staging: gasket: sundry fixes and fixups
v3 cover staging: gasket: sundry fixes and fixups
From: Todd Poynor <[email protected]>
Various fixes mainly from the chromium review of the gasket and apex drivers. More to come.
Todd Poynor (32):
staging: gasket: remove X86 Kconfig restriction
staging: gasket: fix typo in apex_enter_reset
staging: gasket: fix typo in gasket_core.h comments
staging: gasket: whitespace fix in gasket_page_table_init
staging: gasket: remove driver registration on class creation failure
staging: gasket: hold mutex on gasket driver unregistration
staging: gasket: Return EBUSY on mapping create when already in use
staging: gasket: Remove stale pointers on error allocating attr array
staging: gasket: convert gasket_mmap_has_permissions to bool return
staging: gasket: fix gasket_wait_with_reschedule timeout return code
staging: gasket: gasket_wait_with_reschedule use msleep
staging: gasket: gasket_wait_with_reschedule simplify logic
staging: gasket: gasket_wait_with_reschedule use 32 bits of retry
count
staging: gasket: bail out of reset sequence on device callback error
staging: gasket: drop gasket_cdev_get_info, use container_of
staging: gasket: always allow root open for write
staging: gasket: annotate ioctl arg with __user
staging: gasket: gasket_enable_dev remove unnecessary variable
staging: gasket: remove code for no physical device
staging: gasket: fix class create bug handling
staging: gasket: remove unnecessary code in coherent allocator
staging: gasket: don't treat no device reset callback as an error
staging: gasket: gasket_mmap return error instead of valid BAR index
staging: gasket: apex_clock_gating simplify logic, reduce indentation
staging: gasket: apex_ioctl_check_permissions use bool return type
staging: gasket: gasket page table functions use bool return type
staging: gasket: remove else clause after return in if clause
staging: gasket: fix comment syntax in apex.h
staging: gasket: remove unnecessary parens in page table code
staging: gasket: gasket_mmap use PAGE_MASK
staging: gasket: remove extra parens in gasket_write_mappable_regions
staging: gasket: fix multi-line comment syntax in gasket_core.h
Changes from v2:
staging: gasket: Return EBUSY on mapping create when already in use
Update commit text for clarity: why EBUSY is better than EINVAL here.
staging: gasket: gasket_wait_with_reschedule use msleep
Update commit text for clarity: why msleep is better than schedule_timeout here.
staging: gasket: gasket_wait_with_reschedule return when condition hit
Renamed: staging: gasket: gasket_wait_with_reschedule simplify logic Update commit text for clarity: improved description of change. Remove extraneous condition check and a return statement now impossible to reach.
staging: gasket: always allow root open for write
Add change that accidentally went into the below patch in v2. Normalize file mode bitwise ANDs to zero or one values. Dropped Reviewed-by from Dmitry Torokhov since this has changed.
staging: gasket: annotate ioctl arg with __user
Remove stray change from v2, now moved into the above patch in v3.
staging: gasket: remove else clause after return in if clause
Don't break line that doesn't need it after indentation reduced.
staging: gasket: gasket_mmap use PAGE_MASK
Update commit text for clarity: why ~PAGE_MASK is better than PAGE_SIZE - 1 here.
staging: gasket: apex_clock_gating simplify logic, reduce indentation
Unbreak line that no longer needs a break after re-indent.
Unchanged from v2:
staging: gasket: remove X86 Kconfig restriction
staging: gasket: fix typo in apex_enter_reset
staging: gasket: fix typo in gasket_core.h comments
staging: gasket: whitespace fix in gasket_page_table_init
staging: gasket: remove driver registration on class creation failure
staging: gasket: hold mutex on gasket driver unregistration
staging: gasket: Remove stale pointers on error allocating attr array
staging: gasket: convert gasket_mmap_has_permissions to bool return
staging: gasket: fix gasket_wait_with_reschedule timeout return code
staging: gasket: gasket_wait_with_reschedule use 32 bits of retry
count
staging: gasket: bail out of reset sequence on device callback error
staging: gasket: drop gasket_cdev_get_info, use container_of
staging: gasket: gasket_enable_dev remove unnecessary variable
staging: gasket: remove code for no physical device
staging: gasket: fix class create bug handling
staging: gasket: remove unnecessary code in coherent allocator
staging: gasket: don't treat no device reset callback as an error
staging: gasket: gasket_mmap return error instead of valid BAR index
staging: gasket: apex_ioctl_check_permissions use bool return type
staging: gasket: gasket page table functions use bool return type
staging: gasket: fix comment syntax in apex.h
staging: gasket: remove unnecessary parens in page table code
staging: gasket: remove extra parens in gasket_write_mappable_regions
staging: gasket: fix multi-line comment syntax in gasket_core.h
drivers/staging/gasket/Kconfig | 2 +- drivers/staging/gasket/apex.h | 7 +- drivers/staging/gasket/apex_driver.c | 77 ++++++-------- drivers/staging/gasket/gasket_core.c | 115 ++++++++------------- drivers/staging/gasket/gasket_core.h | 11 +- drivers/staging/gasket/gasket_ioctl.c | 95 +++++++++-------- drivers/staging/gasket/gasket_ioctl.h | 4 +- drivers/staging/gasket/gasket_page_table.c | 66 ++++++------ drivers/staging/gasket/gasket_page_table.h | 8 +- drivers/staging/gasket/gasket_sysfs.c | 4 +- 10 files changed, 177 insertions(+), 212 deletions(-)
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 1/32 staging: gasket: remove X86 Kconfig restriction
From: Todd Poynor <[email protected]>
The gasket and apex drivers are to be used on other architectures besides X86.
--- drivers/staging/gasket/Kconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/gasket/Kconfig +1 -1
@@ -2,7 +2,7 @@ menu "Gasket devices" config STAGING_GASKET_FRAMEWORK tristate "Gasket framework" - depends on PCI && X86_64 + depends on PCI help This framework supports Gasket-compatible devices, such as Apex. It is required for any of the following module(s).
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 2/32 staging: gasket: fix typo in apex_enter_reset
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Fix typo in log message.
--- drivers/staging/gasket/apex_driver.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/gasket/apex_driver.c +1 -1
@@ -488,7 +488,7 @@ static int apex_enter_reset(struct gasket_dev *gasket_dev, uint type) APEX_BAR2_REG_USER_HIB_DMA_PAUSED, 1, 1, APEX_RESET_DELAY, APEX_RESET_RETRY)) {gasket_log_error (gasket_dev, -"DMAs did not quiece within timeout (%d ms)" , +"DMAs did not quiesce within timeout (%d ms)" ,APEX_RESET_RETRY *APEX_RESET_DELAY );return -EINVAL ; }
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 3/32 staging: gasket: fix typo in gasket_core.h comments
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Grammar fixup in gasket_core.h comments describing struct gasket_interrupt_desc.
--- drivers/staging/gasket/gasket_core.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/gasket/gasket_core.h +1 -1
@@ -56,7 +56,7 @@ enum gasket_interrupt_type {/* Used to describe a Gasket interrupt. Contains an interrupt index, a register, * and packing data for that interrupt. The register and packing data -* fields is relevant only for PCI_MSIX interrupt type and can be +* fields are relevant only for PCI_MSIX interrupt type and can be * set to 0 for everything else. */ struct gasket_interrupt_desc {
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 4/32 staging: gasket: whitespace fix in gasket_page_table_init
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Tab replaced with space.
--- drivers/staging/gasket/gasket_page_table.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/gasket/gasket_page_table.c +1 -1
@@ -347,7 +347,7 @@ int gasket_page_table_init( pg_tbl->config.mode == GASKET_PAGE_TABLE_MODE_SIMPLE) { pg_tbl->num_simple_entries = total_entries; pg_tbl->num_extended_entries = 0; - pg_tbl->extended_flag = 1ull << page_table_config->extended_bit; + pg_tbl->extended_flag = 1ull << page_table_config->extended_bit; } else { pg_tbl->num_simple_entries =0 ; pg_tbl->num_extended_entries = total_entries;
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 5/32 staging: gasket: remove driver registration on class creation failure
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
If class_create() fails, remove the gasket driver from the global registration table.
--- drivers/staging/gasket/gasket_core.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/staging/gasket/gasket_core.c +3 -1
@@ -335,7 +335,8 @@ int gasket_register_device(const struct gasket_driver_desc *driver_desc)if (IS_ERR_OR_NULL (internal->class)) {gasket_nodev_error ("Cannot register %s class [ret=%ld]" , driver_desc->name,PTR_ERR (internal->class)); -return PTR_ERR (internal->class); + ret =PTR_ERR (internal->class); + goto unregister_gasket_driver; }/*
@@ -369,6 +370,7 @@ int gasket_register_device(const struct gasket_driver_desc *driver_desc) fail1:class_destroy (internal->class); +unregister_gasket_driver: g_descs[desc_idx].driver_desc =NULL ;return ret; }
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 6/32 staging: gasket: hold mutex on gasket driver unregistration
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Take the global mutex on driver unregistration updates for proper ordering of updates and consistent access procedures.
--- drivers/staging/gasket/gasket_core.c | 4 ++++ 1 file changed, 4 insertions(+)
drivers/staging/gasket/gasket_core.c +4 -0
@@ -371,7 +371,9 @@ int gasket_register_device(const struct gasket_driver_desc *driver_desc)class_destroy (internal->class); unregister_gasket_driver: +mutex_lock (&g_mutex); g_descs[desc_idx].driver_desc =NULL ; +mutex_unlock (&g_mutex);return ret; }EXPORT_SYMBOL (gasket_register_device);
@@ -408,7 +410,9 @@ void gasket_unregister_device(const struct gasket_driver_desc *driver_desc)class_destroy (internal_desc->class);/* Finally, effectively "remove" the driver. */ +mutex_lock (&g_mutex); g_descs[desc_idx].driver_desc =NULL ; +mutex_unlock (&g_mutex);gasket_nodev_info ("removed %s driver" , driver_desc->name); }
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 7/32 staging: gasket: Return EBUSY on mapping create when already in use
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
gasket_sysfs_create_mapping() return EBUSY if sysfs mapping already in use, as a more appropriate error code than the current return of EINVAL, which would indicate invalid parameters.
--- drivers/staging/gasket/gasket_sysfs.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/gasket/gasket_sysfs.c +1 -1
@@ -194,7 +194,7 @@ int gasket_sysfs_create_mapping( "0x%p.", device);put_mapping (mapping);mutex_unlock (&function_mutex); -return -EINVAL ; +return -EBUSY ; }/* Find the first empty entry in the array. */
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 8/32 staging: gasket: Remove stale pointers on error allocating attr array
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
If gasket_sysfs_create_mapping() hits errors allocating the attribute array, remove stale pointers to device info from the mapping object.
--- drivers/staging/gasket/gasket_sysfs.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/staging/gasket/gasket_sysfs.c +2 -0
@@ -225,6 +225,8 @@ int gasket_sysfs_create_mapping( mapping->attribute_count =0 ;if (!mapping->attributes) {gasket_nodev_error ("Unable to allocate sysfs attribute array." ); + mapping->device =NULL ; + mapping->gasket_dev =NULL ;mutex_unlock (&mapping->mutex);mutex_unlock (&function_mutex);return -ENOMEM ;
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 9/32 staging: gasket: convert gasket_mmap_has_permissions to bool return
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
gasket_mmap_has_permissions() should return a boolean value.
