards, Guixin Liu
diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h index 14d00ce45bfa..1a7da91eeb80 100644 --- a/drivers/pci/pci.h +++ b/drivers/pci/pci.h @@ -962,4 +962,7 @@ void pcim_release_region(struct pci_dev *pdev, int bar); (PCI_CONF1_ADDRESS(bus, dev, func, reg) | \ PCI_CONF1_EXT_REG(reg)) +void pci_seq_tree_add_dev(struct pci_dev *dev); +void pci_seq_tree_remove_dev(struct pci_dev *dev); + #endif /* DRIVERS_PCI_H */ diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c index 4f68414c3086..1fd9e9022f70 100644 --- a/drivers/pci/probe.c +++ b/drivers/pci/probe.c @@ -2592,6 +2592,7 @@ void pci_device_add(struct pci_dev *dev, struct pci_bus *bus) WARN_ON(ret < 0); pci_npem_create(dev); + pci_seq_tree_add_dev(dev); } struct pci_dev *pci_scan_single_device(struct pci_bus *bus, int devfn) diff --git a/drivers/pci/proc.c b/drivers/pci/proc.c index f967709082d6..30ca071ccad5 100644 --- a/drivers/pci/proc.c +++ b/drivers/pci/proc.c @@ -19,6 +19,9 @@ static int proc_initialized; /* = 0 */ +DEFINE_XARRAY_FLAGS(pci_seq_tree, 0); +static unsigned long pci_max_idx; + static loff_t proc_bus_pci_lseek(struct file *file, loff_t off, int whence) { struct pci_dev *dev = pde_data(file_inode(file)); @@ -334,25 +337,72 @@ static const struct proc_ops proc_bus_pci_ops = { #endif /* HAVE_PCI_MMAP */ }; +void pci_seq_tree_add_dev(struct pci_dev *dev) +{ + int ret; + + if (dev) { + xa_lock(&pci_seq_tree); + pci_dev_get(dev); + ret = __xa_insert(&pci_seq_tree, pci_max_idx, dev, GFP_KERNEL); + if (!ret) { + dev->proc_seq_idx = pci_max_idx; + pci_max_idx++; + } else { + pci_dev_put(dev); + WARN_ON(ret); + } + xa_unlock(&pci_seq_tree); + } +} + +void pci_seq_tree_remove_dev(struct pci_dev *dev) +{ + unsigned long idx = dev->proc_seq_idx; + struct pci_dev *latest_dev = NULL; + struct pci_dev *ret; + + if (!dev) + return; + + xa_lock(&pci_seq_tree); + __xa_erase(&pci_seq_tree, idx); + pci_dev_put(dev); + /* + * Move the latest pci_dev to this idx to keep the continuity. + */ + if (idx != pci_max_idx - 1) { + latest_dev = __xa_erase(&pci_seq_tree, pci_max_idx - 1); + ret = __xa_cmpxchg(&pci_seq_tree, idx, NULL, latest_dev, + GFP_KERNEL); + if (!ret) + latest_dev->proc_seq_idx = idx; + WARN_ON(ret); + } + pci_max_idx--; + xa_unlock(&pci_seq_tree); +} + /* iterator */ static void *pci_seq_start(struct seq_file *m, loff_t *pos) { - struct pci_dev *dev = NULL; + struct pci_dev *dev; loff_t n = *pos; - for_each_pci_dev(dev) { - if (!n--) - break; - } + dev = xa_load(&pci_seq_tree, n); + if (dev) + pci_dev_get(dev); return dev; } static void *pci_seq_next(struct seq_file *m, void *v, loff_t *pos) { - struct pci_dev *dev = v; + struct pci_dev *dev; (*pos)++; - dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev); + dev = xa_load(&pci_seq_tree, *pos); + if (dev) + pci_dev_get(dev); return dev; } diff --git a/drivers/pci/remove.c b/drivers/pci/remove.c index e4ce1145aa3e..257ea46233a3 100644 --- a/drivers/pci/remove.c +++ b/drivers/pci/remove.c @@ -53,6 +53,7 @@ static void pci_destroy_dev(struct pci_dev *dev) pci_npem_remove(dev); device_del(&dev->dev); + pci_seq_tree_remove_dev(dev); down_write(&pci_bus_sem); list_del(&dev->bus_list); diff --git a/include/linux/pci.h b/include/linux/pci.h index 573b4c4c2be6..aeb3d4cce06a 100644 --- a/include/linux/pci.h +++ b/include/linux/pci.h @@ -534,6 +534,8 @@ struct pci_dev { /* These methods index pci_reset_fn_methods[] */ u8 reset_methods[PCI_NUM_RESET_METHODS]; /* In priority order */ + + unsigned long long proc_seq_idx; }; static inline struct pci_dev *pci_physfn(struct pci_dev *dev)
PCI driver will traverse pci device list in pci_seq_start in every sequential file reading, use xarry to store all pci devices to accelerate finding the start. Use "time cat /proc/bus/pci/devices" to test on a machine with 13k pci devices, get an increase of about 90%. Without this patch: real 0m0.917s user 0m0.000s sys 0m0.913s With this patch: real 0m0.093s user 0m0.000s sys 0m0.093s Signed-off-by: Guixin Liu <kanie@linux.alibaba.com> --- drivers/pci/pci.h | 3 +++ drivers/pci/probe.c | 1 + drivers/pci/proc.c | 64 +++++++++++++++++++++++++++++++++++++++----- drivers/pci/remove.c | 1 + include/linux/pci.h | 2 ++ 5 files changed, 64 insertions(+), 7 deletions(-)