| Message ID | 20190402210623.14988-2-jernej.skrabec@siol.net (mailing list archive) |
|---|---|
| State | Awaiting Upstream, archived |
| Headers | show |
| Series | clk: sunxi-ng: H6 related clock fixes | expand |
Hi, On Tue, Apr 02, 2019 at 11:06:21PM +0200, Jernej Skrabec wrote: > Sometimes one of the nkmp factors is unused. This means that one of the > factors shift and width values are set to 0. Current nkmp clock code > generates a mask for each factor with GENMASK(width + shift - 1, shift). > For unused factor this translates to GENMASK(-1, 0). This code is > further expanded by C preprocessor to final version: > (((~0UL) - (1UL << (0)) + 1) & (~0UL >> (BITS_PER_LONG - 1 - (-1)))) > or a bit simplified: > (~0UL & (~0UL >> BITS_PER_LONG)) > > It turns out that result of the second part (~0UL >> BITS_PER_LONG) is > actually undefined by C standard, which clearly specifies: > > "If the value of the right operand is negative or is greater than or > equal to the width of the promoted left operand, the behavior is > undefined." > > Additionally, compiling kernel with aarch64-linux-gnu-gcc 8.3.0 gave > different results whether literals or variables with same values as > literals were used. GENMASK with literals -1 and 0 gives zero and with > variables gives 0xFFFFFFFFFFFFFFF (~0UL). Because nkmp driver uses > GENMASK with variables as parameter, expression calculates mask as ~0UL > instead of 0. This has further consequences that LSB in register is > always set to 1 (1 is neutral value for a factor and shift is 0). > > For example, H6 pll-de clock is set to 600 MHz by sun4i-drm driver, but > due to this bug ends up being 300 MHz. Additionally, 300 MHz seems to be > too low because following warning can be found in dmesg: > > [ 1.752763] WARNING: CPU: 2 PID: 41 at drivers/clk/sunxi-ng/ccu_common.c:41 ccu_helper_wait_for_lock.part.0+0x6c/0x90 > [ 1.763378] Modules linked in: > [ 1.766441] CPU: 2 PID: 41 Comm: kworker/2:1 Not tainted 5.1.0-rc2-next-20190401 #138 > [ 1.774269] Hardware name: Pine H64 (DT) > [ 1.778200] Workqueue: events deferred_probe_work_func > [ 1.783341] pstate: 40000005 (nZcv daif -PAN -UAO) > [ 1.788135] pc : ccu_helper_wait_for_lock.part.0+0x6c/0x90 > [ 1.793623] lr : ccu_helper_wait_for_lock.part.0+0x48/0x90 > [ 1.799107] sp : ffff000010f93840 > [ 1.802422] x29: ffff000010f93840 x28: 0000000000000000 > [ 1.807735] x27: ffff800073ce9d80 x26: ffff000010afd1b8 > [ 1.813049] x25: ffffffffffffffff x24: 00000000ffffffff > [ 1.818362] x23: 0000000000000001 x22: ffff000010abd5c8 > [ 1.823675] x21: 0000000010000000 x20: 00000000685f367e > [ 1.828987] x19: 0000000000001801 x18: 0000000000000001 > [ 1.834300] x17: 0000000000000001 x16: 0000000000000000 > [ 1.839613] x15: 0000000000000000 x14: ffff000010789858 > [ 1.844926] x13: 0000000000000000 x12: 0000000000000001 > [ 1.850239] x11: 0000000000000000 x10: 0000000000000970 > [ 1.855551] x9 : ffff000010f936c0 x8 : ffff800074cec0d0 > [ 1.860864] x7 : 0000800067117000 x6 : 0000000115c30b41 > [ 1.866177] x5 : 00ffffffffffffff x4 : 002c959300bfe500 > [ 1.871490] x3 : 0000000000000018 x2 : 0000000029aaaaab > [ 1.876802] x1 : 00000000000002e6 x0 : 00000000686072bc > [ 1.882114] Call trace: > [ 1.884565] ccu_helper_wait_for_lock.part.0+0x6c/0x90 > [ 1.889705] ccu_helper_wait_for_lock+0x10/0x20 > [ 1.894236] ccu_nkmp_set_rate+0x244/0x2a8 > [ 1.898334] clk_change_rate+0x144/0x290 > [ 1.902258] clk_core_set_rate_nolock+0x180/0x1b8 > [ 1.906963] clk_set_rate+0x34/0xa0 > [ 1.910455] sun8i_mixer_bind+0x484/0x558 > [ 1.914466] component_bind_all+0x10c/0x230 > [ 1.918651] sun4i_drv_bind+0xc4/0x1a0 > [ 1.922401] try_to_bring_up_master+0x164/0x1c0 > [ 1.926932] __component_add+0xa0/0x168 > [ 1.930769] component_add+0x10/0x18 > [ 1.934346] sun8i_dw_hdmi_probe+0x18/0x20 > [ 1.938443] platform_drv_probe+0x50/0xa0 > [ 1.942455] really_probe+0xcc/0x280 > [ 1.946032] driver_probe_device+0x54/0xe8 > [ 1.950130] __device_attach_driver+0x80/0xb8 > [ 1.954488] bus_for_each_drv+0x78/0xc8 > [ 1.958326] __device_attach+0xd4/0x130 > [ 1.962163] device_initial_probe+0x10/0x18 > [ 1.966348] bus_probe_device+0x90/0x98 > [ 1.970185] deferred_probe_work_func+0x6c/0xa0 > [ 1.974720] process_one_work+0x1e0/0x320 > [ 1.978732] worker_thread+0x228/0x428 > [ 1.982484] kthread+0x120/0x128 > [ 1.985714] ret_from_fork+0x10/0x18 > [ 1.989290] ---[ end trace 9babd42e1ca4b84f ]--- > > This commit solves the issue by first checking value of the factor > width. If it is equal to 0 (unused factor), mask is set to 0, otherwise > GENMASK() macro is used as before. > > Fixes: d897ef56faf9 ("clk: sunxi-ng: Mask nkmp factors when setting register") > Signed-off-by: Jernej Skrabec <jernej.skrabec@siol.net> > --- > drivers/clk/sunxi-ng/ccu_nkmp.c | 18 +++++++++++++----- > 1 file changed, 13 insertions(+), 5 deletions(-) > > diff --git a/drivers/clk/sunxi-ng/ccu_nkmp.c b/drivers/clk/sunxi-ng/ccu_nkmp.c > index 9b49adb20d07..69dfc6de1c4e 100644 > --- a/drivers/clk/sunxi-ng/ccu_nkmp.c > +++ b/drivers/clk/sunxi-ng/ccu_nkmp.c > @@ -167,7 +167,7 @@ static int ccu_nkmp_set_rate(struct clk_hw *hw, unsigned long rate, > unsigned long parent_rate) > { > struct ccu_nkmp *nkmp = hw_to_ccu_nkmp(hw); > - u32 n_mask, k_mask, m_mask, p_mask; > + u32 n_mask = 0, k_mask = 0, m_mask = 0, p_mask = 0; > struct _ccu_nkmp _nkmp; > unsigned long flags; > u32 reg; > @@ -186,10 +186,18 @@ static int ccu_nkmp_set_rate(struct clk_hw *hw, unsigned long rate, > > ccu_nkmp_find_best(parent_rate, rate, &_nkmp); > > - n_mask = GENMASK(nkmp->n.width + nkmp->n.shift - 1, nkmp->n.shift); > - k_mask = GENMASK(nkmp->k.width + nkmp->k.shift - 1, nkmp->k.shift); > - m_mask = GENMASK(nkmp->m.width + nkmp->m.shift - 1, nkmp->m.shift); > - p_mask = GENMASK(nkmp->p.width + nkmp->p.shift - 1, nkmp->p.shift); > + if (nkmp->n.width) > + n_mask = GENMASK(nkmp->n.width + nkmp->n.shift - 1, > + nkmp->n.shift); > + if (nkmp->k.width) > + k_mask = GENMASK(nkmp->k.width + nkmp->k.shift - 1, > + nkmp->k.shift); > + if (nkmp->m.width) > + m_mask = GENMASK(nkmp->m.width + nkmp->m.shift - 1, > + nkmp->m.shift); > + if (nkmp->p.width) > + p_mask = GENMASK(nkmp->p.width + nkmp->p.shift - 1, > + nkmp->p.shift); Thanks for the awesome commit log, and for digging this out. I've applied this patch. Could you write a comment just above the code to explain it as well so that it doesn't confuse anyone? You can send a subsequent patch. Thanks! Maxime -- Maxime Ripard, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