--- drivers/staging/gasket/gasket_core.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)
drivers/staging/gasket/gasket_core.c +6 -6
@@ -1241,19 +1241,19 @@ static int gasket_release(struct inode *inode, struct file *file)* that the provided descriptor/range is of adequate size to hold the range to * be mapped. */ -static int gasket_mmap_has_permissions ( +static bool gasket_mmap_has_permissions (struct gasket_dev *gasket_dev,struct vm_area_struct *vma,int bar_permissions) {int requested_permissions;/* Always allow sysadmin to access. */ if (capable (CAP_SYS_ADMIN )) -return 1 ; +return true;/* Never allow non-sysadmins to access to a dead device. */ if (gasket_dev->status !=GASKET_STATUS_ALIVE ) {gasket_log_info (gasket_dev,"Device is dead." ); -return 0 ; +return false; }/* Make sure that no wrong flags are set. */
@@ -1265,7 +1265,7 @@ static int gasket_mmap_has_permissions("Attempting to map a region with requested permissions " "0x%x, but region has permissions 0x%x." , requested_permissions, bar_permissions); -return 0 ; +return false; }/* Do not allow a non-owner to write. */
@@ -1275,10 +1275,10 @@ static int gasket_mmap_has_permissions( gasket_dev,"Attempting to mmap a region for write without owning " "device."); -return 0 ; +return false; } -return 1 ; +return true; }/*
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 10/32 staging: gasket: fix gasket_wait_with_reschedule timeout return code
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Return -ETIMEDOUT, not -EINVAL, on timeout, including callers.
--- drivers/staging/gasket/apex_driver.c | 8 ++++---- drivers/staging/gasket/gasket_core.c | 2 +- 2 files changed, 5 insertions(+), 5 deletions(-)
drivers/staging/gasket/apex_driver.c +4 -4
@@ -490,7 +490,7 @@ static int apex_enter_reset(struct gasket_dev *gasket_dev, uint type)gasket_log_error (gasket_dev,"DMAs did not quiesce within timeout (%d ms)" ,APEX_RESET_RETRY *APEX_RESET_DELAY ); -return -EINVAL ; +return -ETIMEDOUT ; }/* - Enable GCB reset (0x1 to rg_rst_gcb) */
@@ -513,7 +513,7 @@ static int apex_enter_reset(struct gasket_dev *gasket_dev, uint type) gasket_dev,"RAM did not shut down within timeout (%d ms)" , APEX_RESET_RETRY * APEX_RESET_DELAY); -return -EINVAL ; +return -ETIMEDOUT ; }return 0 ;
@@ -562,7 +562,7 @@ static int apex_quit_reset(struct gasket_dev *gasket_dev, uint type) gasket_dev,"RAM did not enable within timeout (%d ms)" , APEX_RESET_RETRY * APEX_RESET_DELAY); -return -EINVAL ; +return -ETIMEDOUT ; }/* - Wait for Reset complete. */
@@ -574,7 +574,7 @@ static int apex_quit_reset(struct gasket_dev *gasket_dev, uint type) gasket_dev,"GCB did not leave reset within timeout (%d ms)" , APEX_RESET_RETRY * APEX_RESET_DELAY); -return -EINVAL ; +return -ETIMEDOUT ; }if (!allow_hw_clock_gating) {
drivers/staging/gasket/gasket_core.c +1 -1
@@ -2106,7 +2106,7 @@ int gasket_wait_with_reschedule("%s timeout: reg %llx timeout (%llu ms)" , __func__, offset, max_retries * delay_ms); -return -EINVAL ; +return -ETIMEDOUT ; }return 0 ; }
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 11/32 staging: gasket: gasket_wait_with_reschedule use msleep
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Replace schedule_timeout() call with msleep() for simplicity.
--- drivers/staging/gasket/gasket_core.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/staging/gasket/gasket_core.c +2 -1
@@ -14,6 +14,7 @@#include "gasket_page_table.h" #include "gasket_sysfs.h" +#include <linux/delay.h> #include <linux/fs.h> #include <linux/init.h> #include <linux/of.h>
@@ -2097,7 +2098,7 @@ int gasket_wait_with_reschedule( tmp =gasket_dev_read_64 (gasket_dev, bar, offset);if ((tmp & mask) == val)break ; -schedule_timeout (msecs_to_jiffies (delay_ms)); +msleep (delay_ms); retries++; }if (retries == max_retries) {
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 12/32 staging: gasket: gasket_wait_with_reschedule simplify logic
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
gasket_wait_with_reschedule() is a little more clear if we just return directly when the waited-for condition is hit. This also allows the following condition check to be removed and identation of the conditionally-executed code to be reduced.
--- drivers/staging/gasket/gasket_core.c | 14 ++++---------- 1 file changed, 4 insertions(+), 10 deletions(-)
drivers/staging/gasket/gasket_core.c +4 -10
@@ -2097,18 +2097,12 @@ int gasket_wait_with_reschedule(while (retries < max_retries) { tmp =gasket_dev_read_64 (gasket_dev, bar, offset);if ((tmp & mask) == val) -break ; +return 0 ;msleep (delay_ms); retries++; } -if (retries == max_retries) { -gasket_log_error ( - gasket_dev, -"%s timeout: reg %llx timeout (%llu ms)" , - __func__, - offset, max_retries * delay_ms); -return -ETIMEDOUT ; - } -return 0 ; +gasket_log_error (gasket_dev,"%s timeout: reg %llx timeout (%llu ms)" , + __func__, offset, max_retries * delay_ms); +return -ETIMEDOUT ; }EXPORT_SYMBOL (gasket_wait_with_reschedule);
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 13/32 staging: gasket: gasket_wait_with_reschedule use 32 bits of retry count
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Don't need a 64-bit retry counter.
--- drivers/staging/gasket/gasket_core.c | 4 ++-- drivers/staging/gasket/gasket_core.h | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/staging/gasket/gasket_core.c +2 -2
@@ -2089,9 +2089,9 @@ struct device *gasket_get_device(struct gasket_dev *dev)**/ int gasket_wait_with_reschedule (struct gasket_dev *gasket_dev,int bar,u64 offset,u64 mask,u64 val, -u64 max_retries,u64 delay_ms) +uint max_retries,u64 delay_ms) { -u64 retries =0 ; +uint retries =0 ;u64 tmp;while (retries < max_retries) {
drivers/staging/gasket/gasket_core.h +1 -1
@@ -702,6 +702,6 @@ struct device *gasket_get_device(struct gasket_dev *dev);/* Helper function, Asynchronous waits on a given set of bits. */ int gasket_wait_with_reschedule (struct gasket_dev *gasket_dev,int bar,u64 offset,u64 mask,u64 val, -u64 max_retries,u64 delay_ms); +uint max_retries,u64 delay_ms);#endif /* __GASKET_CORE_H__ */
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 14/32 staging: gasket: bail out of reset sequence on device callback error
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
If device reset callback returns an error, error out at the gasket level.
--- drivers/staging/gasket/gasket_core.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/staging/gasket/gasket_core.c +3 -1
@@ -1875,9 +1875,11 @@ int gasket_reset_nolock(struct gasket_dev *gasket_dev, uint reset_type)/* Perform a device reset of the requested type. */ ret = driver_desc->device_reset_cb (gasket_dev, reset_type); -if (ret) +if (ret) {gasket_log_error ( gasket_dev,"Device reset cb returned %d." , ret); +return ret; + }/* Reinitialize the page tables and interrupt framework. */ for (i =0 ; i < driver_desc->num_page_tables; ++i)
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 15/32 staging: gasket: drop gasket_cdev_get_info, use container_of
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Remove gasket_cdev_get_info(), use container_of() directly instead, drop unnecessary NULL checks.
--- drivers/staging/gasket/gasket_core.c | 26 ++++---------------------- 1 file changed, 4 insertions(+), 22 deletions(-)
drivers/staging/gasket/gasket_core.c +4 -22
@@ -233,18 +233,6 @@ static inline int gasket_check_and_invoke_callback_nolock(return ret; } -/* -* Retrieve device-specific data via cdev pointer. -* @cdev_ptr: Character device pointer associated with the device. -* -* This function returns the pointer to the device-specific data allocated in -* add_dev_cb for the device associated with cdev_ptr. -*/ -static struct gasket_cdev_info *gasket_cdev_get_info (struct cdev *cdev_ptr) -{ - return container_of(cdev_ptr, struct gasket_cdev_info, cdev); -} -/* * Returns nonzero if the gasket_cdev_info is owned by the current thread group * ID.
@@ -1095,12 +1083,9 @@ static int gasket_open(struct inode *inode, struct file *filp)const struct gasket_driver_desc *driver_desc;struct gasket_ownership *ownership;char task_name[TASK_COMM_LEN ]; -struct gasket_cdev_info *dev_info =gasket_cdev_get_info (inode->i_cdev); +struct gasket_cdev_info *dev_info = + container_of(inode->i_cdev, struct gasket_cdev_info, cdev); -if (!dev_info) { -gasket_nodev_error ("Unable to retrieve device data" ); -return -EINVAL ; - } gasket_dev = dev_info->gasket_dev_ptr; driver_desc = gasket_dev->internal_desc->driver_desc; ownership = &dev_info->ownership;
@@ -1182,11 +1167,8 @@ static int gasket_release(struct inode *inode, struct file *file)const struct gasket_driver_desc *driver_desc;char task_name[TASK_COMM_LEN ];struct gasket_cdev_info *dev_info = - (struct gasket_cdev_info *)gasket_cdev_get_info (inode->i_cdev); -if (!dev_info) { -gasket_nodev_error ("Unable to retrieve device data" ); -return -EINVAL ; - } + container_of(inode->i_cdev, struct gasket_cdev_info, cdev); + gasket_dev = dev_info->gasket_dev_ptr; driver_desc = gasket_dev->internal_desc->driver_desc; ownership = &dev_info->ownership;
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 16/32 staging: gasket: always allow root open for write
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Always allow root to open device for writing.
Drop special-casing of ioctl permissions for root vs. owner.
--- drivers/staging/gasket/apex_driver.c | 9 +-------- drivers/staging/gasket/gasket_core.c | 8 +++++--- drivers/staging/gasket/gasket_ioctl.c | 19 ++++++++----------- 3 files changed, 14 insertions(+), 22 deletions(-)
drivers/staging/gasket/apex_driver.c +1 -8
@@ -629,14 +629,7 @@ static bool is_gcb_in_reset(struct gasket_dev *gasket_dev)*/ static uint apex_ioctl_check_permissions (struct file *filp,uint cmd) { -struct gasket_dev *gasket_dev = filp->private_data; -int root =capable (CAP_SYS_ADMIN ); -int is_owner = gasket_dev->dev_info.ownership.is_owned && - current->tgid == gasket_dev->dev_info.ownership.owner; - -if (root || is_owner) -return 1 ; -return 0 ; +return !!(filp->f_mode &FMODE_WRITE ); }/*
drivers/staging/gasket/gasket_core.c +5 -3
@@ -1085,6 +1085,7 @@ static int gasket_open(struct inode *inode, struct file *filp)char task_name[TASK_COMM_LEN ];struct gasket_cdev_info *dev_info = container_of(inode->i_cdev, struct gasket_cdev_info, cdev); +int is_root =capable (CAP_SYS_ADMIN ); gasket_dev = dev_info->gasket_dev_ptr; driver_desc = gasket_dev->internal_desc->driver_desc;
@@ -1098,7 +1099,7 @@ static int gasket_open(struct inode *inode, struct file *filp)"Attempting to open with tgid %u (%s) (f_mode: 0%03o, " "fmode_write: %d is_root: %u)" , current->tgid, task_name, filp->f_mode, - (filp->f_mode & FMODE_WRITE), capable(CAP_SYS_ADMIN)); + (filp->f_mode & FMODE_WRITE), is_root);/* Always allow non-writing accesses. */ if (!(filp->f_mode &FMODE_WRITE )) {
@@ -1112,8 +1113,9 @@ static int gasket_open(struct inode *inode, struct file *filp) gasket_dev,"Current owner open count (owning tgid %u): %d." , ownership->owner, ownership->write_open_count); -/* Opening a node owned by another TGID is an error (even root.) */ -if (ownership->is_owned && ownership->owner != current->tgid) { +/* Opening a node owned by another TGID is an error (unless root) */ +if (ownership->is_owned && ownership->owner != current->tgid && + !is_root) { gasket_log_error( gasket_dev, "Process %u is opening a node held by %u.",
drivers/staging/gasket/gasket_ioctl.c +8 -11
@@ -171,7 +171,7 @@ long gasket_is_supported_ioctl(uint cmd)*/ static uint gasket_ioctl_check_permissions (struct file *filp,uint cmd) { -uint alive, root, device_owner; +uint alive;fmode_t read, write;struct gasket_dev *gasket_dev = (struct gasket_dev *)filp->private_data;
@@ -183,33 +183,30 @@ static uint gasket_ioctl_check_permissions(struct file *filp, uint cmd) alive, gasket_dev->status); } - root =capable (CAP_SYS_ADMIN ); - read = filp->f_mode &FMODE_READ ; - write = filp->f_mode &FMODE_WRITE ; - device_owner = (gasket_dev->dev_info.ownership.is_owned && - current->tgid == gasket_dev->dev_info.ownership.owner); + read = !!(filp->f_mode &FMODE_READ ); + write = !!(filp->f_mode &FMODE_WRITE );switch (cmd) {case GASKET_IOCTL_RESET :case GASKET_IOCTL_CLEAR_INTERRUPT_COUNTS : -return root || (write && device_owner); +return write;case GASKET_IOCTL_PAGE_TABLE_SIZE :case GASKET_IOCTL_SIMPLE_PAGE_TABLE_SIZE :case GASKET_IOCTL_NUMBER_PAGE_TABLES : -return root || read; +return read;case GASKET_IOCTL_PARTITION_PAGE_TABLE :case GASKET_IOCTL_CONFIG_COHERENT_ALLOCATOR : -return alive && (root || (write && device_owner)); +return alive && write;case GASKET_IOCTL_MAP_BUFFER :case GASKET_IOCTL_UNMAP_BUFFER : -return alive && (root || (write && device_owner)); +return alive && write;case GASKET_IOCTL_CLEAR_EVENTFD :case GASKET_IOCTL_SET_EVENTFD : -return alive && (root || (write && device_owner)); +return alive && write; }return 0 ;/* unknown permissions */
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
Greg Kroah-Hartman
maintainerRe: [PATCH 16/32] staging: gasket: always allow root open for write
Quoted 22 lines · Todd Poynor
From: Todd Poynor <[email protected]>
Always allow root to open device for writing.
Drop special-casing of ioctl permissions for root vs. owner.
Reported-by Dmitry Torokhov <[email protected]>Signed-off-by Zhongze Hu <[email protected]>Signed-off-by Todd Poynor <[email protected]>--- drivers/staging/gasket/apex_driver.c | 9 +-------- drivers/staging/gasket/gasket_core.c | 8 +++++--- drivers/staging/gasket/gasket_ioctl.c | 19 ++++++++----------- 3 files changed, 14 insertions(+), 22 deletions(-)drivers/staging/gasket/apex_driver.c +0
-0@@ -629,14 +629,7 @@ static bool is_gcb_in_reset(struct gasket_dev *gasket_dev)*/ static uint apex_ioctl_check_permissions (struct file *filp,uint cmd)
drivers/staging/gasket/apex_driver.c static bool is_gcb_in_reset(struct gasket_dev *gasket_dev) · line 632
Quoted 10 lines
drivers/staging/gasket/apex_driver.c +1
-8{ -struct gasket_dev *gasket_dev = filp->private_data; -int root =capable (CAP_SYS_ADMIN ); -int is_owner = gasket_dev->dev_info.ownership.is_owned && - current->tgid == gasket_dev->dev_info.ownership.owner; - -if (root || is_owner) -return 1 ; -return 0 ; +return !!(filp->f_mode &FMODE_WRITE );
Shouldn't this return a boolean instead of uint?
Quoted 41 lines
drivers/staging/gasket/apex_driver.c +0
-0}/* drivers/staging/gasket/gasket_core.c +5
-3@@ -1085,6 +1085,7 @@ static int gasket_open(struct inode *inode, struct file *filp)char task_name[TASK_COMM_LEN ];struct gasket_cdev_info *dev_info = container_of(inode->i_cdev, struct gasket_cdev_info, cdev); +int is_root =capable (CAP_SYS_ADMIN ); gasket_dev = dev_info->gasket_dev_ptr; driver_desc = gasket_dev->internal_desc->driver_desc;@@ -1098,7 +1099,7 @@ static int gasket_open(struct inode *inode, struct file *filp)"Attempting to open with tgid %u (%s) (f_mode: 0%03o, " "fmode_write: %d is_root: %u)" , current->tgid, task_name, filp->f_mode, - (filp->f_mode & FMODE_WRITE), capable(CAP_SYS_ADMIN)); + (filp->f_mode & FMODE_WRITE), is_root);/* Always allow non-writing accesses. */ if (!(filp->f_mode &FMODE_WRITE )) {@@ -1112,8 +1113,9 @@ static int gasket_open(struct inode *inode, struct file *filp) gasket_dev,"Current owner open count (owning tgid %u): %d." , ownership->owner, ownership->write_open_count); -/* Opening a node owned by another TGID is an error (even root.) */ -if (ownership->is_owned && ownership->owner != current->tgid) { +/* Opening a node owned by another TGID is an error (unless root) */ +if (ownership->is_owned && ownership->owner != current->tgid && + !is_root) { gasket_log_error( gasket_dev, "Process %u is opening a node held by %u.",drivers/staging/gasket/gasket_ioctl.c +0
-0
drivers/staging/gasket/gasket_ioctl.c long gasket_is_supported_ioctl(uint cmd) · line 173
Quoted 3 lines
drivers/staging/gasket/gasket_ioctl.c +0
-0@@ -171,7 +171,7 @@ long gasket_is_supported_ioctl(uint cmd)*/ static uint gasket_ioctl_check_permissions (struct file *filp,uint cmd) {
Also return bool?
Quoted 6 lines
drivers/staging/gasket/gasket_ioctl.c +1
-1-uint alive, root, device_owner; +uint alive;fmode_t read, write;struct gasket_dev *gasket_dev = (struct gasket_dev *)filp->private_data;
drivers/staging/gasket/gasket_ioctl.c static uint gasket_ioctl_check_permissions(struct file *filp, uint cmd) · line 187
Quoted 10 lines
drivers/staging/gasket/gasket_ioctl.c +2
-5@@ -183,33 +183,30 @@ static uint gasket_ioctl_check_permissions(struct file *filp, uint cmd) alive, gasket_dev->status); } - root =capable (CAP_SYS_ADMIN ); - read = filp->f_mode &FMODE_READ ; - write = filp->f_mode &FMODE_WRITE ; - device_owner = (gasket_dev->dev_info.ownership.is_owned && - current->tgid == gasket_dev->dev_info.ownership.owner); + read = !!(filp->f_mode &FMODE_READ ); + write = !!(filp->f_mode &FMODE_WRITE );
These should be bool as well, right?
I'll drop this patch, thanks.
greg k-h
Re: [PATCH 16/32] staging: gasket: always allow root open for write
Quoted 27 lines · Greg Kroah-Hartman
Quoted 26 lines · Todd Poynor
From: Todd Poynor <[email protected]>
Always allow root to open device for writing.
Drop special-casing of ioctl permissions for root vs. owner.
Reported-by Dmitry Torokhov <[email protected]>Signed-off-by Zhongze Hu <[email protected]>Signed-off-by Todd Poynor <[email protected]>--- drivers/staging/gasket/apex_driver.c | 9 +-------- drivers/staging/gasket/gasket_core.c | 8 +++++--- drivers/staging/gasket/gasket_ioctl.c | 19 ++++++++----------- 3 files changed, 14 insertions(+), 22 deletions(-)drivers/staging/gasket/apex_driver.c +0
-3@@ -629,14 +629,7 @@ static bool is_gcb_in_reset(struct gasket_dev *gasket_dev)*/ static uint apex_ioctl_check_permissions (struct file *filp,uint cmd) { -struct gasket_dev *gasket_dev = filp->private_data; -int root =capable (CAP_SYS_ADMIN ); - int is_owner = gasket_dev->dev_info.ownership.is_owned &&
Quoted 8 lines
drivers/staging/gasket/apex_driver.c static bool is_gcb_in_reset(struct gasket_dev *gasket_dev) · line 632
Quoted 6 lines
drivers/staging/gasket/apex_driver.c +1
-5- current->tgid == gasket_dev->dev_info.ownership.owner; - -if (root || is_owner) -return 1 ; -return 0 ; +return !!(filp->f_mode &FMODE_WRITE );Shouldn't this return a boolean instead of uint?
There's a patch farther down in the stack that does that, I wasn't sure if I should collapse the two things together into a single patch, will fold that patch into this one.