diff --git a/drivers/clk/sunxi-ng/ccu_nkmp.c b/drivers/clk/sunxi-ng/ccu_nkmp.c index 9b49adb20d07..69dfc6de1c4e 100644 --- a/drivers/clk/sunxi-ng/ccu_nkmp.c +++ b/drivers/clk/sunxi-ng/ccu_nkmp.c @@ -167,7 +167,7 @@ static int ccu_nkmp_set_rate(struct clk_hw *hw, unsigned long rate, unsigned long parent_rate) { struct ccu_nkmp *nkmp = hw_to_ccu_nkmp(hw); - u32 n_mask, k_mask, m_mask, p_mask; + u32 n_mask = 0, k_mask = 0, m_mask = 0, p_mask = 0; struct _ccu_nkmp _nkmp; unsigned long flags; u32 reg; @@ -186,10 +186,18 @@ static int ccu_nkmp_set_rate(struct clk_hw *hw, unsigned long rate, ccu_nkmp_find_best(parent_rate, rate, &_nkmp); - n_mask = GENMASK(nkmp->n.width + nkmp->n.shift - 1, nkmp->n.shift); - k_mask = GENMASK(nkmp->k.width + nkmp->k.shift - 1, nkmp->k.shift); - m_mask = GENMASK(nkmp->m.width + nkmp->m.shift - 1, nkmp->m.shift); - p_mask = GENMASK(nkmp->p.width + nkmp->p.shift - 1, nkmp->p.shift); + if (nkmp->n.width) + n_mask = GENMASK(nkmp->n.width + nkmp->n.shift - 1, + nkmp->n.shift); + if (nkmp->k.width) + k_mask = GENMASK(nkmp->k.width + nkmp->k.shift - 1, + nkmp->k.shift); + if (nkmp->m.width) + m_mask = GENMASK(nkmp->m.width + nkmp->m.shift - 1, + nkmp->m.shift); + if (nkmp->p.width) + p_mask = GENMASK(nkmp->p.width + nkmp->p.shift - 1, + nkmp->p.shift); spin_lock_irqsave(nkmp->common.lock, flags);
Sometimes one of the nkmp factors is unused. This means that one of the factors shift and width values are set to 0. Current nkmp clock code generates a mask for each factor with GENMASK(width + shift - 1, shift). For unused factor this translates to GENMASK(-1, 0). This code is further expanded by C preprocessor to final version: (((~0UL) - (1UL << (0)) + 1) & (~0UL >> (BITS_PER_LONG - 1 - (-1)))) or a bit simplified: (~0UL & (~0UL >> BITS_PER_LONG)) It turns out that result of the second part (~0UL >> BITS_PER_LONG) is actually undefined by C standard, which clearly specifies: "If the value of the right operand is negative or is greater than or equal to the width of the promoted left operand, the behavior is undefined." Additionally, compiling kernel with aarch64-linux-gnu-gcc 8.3.0 gave different results whether literals or variables with same values as literals were used. GENMASK with literals -1 and 0 gives zero and with variables gives 0xFFFFFFFFFFFFFFF (~0UL). Because nkmp driver uses GENMASK with variables as parameter, expression calculates mask as ~0UL instead of 0. This has further consequences that LSB in register is always set to 1 (1 is neutral value for a factor and shift is 0). For example, H6 pll-de clock is set to 600 MHz by sun4i-drm driver, but due to this bug ends up