Quoted 39 lines
Quoted 39 lines
drivers/staging/gasket/apex_driver.c +0
-0}/* drivers/staging/gasket/gasket_core.c +5
-3@@ -1085,6 +1085,7 @@ static int gasket_open(struct inode *inode, struct file *filp)char task_name[TASK_COMM_LEN ];struct gasket_cdev_info *dev_info = container_of(inode->i_cdev, struct gasket_cdev_info, cdev); +int is_root =capable (CAP_SYS_ADMIN ); gasket_dev = dev_info->gasket_dev_ptr; driver_desc = gasket_dev->internal_desc->driver_desc;@@ -1098,7 +1099,7 @@ static int gasket_open(struct inode *inode, struct file *filp)"Attempting to open with tgid %u (%s) (f_mode: 0%03o, " "fmode_write: %d is_root: %u)" , current->tgid, task_name, filp->f_mode, - (filp->f_mode & FMODE_WRITE), capable(CAP_SYS_ADMIN)); + (filp->f_mode & FMODE_WRITE), is_root);/* Always allow non-writing accesses. */ if (!(filp->f_mode &FMODE_WRITE )) {@@ -1112,8 +1113,9 @@ static int gasket_open(struct inode *inode, struct file *filp) gasket_dev,"Current owner open count (owning tgid %u): %d." , ownership->owner, ownership->write_open_count); -/* Opening a node owned by another TGID is an error (even root.) */ -if (ownership->is_owned && ownership->owner != current->tgid) { +/* Opening a node owned by another TGID is an error (unless root) */ +if (ownership->is_owned && ownership->owner != current->tgid && + !is_root) { gasket_log_error( gasket_dev, "Process %u is opening a node held by %u.",drivers/staging/gasket/gasket_ioctl.c +0
-0
Quoted 8 lines
drivers/staging/gasket/gasket_ioctl.c long gasket_is_supported_ioctl(uint cmd) · line 173
Quoted 6 lines
drivers/staging/gasket/gasket_ioctl.c +0
-0@@ -171,7 +171,7 @@ long gasket_is_supported_ioctl(uint cmd)*/ static uint gasket_ioctl_check_permissions (struct file *filp,uint cmd) {Also return bool?
Will fix, I had meant to send a patch for this, will just do that here.
Quoted 10 lines
Quoted 10 lines
drivers/staging/gasket/gasket_ioctl.c +1
-1-uint alive, root, device_owner; +uint alive;fmode_t read, write;struct gasket_dev *gasket_dev = (struct gasket_dev *)filp->private_data;@@ -183,33 +183,30 @@ static uint gasket_ioctl_check_permissions(struct file *filp, uint cmd) alive, gasket_dev->status); }
Quoted 9 lines
drivers/staging/gasket/gasket_ioctl.c static uint gasket_ioctl_check_permissions(struct file *filp, uint cmd) · line 187
Quoted 7 lines
drivers/staging/gasket/gasket_ioctl.c +2
-5- root =capable (CAP_SYS_ADMIN ); - read = filp->f_mode &FMODE_READ ; - write = filp->f_mode &FMODE_WRITE ; - device_owner = (gasket_dev->dev_info.ownership.is_owned && - current->tgid == gasket_dev->dev_info.ownership.owner); + read = !!(filp->f_mode &FMODE_READ ); + write = !!(filp->f_mode &FMODE_WRITE );These should be bool as well, right?
Right.
Quoted 4 lines
I'll drop this patch, thanks.
greg k-h
Signature · 1 line
Todd
PATCH v3 17/32 staging: gasket: annotate ioctl arg with __user
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
For sparse checking.
--- drivers/staging/gasket/apex_driver.c | 11 ++-- drivers/staging/gasket/gasket_core.c | 6 ++- drivers/staging/gasket/gasket_core.h | 4 +- drivers/staging/gasket/gasket_ioctl.c | 72 ++++++++++++++------------- drivers/staging/gasket/gasket_ioctl.h | 4 +- 5 files changed, 52 insertions(+), 45 deletions(-)
drivers/staging/gasket/apex_driver.c +6 -5
@@ -5,6 +5,7 @@* Copyright (C) 2018 Google, Inc. */ +#include <linux/compiler.h> #include <linux/delay.h> #include <linux/fs.h> #include <linux/init.h>
@@ -142,9 +143,9 @@ static int apex_get_status(struct gasket_dev *gasket_dev);static uint apex_ioctl_check_permissions (struct file *file,uint cmd); -static long apex_ioctl (struct file *file,uint cmd,ulong arg); +static long apex_ioctl (struct file *file,uint cmd,void __user *arg); -static long apex_clock_gating (struct gasket_dev *gasket_dev,ulong arg); +static long apex_clock_gating (struct gasket_dev *gasket_dev,void __user *arg);static int apex_enter_reset (struct gasket_dev *gasket_dev,uint type);
@@ -635,7 +636,7 @@ static uint apex_ioctl_check_permissions(struct file *filp, uint cmd)/* * Apex-specific ioctl handler. */ -static long apex_ioctl (struct file *filp,uint cmd,ulong arg) +static long apex_ioctl (struct file *filp,uint cmd,void __user *arg) {struct gasket_dev *gasket_dev = filp->private_data;
@@ -655,7 +656,7 @@ static long apex_ioctl(struct file *filp, uint cmd, ulong arg)* @gasket_dev: Gasket device pointer. * @arg: User ioctl arg, in this case to a apex_gate_clock_ioctl struct. */ -static long apex_clock_gating (struct gasket_dev *gasket_dev,ulong arg) +static long apex_clock_gating (struct gasket_dev *gasket_dev,void __user *arg) {struct apex_gate_clock_ioctl ibuf;
@@ -663,7 +664,7 @@ static long apex_clock_gating(struct gasket_dev *gasket_dev, ulong arg)return 0 ;if (allow_sw_clock_gating) { -if (copy_from_user (&ibuf, (void __user *)arg,sizeof (ibuf))) +if (copy_from_user (&ibuf, arg,sizeof (ibuf)))return -EFAULT ; gasket_log_error(
drivers/staging/gasket/gasket_core.c +4 -2
@@ -14,6 +14,7 @@#include "gasket_page_table.h" #include "gasket_sysfs.h" +#include <linux/compiler.h> #include <linux/delay.h> #include <linux/fs.h> #include <linux/init.h>
@@ -1823,14 +1824,15 @@ static long gasket_ioctl(struct file *filp, uint cmd, ulong arg)* check_and_invoke_callback. */ if (driver_desc->ioctl_handler_cb) -return driver_desc->ioctl_handler_cb (filp, cmd, arg); +return driver_desc->ioctl_handler_cb ( + filp, cmd, (void __user *)arg);gasket_log_error ( gasket_dev,"Received unknown ioctl 0x%x" , cmd);return -EINVAL ; } -return gasket_handle_ioctl (filp, cmd, arg); +return gasket_handle_ioctl (filp, cmd, (void __user *)arg); }int gasket_reset (struct gasket_dev *gasket_dev,uint reset_type)
drivers/staging/gasket/gasket_core.h +2 -2
@@ -315,7 +315,7 @@ struct gasket_dev {/* Type of the ioctl permissions check callback. See below. */ typedef int (*gasket_ioctl_permissions_cb_t )( -struct file *filp,uint cmd,ulong arg); +struct file *filp,uint cmd,void __user *arg);/* * Device type descriptor.
@@ -549,7 +549,7 @@ struct gasket_driver_desc {* return -EINVAL. Should return an error status (either -EINVAL or * the error result of the ioctl being handled). */ -long (*ioctl_handler_cb)(struct file *filp,uint cmd,ulong arg); +long (*ioctl_handler_cb)(struct file *filp,uint cmd,void __user *arg);/* * device_status_cb: Callback to determine device health.
drivers/staging/gasket/gasket_ioctl.c +37 -35
200 lines not on this page · raw message
drivers/staging/gasket/gasket_ioctl.h +3 -1
@@ -5,6 +5,8 @@#include "gasket_core.h" +#include <linux/compiler.h> +/* * Handle Gasket common ioctls. * @filp: Pointer to the ioctl's file.
@@ -13,7 +15,7 @@* * Returns 0 on success and nonzero on failure. */ -long gasket_handle_ioctl (struct file *filp,uint cmd,ulong arg); +long gasket_handle_ioctl (struct file *filp,uint cmd,void __user *arg);/* * Determines if an ioctl is part of the standard Gasket framework.
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
Greg Kroah-Hartman
maintainerRe: [PATCH 17/32] staging: gasket: annotate ioctl arg with __user
Quoted 3 lines · Todd Poynor
From: Todd Poynor <[email protected]>
For sparse checking.
Close, but you can do better :)
Quoted 26 lines
Reported-by Dmitry Torokhov <[email protected]>Signed-off-by Zhongze Hu <[email protected]>Signed-off-by Todd Poynor <[email protected]>Reviewed-by Dmitry Torokhov <[email protected]>--- drivers/staging/gasket/apex_driver.c | 11 ++-- drivers/staging/gasket/gasket_core.c | 6 ++- drivers/staging/gasket/gasket_core.h | 4 +- drivers/staging/gasket/gasket_ioctl.c | 72 ++++++++++++++------------- drivers/staging/gasket/gasket_ioctl.h | 4 +- 5 files changed, 52 insertions(+), 45 deletions(-)drivers/staging/gasket/apex_driver.c +1
-0@@ -5,6 +5,7 @@* Copyright (C) 2018 Google, Inc. */ +#include <linux/compiler.h> #include <linux/delay.h> #include <linux/fs.h> #include <linux/init.h>
drivers/staging/gasket/apex_driver.c static int apex_get_status(struct gasket_dev *gasket_dev); · line 148
Quoted 8 lines
drivers/staging/gasket/apex_driver.c +2
-2@@ -142,9 +143,9 @@ static int apex_get_status(struct gasket_dev *gasket_dev);static uint apex_ioctl_check_permissions (struct file *file,uint cmd); -static long apex_ioctl (struct file *file,uint cmd,ulong arg); +static long apex_ioctl (struct file *file,uint cmd,void __user *arg); -static long apex_clock_gating (struct gasket_dev *gasket_dev,ulong arg); +static long apex_clock_gating (struct gasket_dev *gasket_dev,void __user *arg);
Make this a __user pointer to the correct struct type you are handling here. You know what the type is, use it.
Quoted 12 lines
drivers/staging/gasket/apex_driver.c +1
-1static int apex_enter_reset (struct gasket_dev *gasket_dev,uint type);@@ -635,7 +636,7 @@ static uint apex_ioctl_check_permissions(struct file *filp, uint cmd)/* * Apex-specific ioctl handler. */ -static long apex_ioctl (struct file *filp,uint cmd,ulong arg) +static long apex_ioctl (struct file *filp,uint cmd,void __user *arg) {struct gasket_dev *gasket_dev = filp->private_data;
drivers/staging/gasket/apex_driver.c static long apex_ioctl(struct file *filp, uint cmd, ulong arg) · line 659
Quoted 5 lines
drivers/staging/gasket/apex_driver.c +1
-1@@ -655,7 +656,7 @@ static long apex_ioctl(struct file *filp, uint cmd, ulong arg)* @gasket_dev: Gasket device pointer. * @arg: User ioctl arg, in this case to a apex_gate_clock_ioctl struct. */ -static long apex_clock_gating (struct gasket_dev *gasket_dev,ulong arg) +static long apex_clock_gating (struct gasket_dev *gasket_dev,void __user *arg)
As above, this should be a different type.
Quoted 25 lines
drivers/staging/gasket/apex_driver.c +1
-1{struct apex_gate_clock_ioctl ibuf;@@ -663,7 +664,7 @@ static long apex_clock_gating(struct gasket_dev *gasket_dev, ulong arg)return 0 ;if (allow_sw_clock_gating) { -if (copy_from_user (&ibuf, (void __user *)arg,sizeof (ibuf))) +if (copy_from_user (&ibuf, arg,sizeof (ibuf)))return -EFAULT ; gasket_log_error(drivers/staging/gasket/gasket_core.c +1
-0@@ -14,6 +14,7 @@#include "gasket_page_table.h" #include "gasket_sysfs.h" +#include <linux/compiler.h> #include <linux/delay.h> #include <linux/fs.h> #include <linux/init.h>
drivers/staging/gasket/gasket_core.c static long gasket_ioctl(struct file *filp, uint cmd, ulong arg) · line 1828
Quoted 6 lines
drivers/staging/gasket/gasket_core.c +2
-1@@ -1823,14 +1824,15 @@ static long gasket_ioctl(struct file *filp, uint cmd, ulong arg)* check_and_invoke_callback. */ if (driver_desc->ioctl_handler_cb) -return driver_desc->ioctl_handler_cb (filp, cmd, arg); +return driver_desc->ioctl_handler_cb ( + filp, cmd, (void __user *)arg);
You can use a temp variable and then only have to cast things once then, instead of twice, if you care. Not a big deal.
drivers/staging/gasket/gasket_core.c static long gasket_ioctl(struct file *filp, uint cmd, ulong arg) · line 1831
Quoted 3 lines
drivers/staging/gasket/gasket_core.c +0
-0gasket_log_error ( gasket_dev,"Received unknown ioctl 0x%x" , cmd);
This is a fun way to cause a DoS on your system, you should fix this up in later changes.
Quoted 22 lines
drivers/staging/gasket/gasket_core.c +1
-1return -EINVAL ; } -return gasket_handle_ioctl (filp, cmd, arg); +return gasket_handle_ioctl (filp, cmd, (void __user *)arg); }int gasket_reset (struct gasket_dev *gasket_dev,uint reset_type)drivers/staging/gasket/gasket_core.h +1
-1@@ -315,7 +315,7 @@ struct gasket_dev {/* Type of the ioctl permissions check callback. See below. */ typedef int (*gasket_ioctl_permissions_cb_t )( -struct file *filp,uint cmd,ulong arg); +struct file *filp,uint cmd,void __user *arg);/* * Device type descriptor.
drivers/staging/gasket/gasket_core.h struct gasket_driver_desc { · line 552
Quoted 5 lines
drivers/staging/gasket/gasket_core.h +1
-1@@ -549,7 +549,7 @@ struct gasket_driver_desc {* return -EINVAL. Should return an error status (either -EINVAL or * the error result of the ioctl being handled). */ -long (*ioctl_handler_cb)(struct file *filp,uint cmd,ulong arg); +long (*ioctl_handler_cb)(struct file *filp,uint cmd,void __user *arg);
Why are you not using the typedef above?
Quoted 28 lines
drivers/staging/gasket/gasket_core.h +0
-0/* * device_status_cb: Callback to determine device health. drivers/staging/gasket/gasket_ioctl.c +4
-3@@ -7,6 +7,7 @@#include "gasket_interrupt.h" #include "gasket_logging.h" #include "gasket_page_table.h" +#include <linux/compiler.h> #include <linux/fs.h> #include <linux/uaccess.h> @@ -23,17 +24,18 @@#endif static uint gasket_ioctl_check_permissions (struct file *filp,uint cmd); -static int gasket_set_event_fd (struct gasket_dev *dev,ulong arg); +static int gasket_set_event_fd (struct gasket_dev *dev,void __user *arg);static int gasket_read_page_table_size ( -struct gasket_dev *gasket_dev,ulong arg); +struct gasket_dev *gasket_dev,void __user *arg);static int gasket_read_simple_page_table_size ( -struct gasket_dev *gasket_dev,ulong arg); +struct gasket_dev *gasket_dev,void __user *arg); static int gasket_partition_page_table(
drivers/staging/gasket/gasket_ioctl.c line 38
Quoted 10 lines
drivers/staging/gasket/gasket_ioctl.c +5
-4- struct gasket_dev *gasket_dev, ulong arg); -static int gasket_map_buffers (struct gasket_dev *gasket_dev,ulong arg); -static int gasket_unmap_buffers (struct gasket_dev *gasket_dev,ulong arg); + struct gasket_dev *gasket_dev, void __user *arg); +static int gasket_map_buffers (struct gasket_dev *gasket_dev,void __user *arg); +static int gasket_unmap_buffers (struct gasket_dev *gasket_dev, +void __user *arg);static int gasket_config_coherent_allocator ( -struct gasket_dev *gasket_dev,ulong arg); +struct gasket_dev *gasket_dev,void __user *arg);
For most of these you know the real type, please just use it.
drivers/staging/gasket/gasket_ioctl.c static int gasket_config_coherent_allocator( · line 51
Quoted 11 lines
drivers/staging/gasket/gasket_ioctl.c +2
-1/* * standard ioctl dispatch function. @@ -43,9 +45,10 @@ static int gasket_config_coherent_allocator(* * Standard ioctl dispatcher; forwards operations to individual handlers. */ -long gasket_handle_ioctl (struct file *filp,uint cmd,ulong arg) +long gasket_handle_ioctl (struct file *filp,uint cmd,void __user *arg) {struct gasket_dev *gasket_dev; +ulong intarg = (ulong )arg;
Ick, really? That's a horrible name for a variable, especially as it doesn't describe what it is...
and "unsigned long" please, ulong is not a "real" kernel type.
Quoted 11 lines
drivers/staging/gasket/gasket_ioctl.c +2
-2int retval; gasket_dev = (struct gasket_dev *)filp->private_data;@@ -74,16 +77,16 @@ long gasket_handle_ioctl(struct file *filp, uint cmd, ulong arg)*/ switch (cmd) {case GASKET_IOCTL_RESET : -trace_gasket_ioctl_integer_data (arg); - retval =gasket_reset (gasket_dev, arg); +trace_gasket_ioctl_integer_data (intarg); + retval =gasket_reset (gasket_dev, intarg);
drivers/staging/gasket/gasket_ioctl.c long gasket_handle_ioctl(struct file *filp, uint cmd, ulong arg) · line 89
Quoted 10 lines
drivers/staging/gasket/gasket_ioctl.c +2
-2break ;case GASKET_IOCTL_SET_EVENTFD : retval =gasket_set_event_fd (gasket_dev, arg);break ;case GASKET_IOCTL_CLEAR_EVENTFD : -trace_gasket_ioctl_integer_data (arg); +trace_gasket_ioctl_integer_data (intarg); retval =gasket_interrupt_clear_eventfd ( - gasket_dev->interrupt_data, (int )arg); + gasket_dev->interrupt_data, (int )intarg);
See, look at that cast, why would you ever think to cast a variabled with "int" in the name of it, to an int? Naming is hard :(
drivers/staging/gasket/gasket_ioctl.c long gasket_handle_ioctl(struct file *filp, uint cmd, ulong arg) · line 98
Quoted 10 lines
drivers/staging/gasket/gasket_ioctl.c +1
-1break ;case GASKET_IOCTL_PARTITION_PAGE_TABLE :trace_gasket_ioctl_integer_data (arg);@@ -91,7 +94,7 @@ long gasket_handle_ioctl(struct file *filp, uint cmd, ulong arg)break ;case GASKET_IOCTL_NUMBER_PAGE_TABLES :trace_gasket_ioctl_integer_data (gasket_dev->num_page_tables); - if (copy_to_user((void __user *)arg, + if (copy_to_user(arg, &gasket_dev->num_page_tables,
Nit, you can move this line up one now.
I'll stop now, you get the idea. This should be broken up a lot.
thanks,
greg k-h
Re: [PATCH 17/32] staging: gasket: annotate ioctl arg with __user
Quoted 36 lines · Greg Kroah-Hartman
Quoted 3 lines · Todd Poynor
From: Todd Poynor <[email protected]>
For sparse checking.
Close, but you can do better :)
Quoted 29 lines
Reported-by Dmitry Torokhov <[email protected]>Signed-off-by Zhongze Hu <[email protected]>Signed-off-by Todd Poynor <[email protected]>Reviewed-by Dmitry Torokhov <[email protected]>--- drivers/staging/gasket/apex_driver.c | 11 ++-- drivers/staging/gasket/gasket_core.c | 6 ++- drivers/staging/gasket/gasket_core.h | 4 +- drivers/staging/gasket/gasket_ioctl.c | 72 ++++++++++++++------------- drivers/staging/gasket/gasket_ioctl.h | 4 +- 5 files changed, 52 insertions(+), 45 deletions(-)drivers/staging/gasket/apex_driver.c +1
-0@@ -5,6 +5,7 @@* Copyright (C) 2018 Google, Inc. */ +#include <linux/compiler.h> #include <linux/delay.h> #include <linux/fs.h> #include <linux/init.h> @@ -142,9 +143,9 @@ static int apex_get_status(struct gasket_dev *gasket_dev);static uint apex_ioctl_check_permissions (struct file *file,uint cmd);
Quoted 8 lines
drivers/staging/gasket/apex_driver.c static int apex_get_status(struct gasket_dev *gasket_dev); · line 148
Quoted 5 lines
drivers/staging/gasket/apex_driver.c +2
-2-static long apex_ioctl (struct file *file,uint cmd,ulong arg); +static long apex_ioctl (struct file *file,uint cmd,void __user *arg); -static long apex_clock_gating (struct gasket_dev *gasket_dev,ulong arg); +static long apex_clock_gating (struct gasket_dev *gasket_dev,void __user *arg);Make this a
__userpointer to the correct struct type you are handling here. You know what the type is, use it.
Got it.
Quoted 41 lines
Quoted 13 lines
drivers/staging/gasket/apex_driver.c +1
-1static int apex_enter_reset (struct gasket_dev *gasket_dev,uint type);@@ -635,7 +636,7 @@ static uint apex_ioctl_check_permissions(struct file *filp, uint cmd)/* * Apex-specific ioctl handler. */ -static long apex_ioctl (struct file *filp,uint cmd,ulong arg) +static long apex_ioctl (struct file *filp,uint cmd,void __user *arg) {struct gasket_dev *gasket_dev = filp->private_data;drivers/staging/gasket/apex_driver.c static long apex_ioctl(struct file *filp, uint cmd, ulong arg) · line 659
Quoted 5 lines
drivers/staging/gasket/apex_driver.c +1
-1@@ -655,7 +656,7 @@ static long apex_ioctl(struct file *filp, uint cmd, ulong arg)* @gasket_dev: Gasket device pointer. * @arg: User ioctl arg, in this case to a apex_gate_clock_ioctl struct. */ -static long apex_clock_gating (struct gasket_dev *gasket_dev,ulong arg) +static long apex_clock_gating (struct gasket_dev *gasket_dev,void __user *arg)As above, this should be a different type.