being 300 MHz. Additionally, 300 MHz seems to be too low because following warning can be found in dmesg: [ 1.752763] WARNING: CPU: 2 PID: 41 at drivers/clk/sunxi-ng/ccu_common.c:41 ccu_helper_wait_for_lock.part.0+0x6c/0x90 [ 1.763378] Modules linked in: [ 1.766441] CPU: 2 PID: 41 Comm: kworker/2:1 Not tainted 5.1.0-rc2-next-20190401 #138 [ 1.774269] Hardware name: Pine H64 (DT) [ 1.778200] Workqueue: events deferred_probe_work_func [ 1.783341] pstate: 40000005 (nZcv daif -PAN -UAO) [ 1.788135] pc : ccu_helper_wait_for_lock.part.0+0x6c/0x90 [ 1.793623] lr : ccu_helper_wait_for_lock.part.0+0x48/0x90 [ 1.799107] sp : ffff000010f93840 [ 1.802422] x29: ffff000010f93840 x28: 0000000000000000 [ 1.807735] x27: ffff800073ce9d80 x26: ffff000010afd1b8 [ 1.813049] x25: ffffffffffffffff x24: 00000000ffffffff [ 1.818362] x23: 0000000000000001 x22: ffff000010abd5c8 [ 1.823675] x21: 0000000010000000 x20: 00000000685f367e [ 1.828987] x19: 0000000000001801 x18: 0000000000000001 [ 1.834300] x17: 0000000000000001 x16: 0000000000000000 [ 1.839613] x15: 0000000000000000 x14: ffff000010789858 [ 1.844926] x13: 0000000000000000 x12: 0000000000000001 [ 1.850239] x11: 0000000000000000 x10: 0000000000000970 [ 1.855551] x9 : ffff000010f936c0 x8 : ffff800074cec0d0 [ 1.860864] x7 : 0000800067117000 x6 : 0000000115c30b41 [ 1.866177] x5 : 00ffffffffffffff x4 : 002c959300bfe500 [ 1.871490] x3 : 0000000000000018 x2 : 0000000029aaaaab [ 1.876802] x1 : 00000000000002e6 x0 : 00000000686072bc [ 1.882114] Call trace: [ 1.884565] ccu_helper_wait_for_lock.part.0+0x6c/0x90 [ 1.889705] ccu_helper_wait_for_lock+0x10/0x20 [ 1.894236] ccu_nkmp_set_rate+0x244/0x2a8 [ 1.898334] clk_change_rate+0x144/0x290 [ 1.902258] clk_core_set_rate_nolock+0x180/0x1b8 [ 1.906963] clk_set_rate+0x34/0xa0 [ 1.910455] sun8i_mixer_bind+0x484/0x558 [ 1.914466] component_bind_all+0x10c/0x230 [ 1.918651] sun4i_drv_bind+0xc4/0x1a0 [ 1.922401] try_to_bring_up_master+0x164/0x1c0 [ 1.926932] __component_add+0xa0/0x168 [ 1.930769] component_add+0x10/0x18 [ 1.934346] sun8i_dw_hdmi_probe+0x18/0x20 [ 1.938443] platform_drv_probe+0x50/0xa0 [ 1.942455] really_probe+0xcc/0x280 [ 1.946032] driver_probe_device+0x54/0xe8 [ 1.950130] __device_attach_driver+0x80/0xb8 [ 1.954488] bus_for_each_drv+0x78/0xc8 [ 1.958326] __device_attach+0xd4/0x130 [ 1.962163] device_initial_probe+0x10/0x18 [ 1.966348] bus_probe_device+0x90/0x98 [ 1.970185] deferred_probe_work_func+0x6c/0xa0 [ 1.974720] process_one_work+0x1e0/0x320 [ 1.978732] worker_thread+0x228/0x428 [ 1.982484] kthread+0x120/0x128 [ 1.985714] ret_from_fork+0x10/0x18 [ 1.989290] ---[ end trace 9babd42e1ca4b84f ]--- This commit solves the issue by first checking value of the factor width. If it is equal to 0 (unused factor), mask is set to 0, otherwise GENMASK() macro is used as before. Fixes: d897ef56faf9 ("clk: sunxi-ng: Mask nkmp factors when setting register") Signed-off-by: Jernej Skrabec <jernej.skrabec@siol.net> --- drivers/clk/sunxi-ng/ccu_nkmp.c | 18 +++++++++++++----- 1 file changed, 13 insertions(+), 5 deletions(-)