Quoted 20 lines
drivers/staging/gasket/apex_driver.c +1
-1{struct apex_gate_clock_ioctl ibuf;@@ -663,7 +664,7 @@ static long apex_clock_gating(struct gasket_dev *gasket_dev, ulong arg)return 0 ;if (allow_sw_clock_gating) { -if (copy_from_user (&ibuf, (void __user *)arg,sizeof (ibuf))) +if (copy_from_user (&ibuf, arg,sizeof (ibuf)))return -EFAULT ; gasket_log_error(drivers/staging/gasket/gasket_core.c +0
-0@@ -14,6 +14,7 @@#include "gasket_page_table.h" #include "gasket_sysfs.h"
Quoted 14 lines
drivers/staging/gasket/gasket_core.c static long gasket_ioctl(struct file *filp, uint cmd, ulong arg) · line 1828
Quoted 11 lines
drivers/staging/gasket/gasket_core.c +3
-1+#include <linux/compiler.h> #include <linux/delay.h> #include <linux/fs.h> #include <linux/init.h> @@ -1823,14 +1824,15 @@ static long gasket_ioctl(struct file *filp, uint cmd, ulong arg)* check_and_invoke_callback. */ if (driver_desc->ioctl_handler_cb) -return driver_desc->ioctl_handler_cb (filp, cmd, arg); +return driver_desc->ioctl_handler_cb ( + filp, cmd, (void __user *)arg);You can use a temp variable and then only have to cast things once then, instead of twice, if you care. Not a big deal.
But then I have to name it, and I suck at that. :P I'm trying switching to "arg" for ulong/int arguments and "argp" for pointer arguments.
Quoted 7 lines
drivers/staging/gasket/gasket_core.c static long gasket_ioctl(struct file *filp, uint cmd, ulong arg) · line 1831
Quoted 4 lines
drivers/staging/gasket/gasket_core.c +0
-0gasket_log_error ( gasket_dev,"Received unknown ioctl 0x%x" , cmd);This is a fun way to cause a DoS on your system, you should fix this up in later changes.
Yeah there's another patch coming soon that converts a bunch of errors and infos to debugs.
Quoted 20 lines
Quoted 20 lines
drivers/staging/gasket/gasket_core.c +1
-1return -EINVAL ; } -return gasket_handle_ioctl (filp, cmd, arg); +return gasket_handle_ioctl (filp, cmd, (void __user *)arg); }int gasket_reset (struct gasket_dev *gasket_dev,uint reset_type)drivers/staging/gasket/gasket_core.h +1
-1@@ -315,7 +315,7 @@ struct gasket_dev {/* Type of the ioctl permissions check callback. See below. */ typedef int (*gasket_ioctl_permissions_cb_t )( -struct file *filp,uint cmd,ulong arg); +struct file *filp,uint cmd,void __user *arg);
Quoted 10 lines
drivers/staging/gasket/gasket_core.h struct gasket_driver_desc { · line 552
Quoted 8 lines
drivers/staging/gasket/gasket_core.h +1
-1/* * Device type descriptor. @@ -549,7 +549,7 @@ struct gasket_driver_desc {* return -EINVAL. Should return an error status (either -EINVAL or * the error result of the ioctl being handled). */ -long (*ioctl_handler_cb)(struct file *filp,uint cmd,ulong arg); +long (*ioctl_handler_cb)(struct file *filp,uint cmd,void __user *arg);Why are you not using the typedef above?
There's a typedef for the permissions check callback, but not for the handler callback. It's a bit confusing, so I tried adding a typedef for the handler, but now checkpatch is spanking me for adding new typedefs -- maybe I should drop the existing typedef?
Quoted 34 lines
Quoted 34 lines
drivers/staging/gasket/gasket_core.h +0
-0/* * device_status_cb: Callback to determine device health. drivers/staging/gasket/gasket_ioctl.c +6
-6@@ -7,6 +7,7 @@#include "gasket_interrupt.h" #include "gasket_logging.h" #include "gasket_page_table.h" +#include <linux/compiler.h> #include <linux/fs.h> #include <linux/uaccess.h> @@ -23,17 +24,18 @@#endif static uint gasket_ioctl_check_permissions (struct file *filp,uint cmd); -static int gasket_set_event_fd (struct gasket_dev *dev,ulong arg); +static int gasket_set_event_fd (struct gasket_dev *dev,void __user *arg);static int gasket_read_page_table_size ( -struct gasket_dev *gasket_dev,ulong arg); +struct gasket_dev *gasket_dev,void __user *arg);static int gasket_read_simple_page_table_size ( -struct gasket_dev *gasket_dev,ulong arg); +struct gasket_dev *gasket_dev,void __user *arg);static int gasket_partition_page_table ( -struct gasket_dev *gasket_dev,ulong arg); -static int gasket_map_buffers (struct gasket_dev *gasket_dev,ulong arg); -static int gasket_unmap_buffers (struct gasket_dev *gasket_dev,ulong arg); +struct gasket_dev *gasket_dev,void __user *arg); +static int gasket_map_buffers (struct gasket_dev *gasket_dev,void __user *arg);
Quoted 8 lines
drivers/staging/gasket/gasket_ioctl.c line 38
Quoted 5 lines
drivers/staging/gasket/gasket_ioctl.c +3
-1+static int gasket_unmap_buffers (struct gasket_dev *gasket_dev, +void __user *arg);static int gasket_config_coherent_allocator ( -struct gasket_dev *gasket_dev,ulong arg); +struct gasket_dev *gasket_dev,void __user *arg);For most of these you know the real type, please just use it.
Roger.
Quoted 34 lines
drivers/staging/gasket/gasket_ioctl.c static int gasket_config_coherent_allocator( · line 51
Quoted 12 lines
drivers/staging/gasket/gasket_ioctl.c +2
-1/* * standard ioctl dispatch function. @@ -43,9 +45,10 @@ static int gasket_config_coherent_allocator(* * Standard ioctl dispatcher; forwards operations to individual handlers. */ -long gasket_handle_ioctl (struct file *filp,uint cmd,ulong arg) +long gasket_handle_ioctl (struct file *filp,uint cmd,void __user *arg) {struct gasket_dev *gasket_dev; +ulong intarg = (ulong )arg;Ick, really? That's a horrible name for a variable, especially as it doesn't describe what it is...
and "unsigned long" please, ulong is not a "real" kernel type.
Quoted 16 lines
drivers/staging/gasket/gasket_ioctl.c +2
-2int retval; gasket_dev = (struct gasket_dev *)filp->private_data;@@ -74,16 +77,16 @@ long gasket_handle_ioctl(struct file *filp, uint cmd, ulong arg)*/ switch (cmd) {case GASKET_IOCTL_RESET : -trace_gasket_ioctl_integer_data (arg); - retval =gasket_reset (gasket_dev, arg); +trace_gasket_ioctl_integer_data (intarg); + retval =gasket_reset (gasket_dev, intarg);break ;case GASKET_IOCTL_SET_EVENTFD : retval =gasket_set_event_fd (gasket_dev, arg);break ;case GASKET_IOCTL_CLEAR_EVENTFD :
Quoted 8 lines
drivers/staging/gasket/gasket_ioctl.c long gasket_handle_ioctl(struct file *filp, uint cmd, ulong arg) · line 89
Quoted 5 lines
drivers/staging/gasket/gasket_ioctl.c +2
-2-trace_gasket_ioctl_integer_data (arg); +trace_gasket_ioctl_integer_data (intarg); retval =gasket_interrupt_clear_eventfd ( - gasket_dev->interrupt_data, (int )arg); + gasket_dev->interrupt_data, (int )intarg);See, look at that cast, why would you ever think to cast a variabled with "int" in the name of it, to an int? Naming is hard :(
Yeah now that I look at it that was a pretty horrible name. I hate naming things.
Quoted 14 lines
drivers/staging/gasket/gasket_ioctl.c long gasket_handle_ioctl(struct file *filp, uint cmd, ulong arg) · line 98
Quoted 12 lines
drivers/staging/gasket/gasket_ioctl.c +1
-1break ;case GASKET_IOCTL_PARTITION_PAGE_TABLE :trace_gasket_ioctl_integer_data (arg);@@ -91,7 +94,7 @@ long gasket_handle_ioctl(struct file *filp, uint cmd, ulong arg)break ;case GASKET_IOCTL_NUMBER_PAGE_TABLES :trace_gasket_ioctl_integer_data (gasket_dev->num_page_tables); - if (copy_to_user((void __user *)arg, + if (copy_to_user(arg, &gasket_dev->num_page_tables,Nit, you can move this line up one now.
OK.
Quoted 2 lines
I'll stop now, you get the idea. This should be broken up a lot.
I'll split up in some fashion.
Thanks -- Todd
Quoted 4 lines
thanks,
greg k-h
Greg Kroah-Hartman
maintainerRe: [PATCH 17/32] staging: gasket: annotate ioctl arg with __user
Quoted 16 lines · Todd Poynor
Quoted 5 lines
Quoted 5 lines
/* Type of the ioctl permissions check callback. See below. */ typedef int (*gasket_ioctl_permissions_cb_t)(
drivers/staging/gasket/gasket_core.h +1
-1- struct file *filp, uint cmd, ulong arg); + struct file *filp, uint cmd, void __user *arg);Quoted 10 lines
drivers/staging/gasket/gasket_core.h struct gasket_driver_desc { · line 552
Quoted 8 lines
/*
* Device type descriptor.drivers/staging/gasket/gasket_core.h +1
-1@@ -549,7 +549,7 @@ struct gasket_driver_desc {* return -EINVAL. Should return an error status (either -EINVAL or * the error result of the ioctl being handled). */ -long (*ioctl_handler_cb)(struct file *filp,uint cmd,ulong arg); +long (*ioctl_handler_cb)(struct file *filp,uint cmd,void __user *arg);Why are you not using the typedef above?
Quoted 4 lines
There's a typedef for the permissions check callback, but not for the handler callback. It's a bit confusing, so I tried adding a typedef for the handler, but now checkpatch is spanking me for adding new typedefs -- maybe I should drop the existing typedef?
No, typedefs for function pointers is just fine, checkpatch should not be complaining about that. And even if it is, it's just a guideline, not a hard rule that you have to abide by everything it says :)
thanks,
greg k-h
Re: [PATCH 17/32] staging: gasket: annotate ioctl arg with __user
In reply to Todd PoynorQuoted 9 lines · Todd Poynor
Quoted 8 lines
drivers/staging/gasket/gasket_core.h struct gasket_driver_desc { · line 552
Quoted 6 lines
drivers/staging/gasket/gasket_core.h +1
-1@@ -549,7 +549,7 @@ struct gasket_driver_desc {* return -EINVAL. Should return an error status (either -EINVAL or * the error result of the ioctl being handled). */ -long (*ioctl_handler_cb)(struct file *filp,uint cmd,ulong arg); +long (*ioctl_handler_cb)(struct file *filp,uint cmd,void __user *arg);Why are you not using the typedef above?
Quoted 5 lines
There's a typedef for the permissions check callback, but not for the handler callback. It's a bit confusing, so I tried adding a typedef for the handler, but now checkpatch is spanking me for adding new typedefs -- maybe I should drop the existing typedef?
Someone should update checkpatch to not complain about function typedefs because those are obviously OK. It wouldn't have even occured to me that people were avoiding them because of checkpatch...
regards,
dan carpenter
PATCH v3 18/32 staging: gasket: gasket_enable_dev remove unnecessary variable
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Remove unnecessary variable, pass constant param instead.
--- drivers/staging/gasket/gasket_core.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-)
drivers/staging/gasket/gasket_core.c +1 -4
@@ -899,7 +899,6 @@ static int gasket_enable_dev( {int tbl_idx;int ret; -bool has_dma_ops;struct device *ddev;const struct gasket_driver_desc *driver_desc = internal_desc->driver_desc;
@@ -918,8 +917,6 @@ static int gasket_enable_dev(return ret; } - has_dma_ops = true; -for (tbl_idx =0 ; tbl_idx < driver_desc->num_page_tables; tbl_idx++) {gasket_log_debug ( gasket_dev,"Initializing page table %d." , tbl_idx);
@@ -937,7 +934,7 @@ static int gasket_enable_dev( &gasket_dev->bar_data[ driver_desc->page_table_bar_index], &driver_desc->page_table_configs[tbl_idx], - ddev, gasket_dev->pci_dev, has_dma_ops); + ddev, gasket_dev->pci_dev, true);if (ret) { gasket_log_error( gasket_dev,
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 19/32 staging: gasket: remove code for no physical device
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
gasket_enable_dev code for enabling a gasket device with no physical PCI device registered shouldn't be necessary.
--- drivers/staging/gasket/gasket_core.c | 12 +----------- 1 file changed, 1 insertion(+), 11 deletions(-)
drivers/staging/gasket/gasket_core.c +1 -11
@@ -899,7 +899,6 @@ static int gasket_enable_dev( {int tbl_idx;int ret; -struct device *ddev;const struct gasket_driver_desc *driver_desc = internal_desc->driver_desc;
@@ -920,21 +919,12 @@ static int gasket_enable_dev(for (tbl_idx =0 ; tbl_idx < driver_desc->num_page_tables; tbl_idx++) {gasket_log_debug ( gasket_dev,"Initializing page table %d." , tbl_idx); -if (gasket_dev->pci_dev) { - ddev = &gasket_dev->pci_dev->dev; - }else { -gasket_log_error ( - gasket_dev, -"%s with no physical device!!" , __func__); -WARN_ON (1 ); - ddev =NULL ; - } ret =gasket_page_table_init ( &gasket_dev->page_table[tbl_idx], &gasket_dev->bar_data[ driver_desc->page_table_bar_index], &driver_desc->page_table_configs[tbl_idx], - ddev, gasket_dev->pci_dev, true); + &gasket_dev->pci_dev->dev, gasket_dev->pci_dev, true);if (ret) { gasket_log_error( gasket_dev,
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 20/32 staging: gasket: fix class create bug handling
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
class_create() never returns NULL, and this driver should never return PTR_ERR(NULL) anyway.
--- drivers/staging/gasket/gasket_core.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/gasket/gasket_core.c +1 -1
@@ -322,7 +322,7 @@ int gasket_register_device(const struct gasket_driver_desc *driver_desc) internal->class =class_create (driver_desc->module, driver_desc->name); -if (IS_ERR_OR_NULL (internal->class)) { +if (IS_ERR (internal->class)) {gasket_nodev_error ("Cannot register %s class [ret=%ld]" , driver_desc->name,PTR_ERR (internal->class)); ret =PTR_ERR (internal->class);
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 21/32 staging: gasket: remove unnecessary code in coherent allocator
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Remove extraneous statement in gasket_config_coherent_allocator()
--- drivers/staging/gasket/gasket_ioctl.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/staging/gasket/gasket_ioctl.c +1 -3
@@ -419,10 +419,8 @@ static int gasket_config_coherent_allocator(if (ibuf.page_table_index >= gasket_dev->num_page_tables)return -EFAULT ; -if (ibuf.size >PAGE_SIZE *MAX_NUM_COHERENT_PAGES ) { - ibuf.size =PAGE_SIZE *MAX_NUM_COHERENT_PAGES ; +if (ibuf.size >PAGE_SIZE *MAX_NUM_COHERENT_PAGES )return -ENOMEM ; - }if (ibuf.enable ==0 ) { ret = gasket_free_coherent_memory(
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 22/32 staging: gasket: don't treat no device reset callback as an error
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
It is not an error for a device to not have a reset callback registered.
--- drivers/staging/gasket/gasket_core.c | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-)
drivers/staging/gasket/gasket_core.c +2 -5
@@ -1840,11 +1840,8 @@ int gasket_reset_nolock(struct gasket_dev *gasket_dev, uint reset_type)const struct gasket_driver_desc *driver_desc; driver_desc = gasket_dev->internal_desc->driver_desc; -if (!driver_desc->device_reset_cb) { -gasket_log_error ( - gasket_dev,"No device reset callback was registered." ); -return -EINVAL ; - } +if (!driver_desc->device_reset_cb) +return 0 ;/* Perform a device reset of the requested type. */ ret = driver_desc->device_reset_cb (gasket_dev, reset_type);
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 23/32 staging: gasket: gasket_mmap return error instead of valid BAR index
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
When offset to be mapped matches both a BAR region and a coherent mapped region return an error as intended, not the BAR index.
--- drivers/staging/gasket/gasket_core.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/gasket/gasket_core.c +1 -1
@@ -1630,7 +1630,7 @@ static int gasket_mmap(struct file *filp, struct vm_area_struct *vma)"0x%lx" , raw_offset);trace_gasket_mmap_exit (bar_index); -return bar_index; +return -EINVAL ; } vma->vm_private_data = gasket_dev;
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 24/32 staging: gasket: apex_clock_gating simplify logic, reduce indentation
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Collapse together two checks and return immediately, avoid conditional indentation for most of function code.
--- drivers/staging/gasket/apex_driver.c | 43 +++++++++++++--------------- 1 file changed, 20 insertions(+), 23 deletions(-)
drivers/staging/gasket/apex_driver.c +20 -23
@@ -660,33 +660,30 @@ static long apex_clock_gating(struct gasket_dev *gasket_dev, void __user *arg) {struct apex_gate_clock_ioctl ibuf; -if (bypass_top_level) +if (bypass_top_level || !allow_sw_clock_gating)return 0 ; -if (allow_sw_clock_gating) { -if (copy_from_user (&ibuf, arg,sizeof (ibuf))) -return -EFAULT ; +if (copy_from_user (&ibuf, arg,sizeof (ibuf))) +return -EFAULT ; -gasket_log_error ( - gasket_dev,"%s %llu" , __func__, ibuf.enable); +gasket_log_error (gasket_dev,"%s %llu" , __func__, ibuf.enable); -if (ibuf.enable) { -/* Quiesce AXI, gate GCB clock. */ -gasket_read_modify_write_32 ( - gasket_dev,APEX_BAR_INDEX , -APEX_BAR2_REG_AXI_QUIESCE ,0x1 ,1 ,16 ); -gasket_read_modify_write_32 ( - gasket_dev,APEX_BAR_INDEX , -APEX_BAR2_REG_GCB_CLOCK_GATE ,0x1 ,2 ,18 ); - }else { -/* Un-gate GCB clock, un-quiesce AXI. */ -gasket_read_modify_write_32 ( - gasket_dev,APEX_BAR_INDEX , -APEX_BAR2_REG_GCB_CLOCK_GATE ,0x0 ,2 ,18 ); -gasket_read_modify_write_32 ( - gasket_dev,APEX_BAR_INDEX , -APEX_BAR2_REG_AXI_QUIESCE ,0x0 ,1 ,16 ); - } +if (ibuf.enable) { +/* Quiesce AXI, gate GCB clock. */ +gasket_read_modify_write_32 ( + gasket_dev,APEX_BAR_INDEX , +APEX_BAR2_REG_AXI_QUIESCE ,0x1 ,1 ,16 ); +gasket_read_modify_write_32 ( + gasket_dev,APEX_BAR_INDEX , +APEX_BAR2_REG_GCB_CLOCK_GATE ,0x1 ,2 ,18 ); + }else { +/* Un-gate GCB clock, un-quiesce AXI. */ +gasket_read_modify_write_32 ( + gasket_dev,APEX_BAR_INDEX , +APEX_BAR2_REG_GCB_CLOCK_GATE ,0x0 ,2 ,18 ); +gasket_read_modify_write_32 ( + gasket_dev,APEX_BAR_INDEX , +APEX_BAR2_REG_AXI_QUIESCE ,0x0 ,1 ,16 ); }return 0 ; }
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 25/32 staging: gasket: apex_ioctl_check_permissions use bool return type
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Convert from int return to bool.
--- drivers/staging/gasket/apex_driver.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/staging/gasket/apex_driver.c +3 -3
@@ -141,7 +141,7 @@ static int apex_reset(struct gasket_dev *gasket_dev, uint type);static int apex_get_status (struct gasket_dev *gasket_dev); -static uint apex_ioctl_check_permissions (struct file *file,uint cmd); +static bool apex_ioctl_check_permissions (struct file *file,uint cmd);static long apex_ioctl (struct file *file,uint cmd,void __user *arg);
@@ -626,9 +626,9 @@ static bool is_gcb_in_reset(struct gasket_dev *gasket_dev)* @file: File pointer from ioctl. * @cmd: ioctl command. * -* Returns 1 if the current user may execute this ioctl, and 0 otherwise. +* Returns true if the current user may execute this ioctl, and false otherwise. */ -static uint apex_ioctl_check_permissions (struct file *filp,uint cmd) +static bool apex_ioctl_check_permissions (struct file *filp,uint cmd) {return !!(filp->f_mode &FMODE_WRITE ); }
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 26/32 staging: gasket: gasket page table functions use bool return type
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Convert from int to bool return type for gasket page table functions that return values used as booleans.
--- drivers/staging/gasket/gasket_page_table.c | 58 +++++++++++----------- drivers/staging/gasket/gasket_page_table.h | 8 +-- 2 files changed, 33 insertions(+), 33 deletions(-)
drivers/staging/gasket/gasket_page_table.c +29 -29
@@ -262,16 +262,16 @@ static void gasket_perform_unmapping(static void gasket_free_extended_subtable (struct gasket_page_table *pg_tbl,struct gasket_page_table_entry *pte,u64 __iomem *att_reg); -static int gasket_release_page (struct page *page); +static bool gasket_release_page (struct page *page);/* Other/utility declarations */ -static inline int gasket_addr_is_simple ( +static inline bool gasket_addr_is_simple (struct gasket_page_table *pg_tbl,ulong addr); -static int gasket_is_simple_dev_addr_bad ( +static bool gasket_is_simple_dev_addr_bad (struct gasket_page_table *pg_tbl,ulong dev_addr,uint num_pages); -static int gasket_is_extended_dev_addr_bad ( +static bool gasket_is_extended_dev_addr_bad (struct gasket_page_table *pg_tbl,ulong dev_addr,uint num_pages); -static int gasket_is_pte_range_free ( +static bool gasket_is_pte_range_free (struct gasket_page_table_entry *pte,uint num_entries);static void gasket_page_table_garbage_collect_nolock (struct gasket_page_table *pg_tbl);
@@ -558,7 +558,7 @@ int gasket_page_table_lookup_page( }/* See gasket_page_table.h for description. */ -int gasket_page_table_are_addrs_bad ( +bool gasket_page_table_are_addrs_bad (struct gasket_page_table *pg_tbl,ulong host_addr,ulong dev_addr,ulong bytes) {
@@ -567,7 +567,7 @@ int gasket_page_table_are_addrs_bad( pg_tbl,"host mapping address 0x%lx must be page aligned" , host_addr); -return 1 ; +return true; }return gasket_page_table_is_dev_addr_bad (pg_tbl, dev_addr, bytes);
@@ -575,7 +575,7 @@ int gasket_page_table_are_addrs_bad(EXPORT_SYMBOL (gasket_page_table_are_addrs_bad);/* See gasket_page_table.h for description. */ -int gasket_page_table_is_dev_addr_bad ( +bool gasket_page_table_is_dev_addr_bad (struct gasket_page_table *pg_tbl,ulong dev_addr,ulong bytes) {uint num_pages = bytes /PAGE_SIZE ;
@@ -584,7 +584,7 @@ int gasket_page_table_is_dev_addr_bad(gasket_pg_tbl_error ( pg_tbl,"mapping size 0x%lX must be page aligned" , bytes); -return 1 ; +return true; }if (num_pages ==0 ) {
@@ -592,7 +592,7 @@ int gasket_page_table_is_dev_addr_bad( pg_tbl,"requested mapping is less than one page: %lu / %lu" , bytes, PAGE_SIZE); -return 1 ; +return true; }if (gasket_addr_is_simple (pg_tbl, dev_addr))
@@ -1285,23 +1285,23 @@ static void gasket_free_extended_subtable(/* * Safely return a page to the OS. * @page: The page to return to the OS. -* Returns 1 if the page was released, 0 if it was +* Returns true if the page was released, false if it was * ignored. */ -static int gasket_release_page (struct page *page) +static bool gasket_release_page (struct page *page) {if (!page) -return 0 ; +return false;if (!PageReserved (page))SetPageDirty (page);put_page (page); -return 1 ; +return true; }/* Evaluates to nonzero if the specified virtual address is simple. */ -static inline int gasket_addr_is_simple ( +static inline bool gasket_addr_is_simple (struct gasket_page_table *pg_tbl,ulong addr) {return !((addr) & (pg_tbl)->extended_flag);
@@ -1317,7 +1317,7 @@ static inline int gasket_addr_is_simple(* address to/from page + offset) and that the requested page range starts and * ends within the set of currently-partitioned simple pages. */ -static int gasket_is_simple_dev_addr_bad ( +static bool gasket_is_simple_dev_addr_bad (struct gasket_page_table *pg_tbl,ulong dev_addr,uint num_pages) {ulong page_offset = dev_addr & (PAGE_SIZE -1 );
@@ -1328,7 +1328,7 @@ static int gasket_is_simple_dev_addr_bad( pg_tbl, 1, page_index, page_offset) != dev_addr) {gasket_pg_tbl_error ( pg_tbl,"address is invalid, 0x%lX" , dev_addr); -return 1 ; +return true; }if (page_index >= pg_tbl->num_simple_entries) {
@@ -1336,7 +1336,7 @@ static int gasket_is_simple_dev_addr_bad( pg_tbl,"starting slot at %lu is too large, max is < %u" , page_index, pg_tbl->num_simple_entries); -return 1 ; +return true; }if (page_index + num_pages > pg_tbl->num_simple_entries) {
@@ -1344,10 +1344,10 @@ static int gasket_is_simple_dev_addr_bad( pg_tbl,"ending slot at %lu is too large, max is <= %u" , page_index + num_pages, pg_tbl->num_simple_entries); -return 1 ; +return true; } -return 0 ; +return false; }/*
@@ -1359,7 +1359,7 @@ static int gasket_is_simple_dev_addr_bad(* @dev_addr: The device address to which the pages will be mapped. * @num_pages: The number of second-level/sub pages in the range to consider. */ -static int gasket_is_extended_dev_addr_bad ( +static bool gasket_is_extended_dev_addr_bad (struct gasket_page_table *pg_tbl,ulong dev_addr,uint num_pages) {/* Starting byte index of dev_addr into the first mapped page */
@@ -1373,7 +1373,7 @@ static int gasket_is_extended_dev_addr_bad(if (addr >> (GASKET_EXTENDED_LVL0_WIDTH +GASKET_EXTENDED_LVL0_SHIFT )) {gasket_pg_tbl_error (pg_tbl,"device address out of bound, 0x%p" , (void *)dev_addr); -return 1 ; +return true; }/* Find the starting sub-page index in the space of all sub-pages. */
@@ -1391,7 +1391,7 @@ static int gasket_is_extended_dev_addr_bad( pg_tbl, 0, page_global_idx, page_offset) != dev_addr) {gasket_pg_tbl_error ( pg_tbl,"address is invalid, 0x%p" , (void *)dev_addr); -return 1 ; +return true; }if (page_lvl0_idx >= pg_tbl->num_extended_entries) {
@@ -1399,7 +1399,7 @@ static int gasket_is_extended_dev_addr_bad( pg_tbl,"starting level 0 slot at %lu is too large, max is < " "%u", page_lvl0_idx, pg_tbl->num_extended_entries); -return 1 ; +return true; }if (page_lvl0_idx + num_lvl0_pages > pg_tbl->num_extended_entries) {
@@ -1408,10 +1408,10 @@ static int gasket_is_extended_dev_addr_bad("ending level 0 slot at %lu is too large, max is <= %u" , page_lvl0_idx + num_lvl0_pages, pg_tbl->num_extended_entries); -return 1 ; +return true; } -return 0 ; +return false; }/*
@@ -1425,17 +1425,17 @@ static int gasket_is_extended_dev_addr_bad(* * The page table mutex must be held before this call. */ -static int gasket_is_pte_range_free ( +static bool gasket_is_pte_range_free (struct gasket_page_table_entry *ptes,uint num_entries) {int i;for (i =0 ; i < num_entries; i++) {if (ptes[i].status !=PTE_FREE ) -return 0 ; +return false; } -return 1 ; +return true; }/*
drivers/staging/gasket/gasket_page_table.h +4 -4
@@ -162,9 +162,9 @@ int gasket_page_table_lookup_page(* specified by both addresses and the size are valid for mapping pages into * device memory. * -* Returns 1 if true - if the mapping is bad, 0 otherwise. +* Returns true if the mapping is bad, false otherwise. */ -int gasket_page_table_are_addrs_bad ( +bool gasket_page_table_are_addrs_bad (struct gasket_page_table *page_table,ulong host_addr,ulong dev_addr,ulong bytes);
@@ -178,9 +178,9 @@ int gasket_page_table_are_addrs_bad(* specified by the device address and the size is valid for mapping pages into * device memory. * -* Returns 1 if true - if the address is bad, 0 otherwise. +* Returns true if the address is bad, false otherwise. */ -int gasket_page_table_is_dev_addr_bad ( +bool gasket_page_table_is_dev_addr_bad (struct gasket_page_table *page_table,ulong dev_addr,ulong bytes);/*
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 27/32 staging: gasket: remove else clause after return in if clause
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Else after return is unnecessary and may cause static code checkers to complain.
--- drivers/staging/gasket/gasket_page_table.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/staging/gasket/gasket_page_table.c +1 -3
@@ -598,9 +598,7 @@ bool gasket_page_table_is_dev_addr_bad(if (gasket_addr_is_simple (pg_tbl, dev_addr))return gasket_is_simple_dev_addr_bad ( pg_tbl, dev_addr, num_pages); -else -return gasket_is_extended_dev_addr_bad ( - pg_tbl, dev_addr, num_pages); +return gasket_is_extended_dev_addr_bad (pg_tbl, dev_addr, num_pages); }EXPORT_SYMBOL (gasket_page_table_is_dev_addr_bad);
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 28/32 staging: gasket: fix comment syntax in apex.h
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Use kernel-style multi-line comment syntax.
--- drivers/staging/gasket/apex.h | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)
drivers/staging/gasket/apex.h +4 -3
@@ -22,9 +22,10 @@#define APEX_EXTENDED_SHIFT 63/* Extended address bit position. */ -/* Addresses are 2^3=8 bytes each. */ -/* page in second level page table */ -/* holds APEX_PAGE_SIZE/8 addresses */ +/* +* Addresses are 2^3=8 bytes each. Page in second level page table holds +* APEX_PAGE_SIZE/8 addresses. +*/ #define APEX_ADDR_SHIFT 3#define APEX_LEVEL_SHIFT (APEX_PAGE_SHIFT - APEX_ADDR_SHIFT)#define APEX_LEVEL_SIZE BIT(APEX_LEVEL_SHIFT)
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 29/32 staging: gasket: remove unnecessary parens in page table code
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
gasket_alloc_coherent_memory() extra parentheses in statement.
--- drivers/staging/gasket/gasket_page_table.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/gasket/gasket_page_table.c +1 -1
@@ -1639,7 +1639,7 @@ int gasket_alloc_coherent_memory(struct gasket_dev *gasket_dev, u64 size,dma_addr_t handle;void *mem;int j; -unsigned int num_pages = (size +PAGE_SIZE -1 ) / (PAGE_SIZE ); +unsigned int num_pages = (size +PAGE_SIZE -1 ) /PAGE_SIZE ;const struct gasket_driver_desc *driver_desc =gasket_get_driver_desc (gasket_dev);
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 30/32 staging: gasket: gasket_mmap use PAGE_MASK
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
gasket_mmap use PAGE_MASK, instead of performing math on PAGE_SIZE, for simplicity and clarity.
--- drivers/staging/gasket/gasket_core.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/gasket/gasket_core.c +1 -1
@@ -1594,7 +1594,7 @@ static int gasket_mmap(struct file *filp, struct vm_area_struct *vma) } driver_desc = gasket_dev->internal_desc->driver_desc; -if (vma->vm_start & (PAGE_SIZE -1 )) { +if (vma->vm_start & ~PAGE_MASK ) {gasket_log_error ( gasket_dev,"Base address not page-aligned: 0x%p\n" , (void *)vma->vm_start);
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 31/32 staging: gasket: remove extra parens in gasket_write_mappable_regions
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Remove unneeded parentheses around subexpressions.
--- drivers/staging/gasket/gasket_core.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/gasket/gasket_core.c +1 -1
@@ -1893,7 +1893,7 @@ static ssize_t gasket_write_mappable_regions(if (bar_desc.permissions ==GASKET_NOMAP )return 0 ;for (i =0 ; - (i < bar_desc.num_mappable_regions) && (total_written <PAGE_SIZE ); + i < bar_desc.num_mappable_regions && total_written <PAGE_SIZE ; i++) { min_addr = bar_desc.mappable_regions[i].start - driver_desc->legacy_mmap_address_offset;
Signature · 1 line
2.18.0.203.gfac676dfb9-goog
PATCH v3 32/32 staging: gasket: fix multi-line comment syntax in gasket_core.h
In reply to Todd PoynorFrom: Todd Poynor <[email protected]>
Use consistent kernel-style multi-line comment syntax.
--- drivers/staging/gasket/gasket_core.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/staging/gasket/gasket_core.h +2 -1
@@ -54,7 +54,8 @@ enum gasket_interrupt_type { PLATFORM_WIRE = 2, }; -/* Used to describe a Gasket interrupt. Contains an interrupt index, a register, +/* +* Used to describe a Gasket interrupt. Contains an interrupt index, a register, * and packing data for that interrupt. The register and packing data * fields are relevant only for PCI_MSIX interrupt type and can be * set to 0 for everything else.
Signature · 1 line
2.18.0.203.gfac676dfb9-goog